Modelling and Analysis of

Electric Power Systems

Power Flow Analysis
Fault Analysis
Power Systems Dynamics and Stability

Lecture 227-0526-00, ITET ETH Zürich

Göran Andersson
EEH - Power Systems Laboratory
ETH Zürich

September 2008
ii
Contents

Preface vii

I Static Analysis 1

1 Introduction 1
1.1 Power Flow Analysis . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Fault Current Analysis . . . . . . . . . . . . . . . . . . . . . . 3
1.3 Literature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

2 Network Models 5
2.1 Lines and Cables . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.2 Transformers . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.2.1 In-Phase Transformers . . . . . . . . . . . . . . . . . . 10
2.2.2 Phase-Shifting Transformers . . . . . . . . . . . . . . . 12
2.2.3 Unified Branch Model . . . . . . . . . . . . . . . . . . 14
2.3 Shunt Elements . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.4 Loads . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.5 Generators . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.5.1 Stator Current Heating Limit . . . . . . . . . . . . . . 18
2.5.2 Field Current Heating Limit . . . . . . . . . . . . . . 19
2.5.3 Stator End Region Heating Limit . . . . . . . . . . . . 20

3 Active and Reactive Power Flows 21

3.1 Transmission Lines . . . . . . . . . . . . . . . . . . . . . . . . 21
3.2 In-phase Transformers . . . . . . . . . . . . . . . . . . . . . . 23
3.3 Phase-Shifting Transformer with akm = 1 . . . . . . . . . . . 24
3.4 Unified Power Flow Equations . . . . . . . . . . . . . . . . . . 25

4 Nodal Formulation of the Network Equations 27

5 Basic Power Flow Problem 31

5.1 Basic Bus Types . . . . . . . . . . . . . . . . . . . . . . . . . 31
5.2 Equality and Inequality Constraints . . . . . . . . . . . . . . 32

iii
iv Contents

5.3 Problem Solvability . . . . . . . . . . . . . . . . . . . . . . . . 34

6 Solution of the Power Flow Problem 37

6.1 Solution by Gauss-Seidel Iteration . . . . . . . . . . . . . . . 37
6.2 Newton-Raphson Method . . . . . . . . . . . . . . . . . . . . 39
6.2.1 Unidimensional case . . . . . . . . . . . . . . . . . . . 40
6.2.2 Quadratic Convergence . . . . . . . . . . . . . . . . . 41
6.2.3 Multidimensional Case . . . . . . . . . . . . . . . . . . 42
6.3 Newton-Raphson applied to the Power Flow Equations . . . . 44
6.4 P θ − QU Decoupling . . . . . . . . . . . . . . . . . . . . . . . 45
6.5 Approximative Solutions of the Power Flow Problem . . . . . 48
6.5.1 Linearization . . . . . . . . . . . . . . . . . . . . . . . 48
6.5.2 Matrix Formulation of DC Power Flow Equations . . 52
6.5.3 DC Power Flow Model . . . . . . . . . . . . . . . . . . 55

7 Fault Analysis 57
7.1 Transients on a transmission line . . . . . . . . . . . . . . . . 61
7.2 Short circuit of a synchronous machine . . . . . . . . . . . . . 63
7.3 Algorithms for short circuit studies . . . . . . . . . . . . . . . 66
7.3.1 Generator model . . . . . . . . . . . . . . . . . . . . . 66
7.3.2 Simplifications . . . . . . . . . . . . . . . . . . . . . . 66
7.3.3 Solving the linear system equations . . . . . . . . . . . 67
7.3.4 The superposition technique . . . . . . . . . . . . . . . 68
7.3.5 The Takahashi method . . . . . . . . . . . . . . . . . . 71

II Power System Dynamics and Stability 77

8 Classification and Definitions of Power System Stability 79

8.1 Dynamics in Power Systems . . . . . . . . . . . . . . . . . . . 80
8.1.1 Classification of Dynamics . . . . . . . . . . . . . . . . 80
8.1.2 Modelling . . . . . . . . . . . . . . . . . . . . . . . . . 81
8.2 Power System Stability . . . . . . . . . . . . . . . . . . . . . . 82
8.2.1 Definition of Stability . . . . . . . . . . . . . . . . . . 82
8.2.2 Classification of Power System Stability . . . . . . . . 84
8.3 Literature on Power System Dynamics and Stability . . . . . 87

9 Synchronous Machine Models 89

9.1 Design and Operating Principle . . . . . . . . . . . . . . . . . 89
9.1.1 Rotor Types . . . . . . . . . . . . . . . . . . . . . . . 90
9.1.2 Stator Field . . . . . . . . . . . . . . . . . . . . . . . . 91
9.1.3 Magnetic Torque . . . . . . . . . . . . . . . . . . . . . 94
9.2 Stationary Operation . . . . . . . . . . . . . . . . . . . . . . . 95
9.2.1 Stationary Single Phase Equivalent Circuit . . . . . . 95
Contents v

9.2.2 Phasor diagram . . . . . . . . . . . . . . . . . . . . . . 97

9.2.3 Operational Limits . . . . . . . . . . . . . . . . . . . . 98
9.3 Dynamic Operation . . . . . . . . . . . . . . . . . . . . . . . 100
9.3.1 Transient Single Phase Equivalent Circuit . . . . . . . 100
9.3.2 Simplified Mechanical Model . . . . . . . . . . . . . . 100

10 The Swing Equation 103

10.1 Derivation of the Swing Equation . . . . . . . . . . . . . . . . 103
10.2 Analysis of the Swing Equation . . . . . . . . . . . . . . . . . 105
10.3 Swing Equation as System of First Order Differential Equations106

11 Power Swings in a Simple System 109

11.1 The Swing Equation and its Solutions . . . . . . . . . . . . . 109
11.1.1 Qualitative Analysis . . . . . . . . . . . . . . . . . . . 111
11.1.2 Stable and Unstable Solutions . . . . . . . . . . . . . . 113
11.2 Equal Area Criterion . . . . . . . . . . . . . . . . . . . . . . . 121
11.3 Lyapunov Stability Criterion . . . . . . . . . . . . . . . . . . 123
11.4 Small Signal Analysis . . . . . . . . . . . . . . . . . . . . . . 124
11.5 Methods to Improve System Stability . . . . . . . . . . . . . 127

12 Oscillations in Multi-Machine Systems 131

12.1 Simplified Model . . . . . . . . . . . . . . . . . . . . . . . . . 131
12.2 Detailed Model . . . . . . . . . . . . . . . . . . . . . . . . . . 132

13 Voltage Stability 135

13.1 Mechanisms of Voltage Instability . . . . . . . . . . . . . . . . 135
13.1.1 Long Term Voltage Instability . . . . . . . . . . . . . 136
13.1.2 Short Term Voltage Instability . . . . . . . . . . . . . 136
13.2 Simple Systems for Analysis of Voltage Stability . . . . . . . 137
13.3 Analysis of Voltage Stability . . . . . . . . . . . . . . . . . . . 140
13.3.1 Stability Indicators . . . . . . . . . . . . . . . . . . . . 141
13.3.2 Analysis of Simple System . . . . . . . . . . . . . . . . 142

14 Control of Electric Power Systems 151

14.1 Control of Active Power and Frequency . . . . . . . . . . . . 153
14.1.1 Spinning reserve . . . . . . . . . . . . . . . . . . . . . 154
14.1.2 Supplementary Reserves . . . . . . . . . . . . . . . . . 156
14.1.3 Back-Up Reserves . . . . . . . . . . . . . . . . . . . . 157
14.2 Control of Reactive Power and Voltage . . . . . . . . . . . . . 157
14.2.1 Reactive Power Control . . . . . . . . . . . . . . . . . 157
14.2.2 Voltage Control . . . . . . . . . . . . . . . . . . . . . . 158
14.3 Supervisory Control of Electric Power Systems . . . . . . . . 160

A Phase-Shifting Transformers 163

vi Contents

B Protections in Electric Power Systems 167

B.1 Design of Protections . . . . . . . . . . . . . . . . . . . . . . . 167
B.2 Distance Protections . . . . . . . . . . . . . . . . . . . . . . . 169
B.2.1 General Principles . . . . . . . . . . . . . . . . . . . . 169
B.2.2 Automatic Re-Closure . . . . . . . . . . . . . . . . . . 171
B.3 Out of Step Protections . . . . . . . . . . . . . . . . . . . . . 172
B.4 System Protections . . . . . . . . . . . . . . . . . . . . . . . . 172
Preface

These lectures notes are intended to be used in the Modellierung und Analyse
elektrischer Netze (Vorlesungsnummer ETH Zürich 227-0526-00) lectures
given at ETH Zürich in Information Technology and Electrical Engineering.
In the lectures three main topics are covered, i.e.

• Power flow analysis

• Fault current calculations

• Power systems dynamics and stability

In Part I of these notes the two first items are covered, while Part II gives
an introduction to dynamics and stability in power systems. In appendices
brief overviews of phase-shifting transformers and power system protections
are given.
The notes start with a derivation and discussion of the models of the most
common power system components to be used in the power flow analysis.
A derivation of the power flow equations based on physical considerations is
then given. The resulting non-linear equations are for realistic power systems
of very large dimension and they have to be solved numerically. The most
commonly used techniques for solving these equations are reviewed. The role
of power flow analysis in power system planning, operation, and analysis is
discussed.1
The next topic covered in these lecture notes is fault current calcula-
tions in power systems. A systematic approach to calculate fault currents
in meshed, large power systems will be derived. The needed models will be
given and the assumptions made when formulating these models discussed.
It will be demonstrated that algebraic models can be used to calculate the
dimensioning fault currents in a power system, and the mathematical analy-
sis has similarities with the power flow analysis, so it is natural to put these
two items in Part I of the notes.
In Part II the dynamic behaviour of the power system during and after
disturbances (faults) will be studied. The concept of power system stability
is defined, and different types of power system instabilities are discussed.
While the phenomena in Part I could be studied by algebraic equations,
1
As stated above, the presentation of the power flow problem is in these lecture notes
based on physical considerations. However, for implementation of these equations in
computer software, an object oriented approach is more suitable. Such an approach is
presented in the lecture notes Modellierung und Analyse elektrischer Netze, by Rainer
Bacher, is also to be dealt with in the lectures. In those lecture notes the techniques for
handling large but sparse matrices are also reviewed.

vii
viii Preface

the description of the power system dynamics requires models based on

differential equations.
These lecture notes provide only a basic introduction to the topics above.
To facilitate for readers who want to get a deeper knowledge of and insight
into these problems, bibliographies are given in the text.
I want to thank numerous assistants, PhD students, and collaborators of
Power Systems Laboratory at ETH Zürich, who have contributed in various
ways to these lecture notes.

Zürich, September 2008

Göran Andersson
 Part I

Static Analysis

1
1
Introduction

This chapter gives a motivation why an algebraic model can be used to de-
scribe the power system in steady state. It is also motivated why an algebraic
approach can be used to calculate fault currents in a power system.

A POWER SYSTEM is predominantly in steady state operation or in a

state that could with sufficient accuracy be regarded as steady state.
In a power system there are always small load changes, switching actions,
and other transients occurring so that in a strict mathematical sense most
of the variables are varying with the time. However, these variations are
most of the time so small that an algebraic, i.e. not time varying model of
the power system is justified.
A short circuit in a power system is clearly not a steady state condition.
Such an event can start a variety of different dynamic phenomena in the
system, and to study these dynamic models are needed. However, when
it comes to calculate the fault currents in the system, steady state (static)
models with appropriate parameter values can be used. A fault current
consists of two components, a transient part, and a steady state part, but
since the transient part can be estimated from the steady state one, fault
current analysis is commonly restricted to the calculation of the steady state
fault currents.

1.1 Power Flow Analysis

It is of utmost importance to be able to calculate the voltages and currents
that different parts of the power system are exposed to. This is essential
not only in order to design the different power system components such
as generators, lines, transformers, shunt elements, etc. so that these can
withstand the stresses they are exposed to during steady state operation
without any risk of damages. Furthermore, for an economical operation of
the system the losses should be kept at a low value taking various constraints
into account, and the risk that the system enters into unstable modes of
operation must be supervised. In order to do this in a satisfactory way the
state of the system, i.e. all (complex) voltages of all nodes in the system,
must be known. With these known, all currents, and hence all active and

1
2 1. Introduction

reactive power flows can be calculated, and other relevant quantities can be
calculated in the system.
Generally the power flow, or load flow, problem is formulated as a non-
linear set of equations
f (x, u, p) = 0 (1.1)
where

f is an n-dimensional (non-linear) function

x is an n-dimensional vector containing the state variables, or states, as

components. These are the unknown voltage magnitudes and voltage
angles of nodes in the system

u is a vector with (known) control outputs, e.g. voltages at generators with

voltage control

p is a vector with the parameters of the network components, e.g. line

reactances and resistances

The power flow problem consists in formulating the equations f in eq. (1.1)
and then solving these with respect to x. This will be the subject dealt with
in the first part of these lectures. A necessary condition for eq. (1.1) to have
a physically meaningful solution is that f and x have the same dimension,
i.e. that we have the same number of unknowns as equations. But in the
general case there is no unique solution, and there are also cases when no
solution exists.
If the states x are known, all other system quantities of interest can
be calculated from these and the known quantities, i.e. u and p. System
quantities of interest are active and reactive power flows through lines and
transformers, reactive power generation from synchronous machines, active
and reactive power consumption by voltage dependent loads, etc.
As mentioned above, the functions f are non-linear, which makes the
equations harder to solve. For the solution of the equations, the linearization

∂f
∆x = ∆y (1.2)
∂x
is quite often used and solved. These equations give also very useful infor-
∂f
mation about the system. The Jacobian matrix , whose elements are
∂x
given by  
∂f ∂fi
= (1.3)
∂x ij ∂xj
can be used for many useful computations, and it is an important indicator
of the system conditions. This will also be elaborated on.
1.2. Fault Current Analysis 3

1.2 Fault Current Analysis

In the lectures Elektrische Energiesysteme it was studied how to calculate
fault currents, e.g. short circuit currents, for simple systems. This analysis
will now be extended to deal with realistic systems including several gener-
ators, lines, loads, and other system components. Generators (synchronous
machines) are important system components when calculating fault currents
and their modelling will be elaborated on and discussed.

1.3 Literature
The material presented in these lectures constitutes only an introduction
to the subject. Further studies can be recommended in the following text
books:

1. Power Systems Analysis, second edition, by Artur R. Bergen and Vijay

Vittal. (Prentice Hall Inc., 2000, ISBN 0-13-691990-1, 619 pages)

2. Computational Methods for Large Sparse Power Systems, An object

oriented approach, by S.A. Soma, S.A. Khaparde, Shubba Pandit (Kluwer
Academic Publishers, 2002, ISBN 0-7923-7591-2, 333 pages)
4 1. Introduction
2
Network Models

In this chapter models of the most common network elements suitable for
power flow analysis are derived. These models will be used in the subsequent
chapters when formulating the power flow problem.

A LL ANALYSIS in the engineering sciences starts with the formulation

of appropriate models. A model, and in power system analysis we al-
most invariably then mean a mathematical model, is a set of equations or
relations, which appropriately describes the interactions between different
quantities in the time frame studied and with the desired accuracy of a phys-
ical or engineered component or system. Hence, depending on the purpose
of the analysis different models of the same physical system or components
might be valid. It is recalled that the general model of a transmission line
was given by the telegraph equation, which is a partial differential equation,
and by assuming stationary sinusoidal conditions the long line equations,
ordinary differential equations, were obtained. By solving these equations
and restricting the interest to the conditions at the ends of the lines, the
lumped-circuit line models (π-models) were obtained, which is an algebraic
model. This gives us three different models each valid for different purposes.
In principle, the complete telegraph equations could be used when study-
ing the steady state conditions at the network nodes. The solution would
then include the initial switching transients along the lines, and the steady
state solution would then be the solution after the transients have decayed.
However, such a solution would contain a lot more information than wanted
and, furthermore, it would require a lot of computational effort. An alge-
braic formulation with the lumped-circuit line model would give the same
result with a much simpler model at a lower computational cost.
In the above example it is quite obvious which model is the appropriate
one, but in many engineering studies the selection of the “correct” model
is often the most difficult part of the study. It is good engineering practice
to use as simple models as possible, but of course not too simple. If too
complicated models are used, the analysis and computations would be un-
necessarily cumbersome. Furthermore, generally more complicated models
need more parameters for their definition, and to get reliable values of these
requires often extensive work.

5
6 2. Network Models

i R´dx L´dx i+di

u G´dx C´dx u+du

dx

Figure 2.1. Equivalent circuit of a line element of length dx

In the subsequent sections algebraic models of the most common power

system components suitable for power flow calculations will be derived. If
not explicitly stated, symmetrical three-phase conditions are assumed in the
following.

2.1 Lines and Cables

The equivalent π-model of a transmission line section was derived in the
lectures Elektrische Energiesysteme, 35-505. The general distributed model
is characterized by the series parameters
R′ = series resistance/km per phase (Ω/km)
X ′ = series reactance/km per phase (Ω/km)
and the shunt parameters
B ′ = shunt susceptance/km per phase (siemens/km)
G′ = shunt conductance/km per phase (siemens/km)
as depicted in Figure 2.1. The parameters above are specific for the line
or cable configuration and are dependent on conductors and geometrical
arrangements.
From the circuit in Figure 2.1 the telegraph equation is derived, and from
this the lumped-circuit line model for symmetrical steady state conditions,
Figure 2.2. This model is frequently referred to as the π-model, and it is
characterized by the parameters
Zkm = Rkm + jXkm = series impedance (Ω)
sh = Gsh + jB sh = shunt admittance (siemens)
Ykm 1
km km
1
In Figure 2.2 the two shunt elements are assumed to be equal, which is true for
homogenous lines, i.e. a line with equal values of the line parameters along its length, but
this might not be true in the general case. In such a case the shunt elements are replaced
sh sh sh sh
by Ykm and Ymk with Ykm 6= Ymk with obvious notation.
2.1. Lines and Cables 7

zkm
I km I mk
k m

sh
sh
ykm ykm

Figure 2.2. Lumped-circuit model (π-model) of a transmission line

between nodes k and m.

Note. In the following most analysis will be made in the p.u. system. For
impedances and admittances, capital letters indicate that the quantity is ex-
pressed in ohms or siemens, and lower case letters that they are expressed
in p.u.

Note. In these lecture notes complex quantities are not explicitly marked
as underlined. This means that instead of writing z km we will write zkm
when this quantity is complex. However, it should be clear from the context
if a quantity is real or complex. Furthermore, we will not always use specific
type settings for vectors. Quite often vectors will be denoted by bold face type
setting, but not always. It should also be clear from the context if a quantity
is a vector or a scalar.

When formulating the network equations the node admittance matrix

will be used and the series admittance of the line model is needed
−1
ykm = zkm = gkm + jbkm (2.1)
with
rkm
gkm = 2 + x2 (2.2)
rkm km
and
xkm
bkm = − 2 (2.3)
rkm + x2km
For actual transmission lines the series reactance xkm and the series
resistance rkm are both positive, and consequently gkm is positive and bkm
is negative. The shunt susceptance ykmsh and the shunt conductance g sh are
km
sh is so
both positive for real line sections. In many cases the value of gkm
small that it could be neglected.
8 2. Network Models

The complex currents Ikm and Imk in Figure 2.2 can be expressed as
functions of the complex voltages at the branch terminal nodes k and m:
sh
Ikm = ykm (Ek − Em ) + ykm Ek (2.4)
sh
Imk = ykm (Em − Ek ) + ykm Em (2.5)
where the complex voltages are
Ek = Uk ejθk (2.6)
Em = Um ejθm (2.7)
This can also be written in matrix form as
sh
    
Ikm ykm + ykm −ykm Ek
= sh (2.8)
Imk −ykm ykm + ykm Em
As seen the matrix on the right hand side of eq. (2.8) is symmetric and
the diagonal elements are equal. This reflects that the lines and cables are
symmetrical elements.
Example 2.1. The series impedance of a 138 kV transmission line section
is
z = r + jx = 0.0062 + j0.0360 p.u.
The total shunt susceptance (double the susceptance that appears in the
equivalent π-model) is
bsh = 0.0105 p.u.
and the shunt conductance is ignored. Calculate the series conductance and
series susceptance and the ratio between the series and shunt susceptances.

Solution The series conductance is given by

r 0.0062
g= = = 4.64 p.u.
r2 +x2 0.00622 + 0.03602
and the series susceptance by
x 0.0360
b=− =− = −27.0 p.u.
r2 +x2 0.00622 + 0.03602
The b/bsh ratio is given by
−27.0
b/bsh = = −2596
0.0105
Of interest is also the ratio x/r
0.036
x/r = = 5.8
0.0062

2.2. Transformers 9

Example 2.2. The series impedance and the total shunt impedance of a
750 kV line section are

z = r + jx = 0.00072 + j0.0175 p.u.

bsh = 8.77 p.u.

Calculate the series conductance and series susceptance and the ratio be-
tween the series and shunt susceptances and the x/r ratio.

Solution The series conductance and susceptance are given by

0.00072
g= = 2.35 p.u.
0.000722 + 0.01752
0.0175
b=− = −57 p.u.
0.000722 + 0.01752
The x/r ratio and b/bsh ratio are
0.0175
x/r = = 24.3
0.00072
−57
b/bsh = = −6.5
8.77


Note. The 750 kV line has a much higher x/r ratio than the 138 kV line
and, at the same time, a much smaller (in magnitude) b/bsh ratio. (Why?)
Higher x/r ratios mean better decoupling between active and reactive parts
of the power flow problem, while smaller (in magnitude) b/bsh ratios may
indicate the need for some sort of compensation, shunt or series, or both.

2.2 Transformers
We will start with a simplified model of a transformer where we neglect the
magnetizing current and the no-load losses 2 . In this case the transformer
can be modelled by an ideal transformer with turns ratio tkm in series with
a series impedance zkm which represents resistive (load-dependent) losses
and the leakage reactance, see Figure 2.3. Depending on if tkm is real or
non-real (complex) the transformer is in-phase or phase-shifting.
2
Often the magnetizing current and no-load losses are modelled by a shunt impedance,
with much higher impedance than the leakage impedance. The inductive part of this
impedance is then determined by the value of the magetizing current and the resistive
part by the no load losses.
10 2. Network Models

Uk e jqk Up e jqp Um e jqm

1:tkm zkm
k m
p
Ikm Imk

Uk e jqk Up e jqp Um e jqm

tkm:1 zkm
k m
p
Ikm Imk

Figure 2.3. Transformer model with complex ratio tkm = akm ejϕkm
(tkm = a−1
km e
−jϕkm
)

Uk e jqk Up e jqp Um e jqm

1:akm zkm
k m
p
Ikm Imk

Figure 2.4. In-phase transformer model

2.2.1 In-Phase Transformers

Figure 2.4 shows an in-phase transformer model indicating the voltage at the
internal – non-physical – node p. In this model the ideal voltage magnitude
ratio (turns ratio) is
Up
= akm (2.9)
Uk
Since θk = θp , this is also the ratio between the complex voltages at nodes
k and p,
Ep Up ejθp
= = akm (2.10)
Ek Uk ejθk
There are no power losses (neither active nor reactive) in the ideal trans-
former (the k-p part of the model), which yields
∗ ∗
Ek Ikm + Ep Imk =0 (2.11)

Then applying eqs. (2.9) and (2.10) gives

Ikm |Ikm |
=− = −akm , (2.12)
Imk |Imk |
2.2. Transformers 11

A
k
Ikm Imk

B C

Figure 2.5. Equivalent π-model for in-phase transformer

which means that the complex currents Ikm and Imk are out of phase by
180◦ since akm ∈ R.
Figure 2.5 represents the equivalent π-model for the in-phase transformer
in Figure 2.4. Parameters A, B, and C of this model can be obtained by
identifying the coefficients of the expressions for the complex currents Ikm
and Imk associated with the models of Figures 2.4 and 2.5. Figure 2.4 gives

Ikm = −akm ykm (Em − Ep ) = (a2km ykm )Ek + (−akm ykm )Em (2.13)

Imk = ykm (Em − Ep ) = (−akm ykm )Ek + (ykm )Em (2.14)

or in matrix form
   2  
Ikm akm ykm −akm ykm Ek
= (2.15)
Imk −akm ykm ykm Em

As seen the matrix on the right hand side of eq. (2.15) is symmetric, but
the diagonal elements are not equal when a2km 6= 1. Figure 2.5 provides now
the following:
Ikm = (A + B)Ek + (−A)Em (2.16)
Imk = (−A)Ek + (A + C)Em (2.17)
or in matrix form
    
Ikm A+B −A Ek
= (2.18)
Imk −A A+C Em

Identifying the matrix elements from the matrices in eqs. (2.15) and (2.18)
yields
A = akm ykm (2.19)
B = akm (akm − 1)ykm (2.20)
C = (1 − akm )ykm (2.21)
12 2. Network Models

Example 2.3. A 138/69 kV in-phase transformer with a series resistance

of zero, a 0.23 p.u. series reactance, and p.u. turns ratio of 1 : 1.030 (from
the model in Figure 2.4) Calculate the equivalent π-model parameters.

Solution
A = akm ykm = 1.030(j0.230)−1 = −j4.48 p.u.
B = akm (akm − 1)ykm = 1.030(1.030 − 1)(j0.230)−1 = −j0.13 p.u.
C = (1 − akm )ykm = (1 − 1.030)(j0.230)−1 = j0.13 p.u.
Hence, since A, B, and C denote admittances, A and B are inductive, and
C is capacitive. 

Example 2.4. A 500/750 kV in-phase transformer with a series resistance

of zero, a 0.00623 p.u. series reactance, and p.u. turns ratio of 1 : 0.950
(from the model in Figure 2.4) Calculate the parameters for the equivalent
π-model. (100 MVA base)

Solution As in the previous example one obtains:

A = −j152.5 p.u.

B = j7.62 p.u.
C = −j8.03 p.u.
i.e., parameter B is capacitive and parameter C is inductive (762 Mvar
and −803 Mvar, respectively, assuming nominal voltage magnitudes and a
100 MVA base). 

2.2.2 Phase-Shifting Transformers

Phase-shifting transformers, such as the one represented in Figure 2.6, are
used to control active power flows; the control variable is the phase angle
and the controlled quantity can be, among other possibilities, the active
power flow in the branch where the shifter is placed. In Appendix A the
physical design of phase-shifting transformer is described.
A phase-shifting transformer affects both the phase and magnitude of
the complex voltages Ek and Ep , without changing their ratio, i.e.,
Ep
= tkm = akm ejϕkm (2.22)
Ek
Thus, θp = θk + ϕkm and Up = akm Uk , using eqs. (2.11) and (2.22),
Ikm
= −t∗km = −akm e−jϕkm (2.23)
Imk
2.2. Transformers 13

Uk e jqk Um e jqm
1:akme jjkm zkm
k m
p
Ikm Imk

qp = qk + jkm

Up = akmUk

Figure 2.6. Phase-shifting transformer with tkm = akm ejϕkm .

As with in-phase transformers, the complex currents Ikm and Imk can
be expressed in terms of complex voltages at the phase-shifting transformer
terminals:

Ikm = −t∗km ykm (Em − Ep ) = (a2km ykm )Ek + (−t∗km ykm )Em (2.24)

Imk = ykm (Em − Ep ) = (−tkm ykm )Ek + (ykm )Em (2.25)

or in matrix form
  2
akm ykm −t∗km ykm
  
Ikm Ek
= (2.26)
Imk −tkm ykm ykm Em

As seen this matrix is not symmetric if tkm is non-real, and the diagonal
matrix elements are not equal if a2km 6= 1. There is no way to determine
parameters A, B, and C of the equivalent π-model from these equations,
since the coefficient −t∗km ykm of Em in eq. (2.24) differs from −tkm ykm in
eq. (2.25), as long as there is nonzero phase shift, i.e. tkm ∈/ R. A phase-
shifting transformer can thus not be represented by a π-model.

Example 2.5. A 230/138 kV transformer (Figure 2.6) has a series resis-

tance of zero, a 0.0127 p.u. series reactance, and a complex turns ratio of
◦
1 : 1.007ej30 (Y-∆ connection).
Show that this transformer can be seen to consist of a series connection
of two transformers: an ideal in-phase transformer with a turns ratio of
1 : 1.007 (constant voltage phase) and a phase-shifting transformer with a
◦
complex turns ratio of 1 : ej30 (constant voltage amplitude) and a series
reactance of 0.0127 p.u.

Note. If no parallel paths exist, the phase-shifting has no significance. The

introduced phase-shift can in such a case be seen as a shift of the phase
angle of the reference node. Y-∆ connected transformers are often used to
14 2. Network Models

1:1
XtA = 0.2 p.u. A
Iload

1:t12
XtB = 0.4 p.u B E2
Load

Figure 2.7. System for Example 2.6.

provide zero-sequence de-coupling between two parts of the system, and not
for active power flow. For active power flow control usually phase-shifting
much lower than 30◦ is needed. Often the phase-shifting could be varied to
cope with different loading situations in the system.

Example 2.6. Consider two transformers connected in parallel according

to Figure 2.7. Transformer A has a turns-ratio of 1:1 (p.u./p.u.) while the
turns-ratio of transformer B will be varied as described below. The trans-
formers are feeding a load at bus 2, Iload = 1.05∠ − 45◦ p.u. and E2 = 1∠0◦
p.u. The reactances of the transformers are given in the figure.
Calculate the complex power through the transformers when the turns-
ratios of transformer B are (t12 = a12 ejϕ12 )

(a) a12 = 1, ϕ12 = 0◦

(b) a12 = 1.05, ϕ12 = 0◦

(c) a12 = 1, ϕ12 = 3◦

Comments!

2.2.3 Unified Branch Model

The expressions for the complex currents Ikm and Imk for both transformers
and shifters derived above depend on the side where the tap is located; i.e.,
they are not symmetrical. It is however possible to develop unified complex
expressions which can be used for lines, transformers, and phase-shifters,
regardless of the side on which the tap is located (or even in the case when
there are taps on both sides of the device). Consider initially the model
2.2. Transformers 15

in Figure 2.8 in which shunt elements have been temporarily ignored and
tkm = akm ejϕkm and tmk = amk ejϕmk . In this case

Ikm = t∗km Ipq = t∗km (Ep − Eq )ykm = t∗km (tkm Ek − tmk Em )ykm (2.27)

and

Imk = t∗mk Iqp = t∗mk (Eq − Ep )ykm = t∗mk (tmk Em − tkm Ek )ykm (2.28)

which together yield

Ikm = (a2km Ek − t∗km tmk Em )ykm (2.29)

and
Imk = (a2mk Em − t∗mk tkm Ek )ykm (2.30)
(These expressions are symmetrical in the sense that if k and m are inter-
changed, as in the expression for Ikm , the result is the expression for Imk ,
and vice-versa.)
Figure 2.9 shows the unified branch model. All the devices studied above
can be derived from this general model by establishing the appropriate def-
initions of the parameters that appear in the unified model. Thus, for in-
stance, if tkm = tmk = 1 is assumed, the result is an equivalent π-model of
a transmission line; or, if the shunt elements are ignored, and tkm = 1 and
tmk = amk e−jϕmk is assumed, then the result is a phase shifting transformer
with the tap located on the bus m side. The general expressions for and Imk
can be obtained from the model in Figure 2.9:

Ikm = (a2km Ek − t∗km tmk Em )ykm + ykm

sh 2
akm Ek (2.31)

and
Imk = (a2mk Em − t∗mk tkm Ek )ykm + ymk
sh 2
amk Em (2.32)
or
a2km (ykm + ykmsh
−t∗km tmk ykm
    
Ikm ) Ek
= (2.33)
Imk −t∗mk tkm ykm a2mk (ykm + ykm
sh
) Em

The transformers modelled above were all two-winding transformers. In

power systems there are also three-winding and N -winding (N > 3) trans-
formers, and they can be modelled in a similar way. Instead of linear rela-
tionships between two complex currents and two complex voltages, we will
have linear relationships between these quantities modelled by N × N ma-
trices. It should be notes that for N -winding transformers there are leakage
reactances between each pair of windings, i.e. in total N (N − 1)/2 reac-
tances. Since the power ratings of the different windings are not necessarily
equal, as in the two-winding case, it should be noted that the power bases
for expressing the reactances in p.u. are not always equal.
16 2. Network Models

Uk e jqk Up e jqp Uq e jqk Um e jqm

zkm
k m
p q
Ikm 1:tkm tmk:1 Imk

Figure 2.8. Transformer symmetrical model.

Uk e jqk Up e jqp Uq e jqq Um e jqm

1:tkm p zkm q tmk:1
k m
Ikm Imk

sh sh
ykm ymk

Figure 2.9. Unified branch model extended (π-model).

2.3 Shunt Elements

The modelling of shunt elements in the network equations is straightforward
and the main purpose here is to introduce the notation and the sign con-
vention to be used when formulating the network equations in the coming
chapters. As seen from Figure 2.10 the current from a shunt is defined as
positive when injected into the bus. This means that

Iksh = −yksh Ek (2.34)

with Ek being the complex voltage at node k. Shunts are in all practical cases
either shunt capacitors or reactors. From eq. (2.34) the injected complex
power is
sh ∗ sh ∗ 2
S sh sh sh 2



k = P k + jQ k = − y k |Ek | = − y k Uk (2.35)

2.4 Loads
Load modelling is an important topic in power system analysis. When for-
mulating the load flow equations for high voltage systems, a load is most
2.5. Generators 17

sh
Ik
k

yksh

Figure 2.10. A shunt connected to bus k.

often the infeed of power to a network at a lower voltage level, e.g. a dis-
tribution network. Often the voltage in the distribution systems is kept
constant by controlling the tap-positions of the distribution transformers
which means that power, active and reactive, in most cases can be regarded
as independent of the voltage on the high voltage side. This means that the
complex power Ek (Ikload )∗ is constant, i.e. independent of the voltage mag-
nitude Uk . Also in this case the current is defined as positive when injected
into the bus, see Figure 2.11. In the general case the complex load current
can be written as
Ikload = Ikload (Uk ) (2.36)
where the function Ikload (·) describes the load characteristics.3 More often
the load characteristics are given for the active and reactive powers

Pkload = Pkload (Uk ) (2.37)

Qload
k = Qload
k (Uk ) (2.38)

2.5 Generators
Generators are in load flow analysis modelled as current injections, see Fig-
ure 2.12. In steady state a generator is commonly controlled so that the
active power injected into the bus and the voltage at the generator termi-
nals are kept constant. This will be elaborated later when formulating the
load flow equations. Active power from the generator is determined by the
turbine control and must of course be within the capability of the turbine-
generator system. Voltage is primarily determined by the reactive power
3
This refers to the steady state model of the load. For transient conditions other
load models apply. These are usually formulated as differential equations and might also
involve the frequency.
18 2. Network Models

load
Ik
k

Load

Figure 2.11. Model of a load connected to bus k.

gen
Ik
k

Generator

Figure 2.12. Model of a generator connected to bus k.

injection into the node, and since the generator must operate within its
reactive capability curve it is not possible to control the voltage outside cer-
tain limits. The reactive capability of a generator depends on a number of
quantities, such as active power, bus voltage and other operating conditions,
and a typical example is shown in Figure 2.13. The shape of the generator
capability curve is specific for each generator and depends on design char-
acteristics, type of generator, hydro or steam turbine, stability constraints,
etc. In Figure 2.13 it is also indicated what imposes the different limits for
this particular generator. These are briefly discussed below. These limits
are also discussed in Chapter 9.

2.5.1 Stator Current Heating Limit

The losses in the armature windings are given by Rt It2 , with obvious no-
tation. These losses result in a temperature rise in the armature wind-
ings, and this must be limited to a given value otherwise the generator is
2.5. Generators 19

Q Field current
heating limit

Overexcited

Stator current
heating limit

P
Underexcited

End region
heating limit

Figure 2.13. Reactive capability curve of a turbo generator.

damaged or its life time is reduced. Since the complex power is given by
S = P + jQ = Ut It∗ it means that for a given terminal voltage, Ut , circles
in the P − Q-plane with centre at the origin correspond to constant value
of the magnitude of the armature current It . The stator current limit for a
given terminal voltage is thus a circle with the centre at the origin. At high
loading of the generator, this is usually determining the reactive capability
of the synchronous machine.

2.5.2 Field Current Heating Limit

The reactive power that can be generated at low load is determined by the
field current heating limit. It can be shown that the locus for constant field
current is a circle with the centre on the Q-axis at −Et2 /Xs , where Et is the
terminal voltage and Xs is the synchronous reactance. The radius is given
by machine parameters and typical behaviour is shown in Figure 2.13. The
field current heating is usually limiting at overexcited operation at low load.
20 2. Network Models

2.5.3 Stator End Region Heating Limit

When the synchronous machine is underexcited the armature end leakage
flux is increased. This flux enters and leaves in a direction perpendicular to
the stator laminations causing eddy currents in the laminations, and hence
heating. This can limit the capability, particularly for round rotor machines.
A typical example of such a limitation is shown in Figure 2.13.
The limits on reactive power generation and absorption given in this
subsection and the two previous ones are imposed by internal synchronous
machine design. As seen they are dependent on the terminal voltage of the
generator. There are also other considerations that can further limit the re-
active capability of the synchronous machine. These are stability limits that
could further limit the operation of the machine in underexcited operation.
3
Active and Reactive Power Flows

In this chapter the expressions for the active and reactive power flows in
transmission lines, transformers, phase-shifting transformers, and unified
branch models are derived.

T HE SYSTEM COMPONENTS dealt with in this chapter are linear in

the sense that the relations between voltages and currents are linear1 .
However, since one usually is interested rather in powers, active and reactive,
than currents, the resulting equations will be non-linear, which introduces
a complication when solving the resulting equations.

3.1 Transmission Lines

Consider the complex current Ikm in a transmission line

Ikm = ykm (Ek − Em ) + jbsh

km Ek (3.1)

with quantities defined according to Figure 2.2. The complex power, Skm =
Pkm + jQkm , is
∗ ∗
Skm = Ek Ikm = ykm Uk ejθk (Uk e−jθk − Um e−jθm ) − jbsh 2
km Uk (3.2)

where the conductance of ykmsh has been neglected.

The expressions for Pkm and Qkm can be determined by identifying the
corresponding coefficients of the real and imaginary parts of eq. (3.2), which
yields

Pkm = Uk2 gkm − Uk Um gkm cos θkm − Uk Um bkm sin θkm (3.3)
Qkm = −Uk2 (bkm + bsh
km ) + Uk Um bkm cos θkm − Uk Um gkm sin θkm (3.4)

where the notation θkm = θk − θm is introduced.

1
This is at least true for the models analysed here. Different non-linear phenomena,
e.g. magnetic saturation, can sometimes be important, but when studying steady state
conditions the devices to be discussed in this chapter are normally within the region of
linearity.

21
22 3. Active and Reactive Power Flows

The active and reactive power flows in opposite directions, Pmk and Qmk ,
can be obtained in the same way, resulting in:
2
Pmk = Um gkm − Uk Um gkm cos θkm + Uk Um bkm sin θkm (3.5)
2
Qmk = −Um (bkm + bsh
km ) + Uk Um bkm cos θkm + Uk Um gkm sin θkm (3.6)

From these expressions the active and reactive power losses of the lines
are easily obtained as:

Pkm + Pmk = gkm (Uk2 + Um

2
− 2Uk Um cos θkm )
= gkm |Ek − Em |2 (3.7)

Qkm + Qmk = −bsh 2 2 2 2

km (Uk + Um ) − bkm (Uk + Um − 2Uk Um cos θkm )
= −bsh 2 2
km (Uk + Um ) − bkm |Ek − Em |
2
(3.8)

Note that |Ek − Em | represents the magnitude of the voltage drop across
the line, gkm |Ek − Em |2 represents the active power losses, −bsh
km |Ek − Em |
2

represents the reactive power losses; and −bsh 2 2

km (Uk + Um ) represents the
reactive power generated by the shunt elements of the equivalent π-model
(assuming actual transmission line sections, i.e. with bkm < 0 and bsh
km > 0).

Example 3.1. A 750 kV transmission line section has a series impedance

of 0.00072 + j0.0175 p.u., a total shunt impedance of 8.775 p.u., a voltage
magnitude at the terminal buses of 0.984 p.u. and 0.962 p.u., and a voltage
angle difference of 22◦ . Calculate the active and reactive power flows.

Solution The active and reactive power flows in the line are obtained by
applying eqs. (3.3) and (3.4), where Uk = 0.984 p.u., Um = 0.962 p.u., and
θkm = 22◦ . The series impedance and admittances are as follows:

zkm = 0.00072 + j0.0175 p.u.

−1
ykm = gkm + jbkm = zkm = 2.347 − j57.05 p.u.
The π-model shunt admittances (100 MVA base) are:

bsh
km = 8.775/2 = 4.387 p.u.

and

Pkm = 0.9842 ·2.347−0.984·0.962·2.347 cos 22◦ +00.984·0.962·57.05 sin 22◦ p.u.

Qkm = −0.9842 ·(−57.05+4.39)−0.984·0.962·57.05 cos 22◦ −00.984·0.962·2.347 sin 22◦ p.u.

which yield
Pkm = 2044 MW Qkm = 8.5 Mvar
3.2. In-phase Transformers 23

In similar way one obtains:

Pmk = −2012 MW Qmk = −50.5 Mvar

It should be noted that powers are positive when injected into the line. 

3.2 In-phase Transformers

The complex current Ikm in an in-phase transformer is expressed as in
eq. (2.13)
Ikm = akm ykm (akm Ek − Em )
The complex power, Skm = Pkm + jQkm , is given by
∗ ∗
Skm = Ek Ikm = ykm akm Uk ejθk (akm Uk e−jθk − Um e−jθm ) (3.9)

Separating the real and imaginary parts of this latter expression yields
the active and reactive power flow equations:

Pkm = (akm Uk )2 gkm − akm Uk Um gkm cos θkm − akm Uk Um bkm sin θkm
(3.10)
Qkm = −(akm Uk )2 bkm + akm Uk Um bkm cos θkm − akm Uk Um gkm sin θkm
(3.11)

These same expressions can be obtained by comparing eqs. (3.9) and

(3.2); in eq. (3.9) the term jbsh 2
km Uk is not present, and Uk is replaced by
akm Uk . Hence, the expressions for the active and reactive power flows on
in-phase transformers are the same expressions derived for a transmission
line, except the for two modifications: ignore bsh
km , and replace Uk with
akm Uk .

Example 3.2. A 500/750 kV transformer with a tap ratio of 1.050:1.0

on the 500 kV side, see Figure 2.4, has negligible series resistance and a
leakage reactance of 0.00623 p.u., terminal voltage magnitudes of 1.023 p.u.
and 0.968 p.u., and an angle spread of 5.3◦ . Calculate the active and reactive
power flows in the transformer.

Solution The active and reactive power flows in the transformer are given
by eqs. (3.10) and (3.11), where Uk = 1.023 p.u., Um = 0.968 p.u., θkm =
5.3◦ , and akm = 1.0/1.05 = 0.9524. The series reactance and susceptance
are as follows:
xkm = 0.00623 p.u.
bkm = −x−1
km = −160.51 p.u.
24 3. Active and Reactive Power Flows

The active and reactive power flows can be expressed as

Pkm = −0.9524 · 1.023 · 0.968 · (−160.51) sin 5.3◦ p.u.

Qkm = −(0.9524·1.023)2 (−160.51)+0.9524·1.023·0.968·(−160.51) cos 5.3◦ p.u.

which yield
Pkm = 1398 MW Qkm = 163 Mvar
The reader is encouraged to calculate Pmk and Qmk . (The value of Pmk
should be obvious.) 

3.3 Phase-Shifting Transformer with akm = 1

The complex current Ikm in a phase shifting transformer with akm = 1 is as
follows, see Figure 2.6:

Ikm = ykm (Ek − e−jϕkm Em ) = ykm e−jϕkm (Ek ejϕkm − Em ) (3.12)

and the complex power, Skm = Pkm + jQkm , is thus

∗ ∗
Skm = Ek Ikm = ykm Uk ej(θk +ϕkm ) (Uk e−j(θk +ϕkm ) − Um e−jθm ) (3.13)

Separating the real and imaginary parts of this expression, yields the
active and reactive power flow equations, respectively:

Pkm = Uk2 gkm − Uk Um gkm cos(θkm + ϕkm )

− Uk Um bkm sin(θkm + ϕkm ) (3.14)

Qkm = −Uk2 bkm + Uk Um bkm cos(θkm + ϕkm )

− Uk Um gkm sin(θkm + ϕkm ) (3.15)

As with in-phase transformers, these expressions could have been ob-

tained through inspection by comparing eqs. (3.2) and (3.13): in eq. (3.13),
the term jbsh 2
km Uk is not present, and θkm is replaced with θkm + ϕkm . Hence,
the expressions for the active and reactive power flows in phase-shifting
transformers are the same expressions derived for the transmission line, al-
beit with two modifications: ignore bsh
km and replace θkm with θkm + ϕkm .

Example 3.3. A ∆-Y , 230/138 kV transformer presents a 30◦ phase an-

gle shift. Series resistance is neglected and series reactance is 0.0997 p.u.
Terminal voltage magnitudes are 0.882 p.u. and 0.989 p.u., and the total
angle difference is −16.6◦ . Calculate the active and reactive power flows in
the transformer.
3.4. Unified Power Flow Equations 25

Solution The active and reactive power flows in the phase-shifting trans-
former are given by eqs. (3.14) and (3.15), where Uk = 0.882 p.u., Um =
0.989 p.u., θkm = −16.6◦ , and ϕkm = 30◦ . The series reactance and suscep-
tance are as follows:
xkm = 0.0997 p.u.
bkm = −x−1
km = −10.03 p.u.

The active and reactive power flows can be expressed as

Pkm = −0.882 · 0.989 · (−10.03) · (−160.51) sin(−16.6◦ + 30◦ ) p.u.

Qkm = −0.8822 (−10.03) + 0.882 · 0.989 · (−10.03) cos(−16.6◦ + 30◦ ) p.u.

which yield
Pkm = 203 MW Qkm = −70.8 Mvar
The reader is encouraged to calculate Pmk and Qmk . (The value of Pmk
should be obvious.) 

3.4 Unified Power Flow Equations

The expressions for active and reactive power flows on transmission lines,
in-phase transformers, and phase shifting transformers, see Figure 2.9, can
be expressed in the following unified forms:

Pkm = (akm Uk )2 gkm

− (akm Uk )(amk Um )gkm cos(θkm + ϕkm − ϕmk )
− (akm Uk )(amk Um )bkm sin(θkm + ϕkm − ϕmk ) (3.16)

Qkm = (akm Uk )2 (bkm + bsh

km )
+ (akm Uk )(amk Um )bkm cos(θkm + ϕkm − ϕmk )
− (akm Uk )(amk Um )gkm sin(θkm + ϕkm − ϕmk ) (3.17)

Where, for the transmission lines like the one represented in Figure 2.2,
akm = amk = 1 and ϕkm = ϕmk = 0; for in-phase transformers such as the
sh = y sh = 0, a
one represented in Figure 2.4, ykm mk mk = 1 and ϕkm = ϕmk = 0;
sh =
and for a phase-shifting transformer such as the one in Figure 2.6, ykm
sh
ymk = 0, amk = 1 and ϕmk = 0.
26 3. Active and Reactive Power Flows
4
Nodal Formulation of the Network Equa-
tions

In this chapter the basic network equations are derived from Kirchhoff ’s
Current Law (KCL) and put into forms that are suitable for the formulation
of the power flow equations in the subsequent chapter

T HE NET COMPLEX current injection at a network bus, see Figure 4.1,

is related to the current flows in the branches connected to the bus.
Applying Kirchhoff’s Current Law (KCL) yields
X
Ik + Iksh = Ikm , for k = 1, . . . , N (4.1)
m∈Ωk
where k is a generic node, Ik is the net current injection from generators
and loads, Iksh is the current injection from shunts, m is a node adjacent to
k, Ωk is the set of nodes adjacent to k, and N is the number of nodes in the
network.
The complex current Ikm in the unified branch model, Figure 2.9, is
Ikm = (a2km Ek − t∗km tmk Em )ykm + ykm
sh 2
akm Ek (4.2)
where tkm = akm ejϕkm and tmk = amk ejϕmk .

Ik Ikm

sh
yksh Ik

Figure 4.1. Generic bus with sign conventions for currents and power flows.

27
28 4. Nodal Formulation of the Network Equations

Equations (4.1) and (4.2) yield

 X  X
Ik = yksh + sh
a2km (ykm + ykm ) Ek − (t∗km tmk ykm )Em (4.3)
m∈Ωk m∈Ωk

for k = 1, . . . , N . This expression can be written as

I = YE (4.4)
where

• I is the injection vector with elements Ik , k = 1, . . . , N

• E is the nodal voltage vector with elements Ek = Uk ejθk

• Y = G + jB is the nodal admittance matrix, with the following ele-

ments
Ykm = −t∗km tmk ykm (4.5)
X
Ykk = yksh + sh
a2km (ykm + ykm ) (4.6)
m∈Ωk

We see that the nodal admittance matrix defined by eqs. (4.5) and (4.6)
is modified as compared with the nodal admittance matrix without trans-
formers. Particularly it should be noted that Y as defined above is not
necessarily symmetric.
For large practical networks this matrix is usually very sparse. The
degree of sparsity (percentage of zero elements) normally increases with
the dimensions of the network: e.g., a network with 1000 buses and 1500
branches typically presents a degree of sparsity greater than 99 %, i.e. less
than 1 % of the matrix elements have non-zero values.
The kth component of I, Ik , defined in eq. (4.3), can, by using eqs. (4.5)
and (4.6), be written as
X X
Ik = Ykk Ek + Ykm Em = Ykm Em (4.7)
m∈Ωk m∈K

where K is the set of buses adjacent to bus k, including bus k, and Ωk is

the set of buses adjacent to bus k, excluding bus k. Now considering that
Ykm = Gkm + jBkm and Em = Um ejθm , eq. (4.7) can be rewritten as
X
Ik = (Gkm + jBkm )(Um ejθm ) (4.8)
m∈K

The complex power injection at bus k is

Sk = Pk + jQk = Ek Ik∗ (4.9)

 29

and by applying eqs. (4.8) and (4.9) this gives

X
Sk = Uk ejθk (Gkm − jBkm )(Um e−jθm ) (4.10)
m∈K

The expressions for active and reactive power injections are obtained by
identifying the real and imaginary parts of eq. (4.10), yielding
X
Pk = Uk Um (Gkm cos θkm + Bkm sin θkm ) (4.11)
m∈K
X
Qk = Uk Um (Gkm sin θkm − Bkm cos θkm ) (4.12)
m∈K
30 4. Nodal Formulation of the Network Equations
5
Basic Power Flow Problem

In this chapter the basic power flow problem is formulated and the basic
bus types are defined. Also, the conditions for solvability of the problem are
discussed

T HE POWER FLOW PROBLEM can be formulated as a set of non-

linear algebraic equality/inequality constraints. These constraints rep-
resent both Kirchhoff’s laws and network operation limits. In the basic
formulation of the power flow problem, four variables are associated to each
bus (network node) k:

• Uk : voltage magnitude

• θk : voltage angle

• Pk : net active power (algebraic sum of generation and load)

• Qk : net reactive power (algebraic sum of generation and load)

5.1 Basic Bus Types

Depending on which of the above four variables are known (given) and which
ones are unknown (to be calculated), two basic types of buses can be defined:

• PQ bus: Pk and Qk are specified; Uk and θk are calculated

• PU bus: Pk and Uk are specified; Qk and θk are calculated

PQ buses are normally used to represent load buses without voltage control,
and PU buses are used to represent generation buses with voltage control
in power flow calculations1 . Synchronous compensators2 are also treated as
PU buses. A third bus is also needed:
1
Synchronous machines are often equipped with Automatic Voltage Regulators (AVRs),
which controls the excitation of the machine so that the terminal voltage, or some other
voltage close to the machine, is kept at the set value.
2
Synchronous compensators, sometimes also called synchronous condensers, are syn-
chronous machines without any active power generation or load (except for losses) used
for reactive power and voltage control.

31
32 5. Basic Power Flow Problem

• Uθ bus: Uk and θk are specified; Pk and Qk are calculated

The Uθ bus, also called reference bus or slack bus, has double functions in
the basic formulation of the power flow problem:

1. It serves as the voltage angle reference

2. Since the active power losses are unknown in advance, the active power
generation of the Uθ bus is used to balance generation, load, and losses

In “normal” power systems PQ-buses or load buses are the far most common,
typically comprising more than 80% of all buses.
Other possible bus types are P, U, and PQU, with obvious definitions.
The use of multiple Uθ buses may also be required for certain applications.
In more general cases, the given values are not limited to the specific set of
buses (P, Q, U, θ), and branch related variables can also be specified.

Example 5.1. Figure 5.1 shows a 5-bus network with four transmission
lines and two transformers. Generators, with voltage control, are connected
at buses 1, 3, and 5, and loads are connected at buses 4 and 5, and at bus 4
a shunt is also connected. Classify the buses according to the bus types PU,
PQ and Uθ.

Solution Buses 1, 3, and 5 are all candidates for PU or Uθ bus types. Since
only one could be Uθ bus, we select (arbitrarily) bus 5 as Uθ. In a practical
system usually a generator, or generator station, that could produce power
within a large range is selected as reference or slack bus. It should be noted
that even if a load is connected to bus 5 it can only be a PU or Uθ bus, since
voltage control is available at the bus. The reference angle is set at bus 5,
usually to 0. Bus 2 is a transition bus in which both P and Q are equal to
zero, and this bus is consequently of type PQ. Bus 4 is a load bus to which
is also connected a shunt susceptance: since shunts are modelled as part of
the network, see next section, the bus is also classified as a PQ bus. 

5.2 Equality and Inequality Constraints

Eqs. (4.11) and (4.12) can be rewritten as follows
X
Pk = Pkm (Uk , Um , θk , θm ) (5.1)
m∈Ωk
X
Qk + Qsh
k (Uk ) = Qkm (Uk , Um , θk , θm ) (5.2)
m∈Ωk

where
5.2. Equality and Inequality Constraints 33

PU PQ PU
1 2 3

Uq PQ

5 4

Figure 5.1. 5-bus system

• k = 1, . . . , N (N is the number of buses in the network)

• Ωk : set of buses adjacent to k

• Uk , Um : voltage magnitudes at the terminal buses of branch k – m

• θk , θm : voltage angles at the terminal buses of branch k – m

• Pkm : active power flow from bus k to bus m

• Qkm : reactive power flow from bus k to bus m

• Qsh
k : component of reactive power injection due to the shunt element
at bus k (Qsh sh 2 sh
k = bk Um ), where bk is the shunt susceptance.
3

A set of inequality constraints imposes operating limits on variables such

as the reactive power injections at PU buses (generator buses), see section
2.5, and voltage magnitudes at PQ buses:

Qmin
k ≤ Qk ≤ Qmax
k (5.3)

Ukmin ≤ Uk ≤ Ukmax (5.4)

3
It is assumed here that all shunts are reactive without losses. If shunts with resistive
components should be included, then eq. (5.1) must be modified accordingly.
34 5. Basic Power Flow Problem

When no inequality constraints are violated, nothing is affected in the

power flow equations, but if a limit is violated, the bus status is changed
and it is enforced as an equality constraint at the limiting value. This
normally requires a change in bus type: if, for example, a Q limit of a PU
bus is violated, the bus is transformed into an PQ bus (Q is specified and
the U becomes a problem unknown). A similar procedure is adopted for
backing-off when ever appropriate. What is crucial is that bus type changes
must not affect solvability. Various other types of limits are also considered
in practical implementations, including branch current flows, branch power
flows, active power generation levels, transformer taps, phase shifter angles,
and area interchanges.

5.3 Problem Solvability

One problem in the definition of bus type (bus classification) is to guarantee

that the resulting set of power flow equations contains the same number of
equations as unknowns, as are normally necessary for solvability, although
not always sufficient. Consider a system with N buses, where NP U are of
type PU, NP Q are of type PQ, and one is of type Uθ. To fully specify
the state of the system we need to know the voltage magnitudes and volt-
age angles of all buses, i.e. in total 2N quantities. But the voltage angle
and voltage magnitude of the slack bus are given together with the voltage
magnitudes of NP U buses. Unknown are thus the voltage magnitudes of
the PQ buses, and the voltage angles of the PU and the PQ buses, giving
a total of NP U + 2NP Q unknown states. From the PU buses we get NP U
balance equations regarding active power injections, and from the PQ buses
2NP Q equations regarding active and reactive power injections, thus in total
NP U + 2NP Q equations, and hence equal to the number of unknowns, and
the necessary condition for solvability has been fulfilled.
Similar necessary conditions for solvability can be established when other
types of buses, such as P, U, and PQU buses, are used in the formulation of
the power flow problem.

Example 5.2. Consider again the 5-bus in Figure 5.1. Formulate the equal-
ity constraints of the system and the inequality constraints for the generator
buses.

Solution In this case N = 5, NP Q = 2, NP U = 2, and of course NU θ = 1.

The number of equations are thus: NP U + 2NP Q = 2 + 2 · 2 = 6, and these
5.3. Problem Solvability 35

are:

P1 = P12 + P15
P2 = P21 + P23 + P25
Q2 = Q21 + Q23 + Q25
P3 = P32 + P34
P4 = P43 + P45
Q4 + Qsh
4 = Q43 + Q45

In the above equations P1 , P2 , P3 , P4 , Q2 , and Q4 are given. All the other

quantities are functions of the bus voltage magnitudes and phase angles, of
which U1 , U3 , and U5 and θ5 are given. The other six, i.e. U2 , U4 , θ1 , θ2 , θ3 ,
θ4 , in total 6 unknowns, can be solved from the above equations, and from
these all power flows and injections can be calculated.
The inequality constraints of the generator buses are:

Qmin
1 ≤ Q1 ≤ Qmax
1
Qmin
3 ≤ Q3 ≤ Qmax
3
Qmin
5 ≤ Q5 ≤ Qmax
5

The reactive limits above are derived from the generator capability curves as
explained in section 2.5. For the slack bus it must also be checked that the
injected active and reactive powers are within the range of the generator. If
not, the power generation of the other generators must be changed or the
voltage settings of these. 
36 5. Basic Power Flow Problem
6
Solution of the Power Flow Problem

In this chapter the basic methods to solve the non-linear power flow equations
are reviewed. Solution methods based on the observation that active and
reactive power flows are not so strongly coupled are introduced.

I N ALL REALISTIC CASES the power flow problem cannot be solved an-
alytically, and hence iterative solutions implemented in computers must
be used. In this chapter we will review two solutions methods, Gauss iter-
ation with a variant called Gauss-Seidel iterative method, and the Newton-
Raphson method.

6.1 Solution by Gauss-Seidel Iteration

Consider the power flow equations (5.1) and (5.2) which could be written in
complex form as
X
∗ ∗
Sk = Ek Ykm Em , k = 1, 2, . . . , N (6.1)
m∈K

which is a the same as eq. (4.10). The set K is the set of buses adjacent
(connected) to bus k, including bus k, and hence shunt admittances are
included in the summation. Furthermore Ek = Uk ejθk . This equation can
be rewritten as
" #
1 Sk
X
Ek∗ = ∗ − ∗
Ykm ∗
Em , k = 1, 2, . . . , N (6.2)
Ykk Ek
m∈Ωk

where Ωk is the set of all buses connected to bus k excluding bus k. Taking
the complex conjugate of eq. (6.2) yields
" #
1 Sk∗ X
Ek = − Ykm Em , k = 1, 2, . . . , N (6.3)
Ykk Ek∗
m∈Ωk

37
38 6. Solution of the Power Flow Problem

Thus we get N − 1 algebraic (complex) equations in the complex variables

Ek in the form
E2 = h2 (E1 , E2 , . . . , EN )
E3 = h3 (E1 , E2 , . . . , EN ) (6.4)
..
.
EN = hN (E1 , E2 , . . . , EN )
where the functions hi are given by eq. (6.3). It is assumed here that bus
number 1 is the U θ bus, and hence E1 is given and we have no equation for
node 1. For PQ buses both the magnitude and angle of Ek are unknown,
while for PU buses only the angle is unknown. For PQ buses Sk is known,
while for PU buses only Pk is known. This will be discussed below in more
detail. In vector form eq. (6.4) can be written as
x = h(x) (6.5)
and based on this equation the following iterative scheme is proposed
xν+1 = h(xν ) , ν = 0, 1, . . . (6.6)
where the superscript indicates the iteration number. Thus starting with an
initial value x0 , the sequence
x0 , x1 , x2 , . . . (6.7)
is generated. If the sequence converges, i.e. xν → x∗ , then
x∗ = h(x∗ ) (6.8)
and x∗
is a solution of eq. (6.5).
In practice the iteration is stopped when the changes in xν become suf-
ficiently small, i.e. when the norm of ∆xν = xν+1 − xν is less than a pre-
determined value ε.
To start the iteration a first guess of x is needed. Usually, if no a priori
knowledge of the solution is known, one selects all unknown voltage magni-
tudes and phase angles equal to the ones of the reference bus, usually around
1 p.u. and phase angle = 0. This initial solution is often called a flat start.
The difference between Gauss and Gauss-Seidel iteration can be ex-
plained by considering eq. (6.6) with all components written out explicitly1
xν+1
2 = h2 (x1 , xν2 , . . . , xνN )
xν+1
3 = h3 (x1 , xν2 , . . . , xνN ) (6.9)
..
.
xν+1
N = hN (x1 , xν2 , . . . , xνN )
1
In this particular formulation x1 is the value of the complex voltage of the slack
bus and consequently known. For completeness we have included it as a variable in the
equations above, but it is actually known
6.2. Newton-Raphson Method 39

In carrying out the computation (normally by computer) we process the

equations from top to bottom. We now observe that when we solve for xν+1 3
we already know xν+1
2 . Since x ν+1
2 is presumably a better estimate than xν2 ,
it seems reasonable to use the updated value. Similarly when we solve for
xν+1
4 we can use the values of xν+1
2 and xν+1
3 . This is the line of reasoning
called the Gauss-Seidel iteration:

xν+1
2 = h2 (x1 , xν2 , . . . , xνN )
xν+1
3 = h3 (x1 , xν+1 ν
2 , . . . , xN ) (6.10)
..
.
xν+1
N = hN (x1 , xν+1 ν+1 ν
2 , . . . , xN −1 , xN )

It is clear that the convergence of the Gauss-Seidel iteration is faster than

the Gauss iteration scheme.
For PQ buses the complex power Sk is completely known and the cal-
culation of the right hand side of eq. (6.3) is well defined. For PU buses
however, Q is not defined but is determined so that the voltage magnitude
is kept at the specified value. In this case we have to estimate the reactive
power injection and an obvious choice is
" #
X
ν ν ∗ ∗ ν
Qk = ℑ Ek Ykm (Em ) (6.11)
m∈K

In the Gauss-Seidel iteration scheme one should use the latest calculated
values of Em . It should be clear that also for PU buses the above iteration
scheme gives a solution if it converges.
A problem with the Gauss and Gauss-Seidel iteration schemes is that
convergence can be very slow, and sometimes even the iteration does not
converge despite that a solution exists. Furthermore, no general results are
known concerning the the convergence characteristics and criteria. There-
fore more efficient solution methods are needed, and one such method that
is widely used in power flow computations is discussed in the subsequent
sections.

6.2 Newton-Raphson Method

Before applying this method to the power flow problem we review the iter-
ation scheme and some of its properties.
A system of nonlinear algebraic equations can be written as

f (x) = 0 (6.12)

where x is an n-vector of unknowns and f is an n-vector function of x. Given

an appropriate starting value x0 , the Newton-Raphson method solves this
40 6. Solution of the Power Flow Problem

f (x)

f ( x0)

f ( x1)

x
x2 x1 x0

Figure 6.1. Newton-Raphson method in unidimensional case

vector equation by generating the following sequence:

J(xν )∆xν = −f (xν )

(6.13)
ν+1 ν ν
x = x + ∆x

where J(xν ) = ∂f (x)/∂x is the Jacobian matrix with elements

∂fi
Jij = (6.14)
∂xj

6.2.1 Unidimensional case

To get a better feeling for the method we first study the one-dimensional
case, and eq. (6.12) becomes
f (x) = 0 (6.15)

where x is the unknown and f (x) is a scalar function. Figure 6.1 illustrates
a simple case in which there is a single solution to eq. (6.15). Under these
circumstances, the following algorithm can be used to find the solution of
eq. (6.15):

1. Set ν = 0 and choose an appropriate starting value x0 ;

2. Compute f (xν );
6.2. Newton-Raphson Method 41

f (x)

f ( x0)

f ( x1)

x2 x1 x0

Figure 6.2. Dishonest Newton-Raphson method in unidimensional case

3. Compare f (xν ) with specified tolerance ε;

if |f (xν )| ≤ ε, then x = xν is the solution to eq. (6.15);
Otherwise, |f (xν )| > ε, go to the next step;

4. Linearize f (x) at the current solution point [xν , f (xν )], as shown in
Figure 6.1. That is, f (xν + ∆xν ) ≈ f (xν ) + f ′ (xν )∆xν , where f ′ (xν )
is calculated at xν

5. Solve f (xν ) + f ′ (xν )∆xν = 0 for ∆xν , and update the solution esti-
mate, xν+1 = xν + ∆xν , where ∆xν = −f (xν )/f ′ (xν );

6. Update iteration counter ν + 1 → ν and go to step 2.

The dishonest Newton-Raphson method is illustrated in Figure 6.2. In

this case at Step 4 of the algorithm, a constant derivative is assigned and
f ′ (xν ) = f ′ (x0 ). Although the number of iterations required for convergence
usually increases, it is not necessary to recalculate the derivatives for each
iteration and hence the computation burden at each iteration is lower. When
only limited accuracy is needed, the overall performance of the dishonest
version may be better than that of the full Newton-Raphson method.

6.2.2 Quadratic Convergence

Close to the solution point x∗ , the Newton-Raphson method normally presents
a property called quadratic convergence. This can be proved for the unidi-
mensional case discussed above if it is assumed that x∗ is a simple (not a
multiple) root and that its first and second derivatives are continuous.
42 6. Solution of the Power Flow Problem

Hence, f ′ (x∗ ) = 0, and for any x in a certain neighbourhood of x∗ ,

f ′ (x) 6= 0. If εν denotes the error at the ν-th iteration, i.e.

εν = x∗ − x(ν) (6.16)

the Taylor expansion about xν yields

f (x∗ ) = f (x(ν) + εν )
= f (x(ν) ) + f ′ (x(ν) )εν + 1/2f ′′ (x̄)ε2ν (6.17)
=0

where x̄ ∈ [x(ν) , x∗ ]. Dividing by f ′ (x(ν) ), this expression can be written as

f (x(ν) ) f ′′ (x̄) 2
+ ε ν + 1/2 ε =0 (6.18)
f ′ (x(ν) ) f ′ (x(ν) ) ν
Since,

f (x(ν) ) f (x(ν) )
+ ε ν = + x∗ − x(ν) = x∗ − x(ν+1) = εν+1 (6.19)
f ′ (x(ν) ) f ′ (x(ν) )
the following relationship between εν and εν+1 results:
εν+1 1 f ′′ (x̄)
= − (6.20)
ε2ν 2 f ′ (xν )
In the vicinity of the root, i.e. as xν → x∗ , x̄ → x∗ , and we thus have
1 |f ′′ (x∗ )| 2
|εν+1 | = εν (6.21)
2 |f ′ (x∗ )|
From eq. (6.21) it is clear that the convergence is quadratic with the as-
sumptions stated above.

6.2.3 Multidimensional Case

Reconsider now the n-dimensional case

f (x) = 0 (6.22)

where
f (x) = (f1 (x), f2 (x), . . . , fn (x))T (6.23)
and
x = (x1 , x2 , . . . , xn )T (6.24)
Thus f (x) and x are n-dimensional (column) vectors.
The Newton-Raphson method applied to to solve eq. (6.22) follows ba-
sically the same steps as those applied to the unidimensional case above,
6.2. Newton-Raphson Method 43

except that in Step 4, the Jacobian matrix J(xν ) is used, and the lineariza-
tion of f (x) at xν is given by the Taylor expansion

f (xν + ∆xν ) ≈ f (xν ) + J(xν )∆xν (6.25)

where the Jacobian matrix has the general form

∂f1 ∂f1 ∂f1

 
 ∂x1 ···
 ∂x2 ∂xn 

 
 ∂f2 ∂f2 ∂f2 
 
∂f ···
J= = ∂x ∂x2 ∂xn 
 (6.26)
∂x  .1
 .. .. .. .. 
 . . . 

 ∂fn ∂fn ∂fn 
···
∂x1 ∂x2 ∂xn

The correction vector ∆x is the solution to

f (xν ) + J(xν )∆xν = 0 (6.27)

Note that this is the linearized version of the original problem f (xν +∆xν ) =
0. The solution of eq. (6.27) involves thus the solution of a system of linear
equations, which usually is done by Gauss elimination (LU factorization).
The Newton-Raphson algorithm for the n-dimensional case is thus as
follows:

1. Set ν = 0 and choose an appropriate starting value x0 ;

2. Compute f (xν );

3. Test convergence:
If |fi (xν )| ≤ ε for i = 1, 2, . . . , n, then xν is the solution
Otherwise go to 4;

4. Compute the Jacobian matrix J(xν );

5. Update the solution

∆xν = −J−1 (xν )f (xν )

(6.28)
xν+1 = xν + ∆xν

6. Update iteration counter ν + 1 → ν and go to step 2.

44 6. Solution of the Power Flow Problem

6.3 Newton-Raphson applied to the Power Flow Equa-

tions
In this section we will now formulate the Newton-Raphson iteration of the
power flow equations. Firstly, the state vector of unknown voltage angles
and magnitudes is ordered such that
 
θ
x= (6.29)
U

and the nonlinear function f is ordered so that the first components corre-
spond to active power and the last ones to reactive power:

P(x) − P(s)
   
∆P(x)
f (x) = = (6.30)
∆Q(x) Q(x) − Q(s)

with
 (s) 
P2 (x) − P2
 .. 

 . 

(s)
 Pm (x) − Pm
 

f (x) =  − − − − −− (6.31)
 

 Q2 (x) − Q(s)
 
2

..
 
 
 . 
(s)
Qn (x) − Qn

In eq. (6.31) the functions Pk (x) are the active power flows out from bus k
(s)
given by eq. (4.11) and the Pk are the known active power injections into
bus k from generators and loads, and the functions Qk (x) are the reactive
(s)
power flows out from bus k given by eq. (4.12) and Qk are the known
reactive power injections into bus k from generators and loads. The first
m − 1 equations are formulated for PU and PQ buses, and the last n − 1
equations can only be formulated for PQ buses. If there are NP U PU buses
and NP Q PQ buses, m − 1 = NP U + NP Q and n − 1 = NP Q . The load flow
equations can now be written as
 
∆P(x)
f (x) = =0 (6.32)
∆Q(x)

and the functions ∆P(x) and ∆Q(x) are called active and reactive (power)
mismatches. The updates to the solutions are determined from the equation

∆θν ∆P(xν )
   
J(xν ) + =0 (6.33)
∆Uν ∆Q(xν )
6.4. P θ − QU Decoupling 45

The Jacobian matrix J can be written as

∂∆P ∂∆P
 
 ∂θ ∂U 
J= (6.34)
 

 ∂∆Q ∂∆Q 
∂θ ∂U
which is equal to
∂P(x) ∂P(x)
 
 ∂θ ∂U 
J= (6.35)
 

 ∂Q(x) ∂Q(x) 
∂θ ∂U
or simply
∂P ∂P
 
 ∂θ ∂U 
J= (6.36)
 

 ∂Q ∂Q 
∂θ ∂U
In eq. (6.34) the matrices ∂P/∂θ and ∂Q/∂U are always quadratic, and so
is of course J.

6.4 P θ − QU Decoupling
The ac power flow problem above involves four variables associated with
each network node k:

• Uk , the voltage magnitude

• θk , the voltage angle

• Pk , the net active power (generation – load)

• Qk , the net reactive power (generation – load)

For transmission systems, a strong coupling is normally observed between

P and θ, as well as between Q and U . This property will in this section be
employed to simplify and speed up the computations. In the next section
we will derive a linear approximation called dc power flow (or dc load flow).
This linear model relates the active power P to the bus voltage angle θ.
Let us consider a π-model of a transmission line, where the series resis-
tance and the shunt admittance both are neglected and put to zero. In this
case, the active and reactive power flows are given by the following simplified
expressions of eqs. (3.3) and (3.4)
Uk Um sin θkm
Pkm = (6.37)
xkm
46 6. Solution of the Power Flow Problem

Uk2 − Uk Um cos θkm

Qkm = (6.38)
xkm
where xkm is the series reactance of the line.
The sensitivities between power flows Pkm and Qkm and the state vari-
ables U and θ are for this approximation given by
∂Pkm Uk Um cos θkm ∂Pkm Um sin θkm
= = (6.39)
∂θk xkm ∂Uk xkm
∂Qkm Uk Um sin θkm ∂Qkm 2Uk − Um cos θkm
= = (6.40)
∂θk xkm ∂Uk xkm
When θkm = 0, perfect decoupling conditions are observed, i.e.
∂Pkm U k Um ∂Pkm
= =0 (6.41)
∂θk xkm ∂Uk
∂Qkm ∂Qkm 2Uk − Um
=0 = (6.42)
∂θk ∂Uk xkm
As illustrated in Figure 6.3, in the usual range of operations (relatively small
voltage angles), a strong coupling between active power and voltage angle as
well as between reactive power and voltage magnitudes exists, while a much
weaker coupling between reactive power and voltage angle, and between
voltage magnitude and active power exists. Notice, however, that for larger
angles this is no longer true. In the neighbourhood of θkm = 90◦ , there is
strong coupling between P and U as well as between Q and θ.
Example 6.1. A 750 kV transmission line has 0.0175 p.u. series reactance
(the series resistance and the shunt admittance are ignored in this example).
The terminal bus voltage magnitudes are 0.984 and 0.962 p.u. and the angle
difference is 10◦ . Calculate the sensitivities of the active and reactive power
flows with respect to voltage magnitude and phase angle.

Solution The four sensitivities are calculated by using eqs. (6.39) and
(6.40):
∂Pkm Uk Um cos θkm 0.984 · 0.962 cos 10◦
= = = 54.1
∂θk xkm 0.0175

∂Pkm Um sin θkm 0.962 sin 10◦

= = = 9.5
∂Uk xkm 0.0175

∂Qkm Uk Um sin θkm 0.984 · 0.962 sin 10◦

= = = 9.4
∂θk xkm 0.0175

∂Qkm 2Uk − Um cos θkm 2 · 0.984 − 0.962 cos 10◦

= = = 58.3
∂Uk xkm 0.0175
6.4. P θ − QU Decoupling 47

P,Q

VkVm / xkm

Operation range Pkm

qkm
-p/2 0 p /2 p

Qkm

Figure 6.3. P − θ and Q − θ curves for a line with a series resistance

and a shunt admittance of zero and considering terminal voltages Uk =
Um = 1.0 p.u.

As seen the P − θ and Q − U couplings are much greater than the other
couplings. 

If the couplings Q−θ and P −U are neglected the Newton-Raphson iter-

ation scheme can be simplified. With this assumption the Jacobian Matrix
can be written as
∂P
 
 ∂θ 0 
JDEC =  (6.43)
 

 ∂Q 
0
∂U
Thus there is no coupling between the updates of voltage magnitudes and
angles and eq. (6.33) can be written as two uncoupled equations:

∂P ν
∆θ + ∆P(θν , Uν ) = 0 (6.44)
∂θ

∂Q
∆Uν + ∆Q(θν+1 , Uν ) = 0 (6.45)
∂U
48 6. Solution of the Power Flow Problem

In this formulation two systems of linear equations have to be solved instead

of one system. Of course, the total number of equations to be solved is
the same, but since the needed number of operations to solve a system of
linear equations increases more than linearly with the dimension, it takes less
operations to solve eqs. (6.44) and (6.45) than eq. (6.33) with the complete
Jacobian matrix, J. It should be noted that if the iterations of eqs. (6.44) and
(6.45) converge, it converges to a correct solution of the load flow equations.
No approximations have been introduced in the functions P(x) or Q(x),
only in the way we calculate the updates. The convergence of the decoupled
scheme is somewhat slower than the full scheme, but often the faster solution
time for the updates compensates for slower convergence, giving as faster
overall solution time. For not too heavily loaded systems a faster overall
solution time is almost always obtained. The two equations (6.44) and (6.45)
are solved sequentially, and then the updated unknowns of the first equation,
eq. (6.44), can be used to calculate the mismatches of the second system of
equations, eq. (6.45), resulting in an increased speed of convergence.
A number of approximations can be made to calculate the matrix ele-
ments of the the two sub-matrices of the Jacobian in eqs. (6.44) and (6.45).
This will only influence the speed of convergence of the solution. If the
method converges to a solution, this is the correct solution as long as the
accurate expression for ∆P and ∆Q are used.
If approximations regarding the active and reactive power mismatches
are used, the solution can be even faster, but then only an approximative
solution will be obtained. This is further elaborated in the next section.

6.5 Approximative Solutions of the Power Flow Prob-

lem
In the previous section the exact expressions of the power flow equations
were used. However, since the power flow equations are solved frequently
in the operation and planning of electric power systems there is a need
that the equations can be solved fast, and for this purpose the approxima-
tions introduced in this chapter have proved to be of great value. Often
the approximations described here are used together with exact methods.
Approximative methods could be used to identify the most critical cases,
which are then further analysed with the full models. The fast, approxima-
tive methods can also used to provide good initial guesses for a complete
solution of the equations.

6.5.1 Linearization
In this subsection the linearized dc power flow equations will be derived.
6.5. Approximative Solutions of the Power Flow Problem 49

Transmission Line
Consider again expressions for the active power flows (Pkm and Pmk ) in a
transmission line:

Pkm = Uk2 gkm − Uk Um gkm cos θkm − Uk Um bkm sin θkm (6.46)
2
Pmk = Um gkm − Uk Um gkm cos θkm + Uk Um bkm sin θkm (6.47)

These equations can be used to determine the real power losses in a

transmission line

Pkm + Pmk = gkm (Uk2 + Um

2
− 2Uk Um cos θkm ) (6.48)

If the terms corresponding to the active power losses are ignored in eqs.
(6.46) and (6.47), the result is

Pkm = −Pmk = −Uk Um bkm sin θkm (6.49)

The following additional approximations are often valid, particularly during

light load conditions
Uk ≈ Um ≈ 1 p.u. (6.50)
sin θkm ≈ θkm (6.51)
And since
bkm = −1/xkm (6.52)
we can simplify the expression for the active power flow Pkm to
θk − θm
Pkm = θkm /xkm = (6.53)
xkm
This equation is analogous to Ohm’s law applied to a resistor carrying a
dc current:

• Pkm is the dc current;

• θk and θm are the dc voltages at the resistor terminals;

• xkm is the resistance.

This is illustrated in Figure 6.4.

Series Capacitor
For a given voltage angle spread, the active power flow in a transmission line
decreases with the line reactance (and series reactance normally increases
with line length). Series compensation aims at reducing the effective electric
length of the line: a series capacitor connected in series with the line. If, for
example, a 40% compensation corresponds to a capacitor with a reactance
50 6. Solution of the Power Flow Problem

k m
rkm + jxkm

sh sh (a)
jbkm jbkm

k xkm m

(b)

Pkm = (qk - q m ) x
km

Figure 6.4. Transmission line. (a) Equivalent π-model. (b) DC power

flow model.

of 40% of the original line reactance, but with opposite sign, the resulting
reactance of the compensated line becomes 60% of the original value. Thus

xcomp
km = xkm − xsc (6.54)

with obvious notation. In the dc power flow model the series capacitor
can thus be regarded as a negative resistance inserted in series with the
equivalent line resistance.

In-Phase Transformer
The active power flows, Pkm and Pmk , in an in-phase transformer are given
by eq. (3.10)

Pkm = (akm Uk )2 gkm − akm Uk Um gkm cos θkm − akm Uk Um bkm sin θkm (6.55)

Neglecting the terms associated with losses and introducing the same ap-
proximations used for transmission lines yields
θkm
Pkm = (6.56)
xkm /akm
where further approximating akm ≈ 1, i.e. the transformer tap ratio is close
to the relation between the nominal voltages of the two sides, yields the
6.5. Approximative Solutions of the Power Flow Problem 51

k 1:akm m
jxkm
(a)

k m
xkm akm
(b)

Pkm = (qk - q m ) (xkm akm)

Figure 6.5. In-phase transformer. (a) Transformer comprising ideal

transformer and series reactance. (b) DC power flow model.

same expression as for transmission lines

θkm
Pkm = (6.57)
xkm

This is illustrated in Figure 6.5.

Phase Shifter
Let us consider again the expression for the active power flow Pkm in a phase-
shifting transformer of the type represented in Figure 2.6 with akm = 1
(eq. (3.14)):

Pkm = Uk2 gkm − Uk Um gkm cos(θkm + ϕkm ) − Uk Um bkm sin(θkm + ϕkm )

As with transmission lines and in-phase transformers, if the terms as-

sociated with active power losses are ignored and Uk = Um = 1 p.u. and
bkm = −x−1
km , the result is

sin(θkm + ϕkm )
Pkm = (6.58)
xkm

and if (θkm + ϕkm ) ≪ π/2, then linear approximation can be used, giving

(θkm + ϕkm )
Pkm = (6.59)
xkm
52 6. Solution of the Power Flow Problem

k m
qk +jkm jxkm
(a)
jjkm
1:e

k xkm m

(b)

Pkm = (qk + jkm - qm ) /xkm

xkm m
k

(c)

jkm / xkm

Figure 6.6. Phase-shifting transformer. (a) Phase-shifting trans-

former model (b) Thévénin dc power flow model. (c) Norton dc power
flow model.

Note that Pkm has two components, the first depending on the terminal
bus voltage angles, θkm /xkm , and the other depending only on the phase-
shifting transformer angle, ϕkm /xkm . If ϕkm is considered to be a constant,
eq. (6.59) can be represented by the linearized model shown in Figure 6.6,
where the constant part of the active power flow, ϕkm /xkm , appears as an
extra load on the terminal bus k and an extra generation on the terminal
bus m if ϕkm > 0, or vice-versa if ϕkm < 0.

6.5.2 Matrix Formulation of DC Power Flow Equations

In this section, the dc model developed above is expressed in the form I =
YE. According to the dc model, the active power flow in a branch is given
by
Pkm = x−1
km θkm (6.60)

where xkm is the series reactance of the branch (parallel equivalent of all the
circuits existing in the branch).
6.5. Approximative Solutions of the Power Flow Problem 53

The active power injection at bus k is thus given by

X X X
Pk = x−1
km θkm = ( x−1
km ) θk + (−x−1
km θm ) (6.61)
m∈Ωk m∈Ωk m∈Ωk

for k = 1, 2, . . . , N , where N is the number of buses in the network. This

can be put into matrix form as follows:

P = B′ θ (6.62)

where

• P is the vector of the net injections Pk

• B′ is the nodal admittance matrix with the following elements:

′
Bkm = −x−1
km
X
′
Bkk = x−1
km
m∈Ωk

• θ is the vector of voltage angles θk

The matrix B′ in eq. (6.62) is singular, i.e. its determinant is equal

to zero. This means that the system of equations in eq. (6.62) has no
unique solution and that the rows of B′ are linearly dependent. To make
the system solvable, one of the equations in the system is removed, and the
bus associated with that row is chosen as the angle reference, i.e. θref = 0.
Since the network losses are ignored in this approximation a slack node is
not required to compensate for these a priori unknown losses, but of course
still an angle reference is needed.
In forming the matrix B′ , in-phase and phase-shifting transformers are
treated like transmission lines. The phase-shifting transformers also con-
tribute to the construction of the independent vector P with the Norton
equivalent injections shown in Figure 6.6(c).

Example 6.2. Consider the network given in Figure 6.7 in which the ref-
erence angle is θ1 = 0. Use the dc power flow method to calculate the power
flows in the lines.

Solution In this case, the elements of the matrix B′ are calculated as

B22 = x−1 −1
21 + x23 = (1/3)
−1
+ (1/2)−1 = 5
B23 = −x−1
23 = −(1/2)
−1
= −2
B32 = −x−1
32 = −(1/2)
−1
= −2
B33 = x−1 −1
31 + x32 = (1/2)
−1
+ (1/2)−1 = 4
54 6. Solution of the Power Flow Problem

P1 = 1.5 P2 = -0.5

P12 2
1

x12 = 1/3
P23
P13
x13 = 1/2 x23 = 1/2

3
P3 = -1.0

Figure 6.7. 3-bus network. (Active power in p.u.; branch reactances in p.u.)

and thus  
′ 5 −2
B =
−2 4
and  
′ −1 1/4 1/8
(B ) =
1/8 5/16
The nodal voltage angles (in radians) can now easily be calculated
 
θ2
θ= = (B′ )−1 P
θ3
    
1/4 1/8 −0.5 −0.250
= =
1/8 5/16 −1.0 −0.375

The power flows in the transmission lines are according to the dc power
flow model

P12 = x−1
12 θ12 = 3 · 0.25 = 0.75 p.u.
P13 = x−1
13 θ13 = 2 · 0.375 = 0.75 p.u.
P23 = x−1
23 θ23 = 2 · 0.125 = 0.25 p.u.


6.5. Approximative Solutions of the Power Flow Problem 55

6.5.3 DC Power Flow Model

The linearized model P = B′ θ can be interpreted as the model for a net-
work of resistors fed by dc current sources where P is the vector of nodal
current injections, θ is the nodal vector of dc voltages, and B′ is the nodal
conductance matrix, as illustrated in Figure 6.8.
56 6. Solution of the Power Flow Problem

1 2 3
P1 x12 P2
P3
x23

x15 x34 (a)

x25

P5 P4

x45
5 4

I2

I1 I3

r12 r23

r15 r34
r25 (b)

I4

r45

Figure 6.8. 6-bus network. (a) power network. (b) dc power flow model.
7
Fault Analysis

This chapter presents computation techniques for the currents that may oc-
cur in the network under symmetrical short circuit conditions. These short
circuit currents determine the rating of circuit breakers that must be able to
clear the fault in order to avoid permanent damage to the equipment.

S O FAR we have dealt with steady state behavior of power systems under
normal operating conditions. This chapter is devoted to abnormal sys-
tem behavior under conditions of faults. Such conditions are caused in the
system accidentally through insulation failure of equipment or flashover of
lines initiated by a lightning stroke or through accidental faulty operation.
In high voltage networks, short circuits are the most frequent type of
faults. Short circuits may be solid or may involve an arc impedance. Fig-
ure 7.1 illustrates different types of short circuits.
Depending on the location, the type, the duration, and the system
grounding short circuits may lead to
• electromagnetic interference with conductors in the vicinity (distur-
bance of communication lines),

• stability problems,

• mechanical and thermal stress (i.e. damage of equipment, personal

danger)

• danger for personnel

The system must be protected against flow of heavy short circuit currents
by disconnecting the faulty part of the system by means of circuit breakers
operated by protective relaying. The safe disconnection can only be guar-
anteed if the current does not exceed the capability of the circuit breaker.
Therefore, the short circuit currents in the network must be computed and
compared with the ratings of the circuit breakers at regular intervals as part
of the normal operation planning.
As illustrated in Figure 7.2, the short circuit currents at network nodes
are generally increasing over the years due to
• more generators,

57
58 7. Fault Analysis

a) Symmetrical three-phase short circuit b) Two-phase without ground contact

a a
b b
c c

c) Two-phase with ground contact d) Single-phase earth fault

a a
b b
c c

partial SC current in
SC current
conductor or ground

Figure 7.1. Examples for different types of short circuits.

• new lines in existing networks,

• interconnection of isolated networks to an integrated one.

This is primarily a problem for the expansion planning, where the impacts
of long-term changes on the short circuit currents have to be assessed. If
the short circuit current exceeds the admissible limit at a network node, the
circuit breakers have to be replaced by breakers with higher ratings. Alter-
natively, the impedance between feeder and fault location can be increased
in order to reduce the short circuit current. This is commonly achieved by

• introducing a higher voltage level while splitting the existing lower

voltage network (Figure 7.3),

• use of multiple busbars (Figure 7.4),

• fast decoupling of busbars during faults (Figure 7.5).

Since changing the circuit breakers involves very high costs, the pro-
posed means to reduce the short circuit currents are the generally preferred
solution. However, this results in a more complex network structure. It
 59

Figure 7.2. Development of short circuit currents over the years.

Figure 7.3. Introduction of a higher voltage level.

also leads to more possibilities to reconfigure the network topology during

operation. When a multiple busbar is introduced, for example, a line can
be switched from one busbar to another.
Switching actions have a significant influence on the short circuit cur-
60 7. Fault Analysis

Figure 7.4. Multiple busbar operation.

ts = 70 . . . 80ms

Figure 7.5. Fast busbar decoupling.

rents, but not all possible topologies can be studied during network planning.
Therefore, calculating the short circuit currents has become more and more
a problem for the network operation planning. Prior to each switching action
all short circuit currents of the new topology must be calculated in order to
decide if the switching action may be carried out. This requires computation
algorithms that are sufficiently fast for real time applications.
The majority of system faults are not three-phase faults but faults in-
volving one line to ground or occasionally two lines to ground. These are
unsymmetrical faults requiring special computational methods like symmet-
rical components. Though symmetrical faults are rare, symmetrical short
circuit analysis must be carried out, as this type of fault generally leads
to the most severe fault current flow against which the system must be
protected. Symmetrical fault analysis is, of course, simpler to carry out.
A power network comprises synchronous generators, transformers, lines,
and loads. Though the operating conditions at the time of fault are impor-
tant, the loads can usually be neglected during short circuits, as voltages
dip very low so that currents drawn by loads can be neglected in comparison
with short circuit currents.
The synchronous generator during short circuit has a characteristic time-
varying behavior. In the event of a short circuit, the flux per pole undergoes
dynamic change with associated transients in damper and field windings.
The reactance of the circuit model of the machine changes in the first few
7.1. Transients on a transmission line 61

cycles from a low subtransient reactance to a higher transient value, finally

settling at a still higher synchronous (steady state) value. Depending upon
the arc interruption time of the circuit breakers, an appropriate reactance
value is used for the circuit model of synchronous generators for the short
circuit analysis.
In a modern large interconnected power system, heavy currents flowing
during a short circuit must be interrupted much before the steady state con-
ditions are established. Furthermore, from the considerations of mechanical
forces that act on the circuit breaker components, the maximum current that
a breaker has to carry momentarily must also be determined. For selecting
a circuit breaker we must, therefore, determine the initial current that flows
on occurrence of a short circuit and also the current in the transient that
flows at the time of circuit interruption.
We distinguish between two different approaches to calculate the short
circuits in a power system:

• Calculation of transient currents

• Calculation of stationary currents

First, we will focus on the calculation of transient currents since this

will help us to understand the physical phenomena during short circuits.
However, for large power systems, the computation of transient currents is
not feasible. For this reason simplified techniques for short circuit current
computation will be presented that are based on stationary models.

7.1 Transients on a transmission line

Let us consider the short circuit transient on a transmission line. Certain
simplifying assumptions are made at this stage:

1. The line is fed from a constant voltage source.

2. Short circuit takes place when the line is unloaded.

3. Line capacitance is negligible and the line can be represented by a

lumped RL series circuit.

R L i

√
v= 2U sin(ωt + α)

Figure 7.6. Transmission line model.

62 7. Fault Analysis

With the above assumptions the line can be represented by the circuit
model of Figure 7.6. The short circuit is assumed to take place at t = 0.
The parameter α controls the instant on the voltage wave when short circuit
occurs. It is known from circuit theory that the current after short circuit
is composed of two parts, i.e.

i = is + it (7.1)

where is represents the steady state alternating current

√
2U
is = sin(ωt + α − θ) (7.2)
|Z|
and it represents the transient direct current
√
−(R/L)t 2U
it = −is (0) e = sin(θ − α) e−(R/L)t (7.3)
|Z|
with  
−1 ωL
p
2 2 2
Z = R + ω L ∠ θ = tan . (7.4)
R
A plot of i = is + it is shown in Figure 7.7. In power system terminology,
the sinusoidal steady state current is called the symmetrical short circuit
current and the unidirectional transient component is called the DC off-set
current, which causes the total short circuit current to be unsymmetrical till
the transient decays.

imm i = is + it

Figure 7.7. Waveform of a short circuit current on a transmission line.

It follows easily from Figure 7.7 that the maximum momentary short
circuit current imm corresponds to the first peak. If the decay of transient
current in this short time is neglected,
√ √
2U 2U
imm = sin(θ − α) + (7.5)
|Z| |Z|
7.2. Short circuit of a synchronous machine 63

Since transmission line resistance is small, θ ≈ 90◦ .

√ √
2U 2U
imm = cos α + (7.6)
|Z| |Z|
This has the maximum possible value for α = 0, i.e. short circuit occurring
when the voltage wave is going through zero. Thus imm may be a high as
twice the maximum of the symmetrical short circuit current:
√
2U
imm ≤ 2 (7.7)
|Z|
For the selection of circuit breakers, momentary short circuit current is taken
corresponding to its maximum possible value.
The next question is ‘what is the current to be interrupted’ ? As has
been pointed out earlier, modern circuit breakers are designed to interrupt
the current in the first few cycles (five cycles or less). With reference to
Figure 7.7 it means that when the current is interrupted, the DC off-set it
has not yet died out and contributes thus to the current to be interrupted.
Rather than computing the value of the DC off-set at the time of interruption
(this would be highly complex in a network of even moderately large size),
the symmetrical short circuit current alone is calculated. This current is
then increased by an empirical multiplying factor to account for the DC
off-set current.

7.2 Short circuit of a synchronous machine

Under steady state short circuit conditions, the armature reaction of a syn-
chronous generator produces a demagnetizing flux. In terms of a circuit
this effect is modelled as a reactance Xa in series with the induced electro-
magnetic field. This reactance when combined with the leakage reactance
Xl of the machine is called synchronous reactance Xd . The index d denotes
the direct axis. Since the armature reactance is small, it can be neglected.
The steady state short circuit model of a synchronous machine is shown in
Figure 7.8.
Xl Xa

Xd
Eg Synchronous
reactance

Figure 7.8. Steady state short circuit model of a synchronous machine.

Consider now the sudden short circuit of a synchronous generator that

has initially been operating under open circuit conditions. The machine
64 7. Fault Analysis

undergoes a transient in all the three phases finally ending up in the steady
state condition discribed above. The circuit breaker must interrupt the
current long before the steady state condition is reached. Immediately upon
short circuit, the DC off-set currents appear in all three phases, each with a
different magnitude since the point on the voltage wave at which short circuit
occurs is different for each phase. These DC off-set currents are accounted
for separately on an empirical basis. Therefore, for short circuit studies, we
need to concentrate our attention on the symmetrical short circuit current
only.
In the event of a short circuit, the symmetrical short circuit current is
limited initially only by the leakage reactance of the machine. Since the air
gap flux cannot change instantaneously, to counter the demagnetization of
the armature short circuit current, currents appear in the field winding as
well as in the damper winding in a direction to help the main flux. These
currents decay in accordance with the winding time constants. The time
constant of the damper winding which has low X/R-ratio is much less than
the one of the field winding, which has high leakage inductance with low
resistance. Thus, during the initial part of the short circuit, the damper
and field windings have transformer currents induced in them. In the circuit
model their reactances—Xf of field winding and Xdw of damper winding—
appear in parallel with Xa as shown in Figure 7.9.
Xdw

Xl Xf

Xa
Eg
Subtransient
reactance

Figure 7.9. Approximate circuit model during subtransient period of

short circuit.

Xf
Xl
Xa

Eg Transient
reactance

Figure 7.10. Approximate circuit model during transient period of short circuit.

As the damper winding currents are first to die out, Xdw effectively
becomes open circuited and at a later stage Xf becomes open circuited.
7.2. Short circuit of a synchronous machine 65

The machine reactance thus changes from the parallel combination of Xa ,

Xf , and Xdw during the initial period of the short circuit to Xa and Xf
in parallel (Figure 7.10) during the middle period. The machine reactance
finally becomes Xa in steady state (Figure 7.8). The reactance presented by
the machine in the initial period of the short circuit, i.e.
1
Xd′′ = Xl + (7.8)
1/Xa + 1/Xf + 1/Xdw
is called the subtransient reactance of the machine; while the reactance ef-
fective after the damper winding currents have died out, i.e.
1
Xd′ = Xl + (7.9)
1/Xa + 1/Xf
is called the transient reactance. Of course, the reactance under steady state
conditions is the synchronous reactance. Obviously Xd′′ < Xd′ < Xd . The
machine thus offers a time-varying reactance which changes from Xd′′ to Xd′
and finally to Xd .

i
√ ′′ Subtransient period
2Isc
√ ′
2Isc Transient period Steady state

√
2Isc

0
t

Figure 7.11. Symmetrical short circuit armature current in syn-

chronous machine.

If we examine the oscillogram of the short circuit current of a syn-

chronous machine after the DC off-set has been removed, we will find the
current wave shape as given in Figure 7.11. The short circuit current can be
divided into three periods—the initial subtransient period when the current
is large as the machine offers subtransient reactance, the middle transient
period where the machine offers transient reactance, and finally the steady
state period when the machine offers synchronous reactance. Hence we dis-
tinguish between
66 7. Fault Analysis

• the steady state current Isc ,

′ , and
• the transient current Isc
′′ .
• the initial subtransient current Isc

None of the three currents includes a DC component. The steady state and
transient currents are obtained by extrapolating their envelopes backwards
in time. Since we are normally interested in maximum short circuit current,
most short circuit computations are based only on the subtransient current.
In some systems a weighted average of the subtransient and the transient
reactance is used.

7.3 Algorithms for short circuit studies

7.3.1 Generator model
In the power flow analysis we are modeling generators as constant power
sources (PU characteristics) which leads to nonlinear power flow equations.
However, for the short transient period during a short circuit this model
is no longer valid since the power and the voltage regulators operate with
much larger time constants. Here, the linear model outlined in Figure 7.12 is
more appropriate. It is used for generators and infeeds, where E ′′ represents
the subtransient electro-magnetic field and Xd′′ represents the subtransient
reactance of generators or the internal grid impedance of a feeder. This
model may also be applied to large motor loads.

R Xd′′

E ′′

Figure 7.12. Linear generator model for short circuit computation.

7.3.2 Simplifications
When computing short circuits in a power system further simplifications
can be made. The following simplifications are also proposed in the German
standard VDE 0102:

• All line capacitances are ignored.

• All non-motor shunt impedances are ignored; motor loads are treated
the same way as generators.
7.3. Algorithms for short circuit studies 67

• The voltage magnitude and phase angle of generators and infeeds are
all set to the same value

Ei′′ = c · Unom (7.10)

where Unom is the nominal voltage of the system in which the short
circuit occurs. In high voltage systems (≥ 35 kV) c = 1.1 represents
the difference between the effective voltage and the system voltage.
If the minimum rather than the maximum initial AC short circuit
current is to be calculated, c is set to c = 0.95.

• All tap changing transformers are in middle position.

These simplifications are indicated for studies regarding medium- and

long-term network planning. In the planning stage, the calculations are
based on estimated and hence inaccurate data. Therefore, the demands
on the short circuit computation algorithm are lower than for real-time
applications in the network operation, where accurate results are desired.
Studies have shown that the shunt elements and loads have little influence
on the short circuit currents (0.5 % . . . 4 %) and may compensate each other.
However, disregarding the actual generator pole voltages and the actual
positions of tap changing transformers may sometimes lead to errors of up
to 30 %.

7.3.3 Solving the linear system equations

With the linear models of the network elements, a system of linear equations
can be set up for a short circuit at any node i. Figure 7.13 illustrates the
network model. The voltage sources of the generators (Figure 7.12) must be
transformed into equivalent current sources. The admittance of the current
source is considered in the respective element of the admittance matrix.

Generator Load
nodes nodes
′′
E1
I1 = ′′
Xd 1 j Ij = 0

Network
′′
Eh
Ih = ′′
Xd h n In = 0
Short circuit node
i

′′
Isc

Figure 7.13. Network with short circuit at node i.

68 7. Fault Analysis

′′ at node i the equation

In order to calculate the short circuit current Isc
Y·U=I (7.11)
must be solved, where
• Y is the nodal admittance matrix
• U is the nodal voltage vector
(
Uj if j 6= i
U= (7.12)
0 if j = i

• I is the vector of injection currents

Ej


 ′′ for generator nodes j 6= i



 Xdj

 0 for load nodes j 6= i
I= ′′

 −Isc for short circuit node without generation j = i
E


 ′′ j
−Isc + ′′ for short circuit node with generation j = i


Xdj
(7.13)
All elements Uj of vector U are unknown variables with the exception of
voltage Ui = 0 at the short circuit node i. On the contrary, all elements
of vector I are known except for the current Ii at the short circuit node i.
To solve the system of equations (7.11), the ith row and the ith column are
removed. The remaining admittance matrix is then factorized into trian-
gular matrices. Backward substitution yields the unknown voltages. The
unknown current Ii at the short circuit node i can be calculated from the
voltages using the ith row of eq. (7.11).
′′ are calculated for all nodes i = 1, . . . , n
If the short circuit currents Isci
the equation system has to be set up and solved for every short circuit node.
Since always another row and another column are removed the matrix must
be factorized each time anew. Therefore, with the proposed method the
computation of all short circuit currents is considerably expensive.

7.3.4 The superposition technique

If two equal voltage sources in opposite direction are introduced into the
short circuit path at node i according to Figure 7.14, neither currents nor
voltages in the network are affected. Since the network consists only of lin-
ear elements the short circuit calculation is a linear problem. Therefore, the
superposition principle can be applied, that means the computation of the
short circuit current may be performed in two steps. These steps are illus-
trated in Figure 7.15 and Figure 7.16. The superposition, i.e. the addition of
the results finally yields the desired values. The advantage of inserting two
opposite voltage sources becomes clear if both steps are regarded separately.
7.3. Algorithms for short circuit studies 69

Generator Load
nodes nodes
I1 1 j

Network

Ih h n
Short circuit node
i

Uli
Ui = 0
Uli
′′
Isc

Figure 7.14. Network with short circuit at node i.

First step
Principally, the voltage Uli can be chosen arbitrarily. If it is chosen in such
′′(1)
a way that Isc = 0, the first step corresponds to the computation of node
voltages and branch currents in a power flow analysis. The power flow
analysis yields the voltage Uli , i.e. the voltage at the short circuit node i
before the fault occurs.
If the above simplifications are applied, the first step can be omitted. For
all generators, infeeds and loads the voltage is then uniformely presumed

Ei′′ = c · Unom . (7.14)

As already mentioned, disregarding the actual generator voltages may

lead to considerable errors. Therefore, this approximation is not always rec-
ommended. The first step, i.e. calculating the pre-fault voltages, is identical
for all short circuit nodes i = 1, . . . , n and needs to be done only once.
Thus, the additional effort for an exact calculation of the node voltages is
comparatively small.

Second step
In the second step the voltage Uli is inserted in reverse direction at the short
circuit node i. All other sources like generators and infeeds are disregarded.
This yields the equation system

Y · U(2) = I(2) (7.15)

with
70 7. Fault Analysis

Generator Load Generator Load

nodes nodes nodes nodes
I1 1 j 1 j

Network Network

Ih h n h n
Short circuit node Short circuit node
i i

Uli Uli
′′
I=0 Isc

Figure 7.15. Superposition 1st step. Figure 7.16. Superposition 2nd step.

• Elements of vector U(2) :

(
(2)
(2) Uj if j 6= i
U = (7.16)
−Uli if j = i

• Elements of vector I(2) :

(
0 if j 6= i
I(2) = ′′
(7.17)
−Isc if j = i

(2)
Vector U(2) contains the node voltages Uj , all of which are unknown with
(2) (2)
the exception of Ui = −Uli . In vector I(2) the only non-zero current is Ii .
Dividing the system of equations (7.15) by −Isc ′′ yields the modified system

of equations (7.18) where Û contains only unknown and Î(2) only given
(2)

values.
Y · Û(2) = Î(2) (7.18)
with
• Elements of vector Û(2) :

(2)
 Uj
− if j 6= i


′′

 Isc

Û(2) = (7.19)

Uli


if j = i


 ′′
Isc

• Elements of vector Î(2) :

(
(2) 0 if j 6= i
Î = (7.20)
1 if j = i
7.3. Algorithms for short circuit studies 71

The transformation from eq. (7.15) to eq. (7.18) leaves the admittance
matrix Y unchanged. As a consequence, the triangular factorization of the
admittance matrix must be done only once, even if multiple short circuit
cases are investigated. The superposition technique should therefore be
preferred to the direct solution of the linear system of equations (7.11).
However, if we compute the solution of eq. (7.18) with the known tech-
nique, i.e. triangular factorization and forward-backward-substitution, a new
forward-backward-substitution is required for every short circuit case. This
disadvantage can be avoided with the Takahashi method.

7.3.5 The Takahashi method

The principle idea behind the Takahashi method is the solution of eq. (7.18)
by means of inversion. This yields

Z · Î(2) = Û(2) (7.21)

where
Z = Y−1 (7.22)

is the impedance matrix of the system. The unknown initial short circuit
′′ at node i can be found in the ith element of vector Û(2)
current Isc

(2) Uli
Ûi = ′′
(7.23)
Isc
resp.
′′ Uli
Isc = (2)
. (7.24)
Ûi
From eqs. (7.20) and (7.21) follows

(2)
Ûi = zii (7.25)

where zii is an element on the main diagonal of matrix Z. Inserting eq. (7.25)
in eq. (7.16) finally yields:
′′ Ui
Isc = (7.26)
zii
For the calculation of the short circuit current only the main diagonal
element zii of the impedance matrix Z is required. Knowing all main diag-
onal elements zii (i = 1, . . . , n) is therefore sufficient to compute the short
′′ at all nodes i.
circuit currents Isci
Often, also the partial short circuit currents Iji are to be determined.
These are the currents that flow from adjacent nodes j to the short circuit
72 7. Fault Analysis

node i. With the simplification that all generator voltages are identical the
partial short circuit currents are
 
(2) (2)
Iji = Uj − Ui yji
 
(2) (2)
= Uj − Uli yji (7.27)

where yji is the admittance of the branch between j and i. If this simplifi-
cation is not permitted the terms from the first step must be added here.
With eq. (7.21) follows:

′′ ′′


Iji = −zji Isci + zii Isci yji
′′
= Isci (zii − zji ) yji (7.28)

For the calculation of the partial short circuit currents from node j to the
short circuit node i in addition to the main diagonal element zii all elements
zji are required. Because node j is connected directly to node i, the element
yji in the admittance matrix is non-zero. Although the impedance matrix Z
is not sparse like the admittance matrix Y relatively few elements of Z must
be known to calculate the short circuit currents and the partial short circuit
currents for all nodes. Apart from the main diagonal elements zii these are
the elements zji that are non-zero in the corresponding admittance matrix.
In the Takahashi method only those elements of Z are determined that
have non-zero counterparts in the triangularly factorized admittance matrix.
Except for the few fill-ins that emerge during factorization these are the
elements of Z essential for the short circuit computation. The method shall
be demonstrated here:
Eq. (7.22) can be written as

Y·Z=E (7.29)

where E is the identity matrix. Y is factorized into triangular matrices and

eq. (7.29) can be transformed in the following way:

L·D·R·Z=E | · L−1 (7.30)

−1 −1
D·R·Z=L |·D
R · Z = D−1 · L−1 |+Z−R·Z
−1 −1
Z=D ·L +Z−R·Z
−1 −1
Z=D ·L + (E − R) Z (7.31)

The inverse of diagonal matrix D is again a diagonal matrix. Each element

on the main diagonal contains the reciprocal value of the element before the
7.3. Algorithms for short circuit studies 73

inversion:
−1 1
0
  
d11 0 d11
1
 d22  
d22

D−1 = = (7.32)
   
..  .. 
 .   . 
0 dnn 1
0 dnn

The inverse of the lower triangular matrix L is a triangular matrix with the
same structure. Since all elements on the main diagonal of L are 1, the
determinant is also 1. Thus, the main diagonal remains unchanged after
inversion. The values of the remaining elements are of no interest due to
the nesting of the system of equations (7.31).
 −1  
1 0 1 0
· 1  · 1 
−1
L = = (7.33)
   
..  .. 
· · .  · · . 
· · · 1 · · · 1

The product of the diagonal matrix D−1 with the triangular matrix L−1 is
a triangular matrix with modified main diagonal. Therefore, eq. (7.31) has
the following form:
   1 
z11 z12 · · · z1n d11 0
 z21 z22 · · ·  · 1
 z2n   d22


 .. ..   .. 
 . .  = ·
 · . − (7.34)
 .. ..  
 
.. 
 . .   · · · . 
zn1 zn2 · · · znn 1
· · · · dnn
  
z11 z12 · · · z1n

0 r12 ··· r1n
0 0 ···   z 21 z22 · · ·
r2n z2n 
  
  .. .. 
−  ... .. ..
 
· . . 

. .
r(n−1)n   ... .. 
   
0 0 ··· . 
0 0 ··· 0 zn1 zn2 · · · znn

With the system of equations (7.34) all elements of the impedance matrix
Z can be determined recursively. We start with the main diagonal element
znn on the nth row and continue with the row (n − 1). For each row, first
the secondary diagonal elements zji and then the main diagonal element zii
are calculated.
Here we take advantage of the fact that the admittance matrix Y and
also their inverse—the impedance matrix Z—are symmetrical. This is due to
the simplifying assumption that all tap changing transformers are in middle
74 7. Fault Analysis

position.


 not calculated if rij = 0

n
zij  = zji = (7.35)
 X
i<j 
− ril zlj
l=i+1
n
1 X
zii = − ril zlj (7.36)
dii
l=i+1

5 1
✉ ✉
❅
❅
❅
❅
❅
❅
✉ ❅
❅✉ ✉
4 3 2

Figure 7.17. Sample network.

Example 7.1. The whole procedure shall be demonstrated in single steps on

the sample network given in Figure 7.17. With the arbitrarily chosen node
numbering the admittance matrix Y has the following structure.

 
y11 y12 0 0 y15
y21 y22 y23 0 0 
 
 0 y32
Y= y33 y34 y35 
 (7.37)
 0 0 y43 y44 y45 
y51 0 y53 y54 y55

Solution During the triangular factorization of Y a fill-in element c25 is

generated in the upper triangular matrix. The recursive solution of the
7.3. Algorithms for short circuit studies 75

system of equations (7.34) is then

1
z55 = (7.38)
d55
z45 = z54 = −r45 z55 (7.39)
1
z44 = − r45 z54 (7.40)
d44
z35 = −r34 z45 − r35 z55 (7.41)
z34 = −r34 z44 − r35 z54 (7.42)
1
z33 = − r34 z43 − r35 z53 (7.43)
d33
z25 = −r23 z35 − r25 z55 (7.44)
z23 = −r23 z33 − r25 z53 (7.45)
1
z22 = − r23 z32 − r25 z52 (7.46)
d22
z15 = −r12 z25 − r15 z55 (7.47)
z12 = −r12 z22 − r15 z52 (7.48)
1
z11 = − r12 z21 − r15 z51 . (7.49)
d11

Figure 7.18 illustrates this sequence. 

✉✛ ✉✰ ❡ ✏
✶ ✉
✏✏
❡
✏ ✏
✏✏
✛ ✉✰ ❡
✉✏ ✉
✟✯
✟
✟
✟✟ ✉✛ ✉
✉✛
✒
✉✛ ✉
✉ Calculated Element
✻
✉ ❡ Element not required

Figure 7.18. Solution sequence with Takahashi method.

Until now we have assumed that the admittance matrix Y is symmet-

ric. This assumption is valid only if all tap changing transformers are in
middle position. However, the Takahashi method can also be applied to un-
symmetric matrices. In this more general case we get instead of eq. (7.31)
two similar matrix equations. With one of these equations the secondary
diagonal elements above the main diagonal are calculated, with the other
76 7. Fault Analysis

equation the elements below the main diagonal. The elements on the main
diagonal can be calculated with either equation.
Unsymmetric matrices require about double the effort of symmetric ma-
trices. Since neglecting the actual tap positions can lead to errors of 30 %,
considering the actual tap positions is indicated for network operation ap-
plications, that require short circuit computations with high accuracy.

Concluding remarks
The Takahashi method is an algorithm especially designed for the demands
of short circuit computations. Having the same accuracy it needs only a
fraction of the computation time of conventional methods. This advantage
is important in real-time applications.
However, with the Takahashi method only the initial short circuit cur-
rents can be calculated. This includes the partial short circuit currents
Iji that flow from the adjacent nodes to the short circuit node i. It is not
possible to determine the short current shares delivered from individual gen-
erators or infeeds. Thus the breaking current at the generators can not be
determined with the Takahashi method but requires Gauss elimination.
 Part II

Power System Dynamics and

Stability

77
8
Classification and Definitions of Power
System Stability

This chapter contains a brief overview of some of the basic concepts in power
system dynamics. A classification of different dynamic phenomena is given
and the requirements on modelling are discussed. Stability is a fundamental
concept and a definition of stability in power systems is introduced. Insta-
bilities in power systems can be classified according to their physical origins,
and this classification is reviewed. Finally, a list of further reading in the
subject is given.

I N A ELECTRIC POWER SYSTEM a great variety of different dynam-

ics occurs. These dynamic phenomena have different physical origin and
they occur in different time scales. A system is in a dynamic state if the
time derivative of any system quantity is non zero, and to describe a dy-
namic system mathematically differential or difference equations must be
employed. As an example, the electric consumer loads in a power system
vary spontaneously all the time, and consequently one could state that the
power system is never in steady state in a strict mathematical sense. Such
a point of view is however unpractical in most cases, and one must, as when
it comes to most large and complex systems, study and analyse more or less
simplified models of the system. To do relevant and adequate simplifications
is beneficial for the analysis, and contributes also to achieving results that
are easy to understand and easy to interpret. In addition, when deriving
simplified models one is forced to identify the most important processes and
phenomena in the studied systems, which provides insight and deeper under-
standing. Thus, as an example, the assumption that the loads are constant
and that the electric power output from the generators are constant are rel-
evant approximations for many studies of a power system, and the power
system is under these assumptions described by a solution to the power flow
equations. In the following we will refer to this state as the steady state of
the power system.1
1
In order to emphasise that this state is an approximation or idealisation, it is some-
times referred to as quasi steady state. More generally, a quasi steady state is a state that
strictly is a dynamic state, but could with sufficient accuracy be described by algebraic

79
80 8. Classification and Definitions of Power System Stability

8.1 Dynamics in Power Systems

There are a lot of different dynamical phenomena with different character-
istics in a power system. The phenomena could be local, in which case they
only involve a minor part of the system or a single component. But they can
also involve interactions between different parts of the system that might be
geographically far from each other. In many cases these system-wide inter-
actions are initiated by a local disturbance causing e.g. an earth fault with
subsequent change in network topology. In this compendium interactions
and phenomena that involve many power system components, e.g. genera-
tors and loads, or parts of the system are dealt with. These interactions
have in common that they can cause system instabilities that can lead to
black outs in large parts of the system, i.e. to interruptions of power supply
for many consumers.
Dynamics can also be initiated by actions of different controllers or by
switchings of lines or other components by system operators. Such “distur-
bances” should be regarded as normal and should consequently not endanger
the stability of the system.

8.1.1 Classification of Dynamics

Dynamic phenomena in power systems are usually classified as
1. Electro-magnetic transients (100 Hz – MHz)
2. Electro-mechanical swings (rotor swings in synchronous machines) (0.1 –
3 Hz)
3. Non-electric dynamics, e.g. mechanical phenomena and thermodynam-
ics (up to tens of Hz)
Approximate values of typical frequencies are given in brackets.
One single initial event in the power system can give rise to dynamics in
all the three groups above. A lightning stroke in a power line can induce so
high over-voltages that the insulation fails, causing an earth fault. The earth
fault can cause rotor swings in synchronous machines with high amplitudes.
This can trigger protections to disconnect generators, so that an unbalance
between produced and consumed power in the system arises. The frequency
in the system drops and generators participating in the frequency control
compensate this by increasing their power outputs. Thus the initial lightning
stroke has initiated dynamics in all the three groups above.
Another way to classify dynamic phenomena is given in Figure 8.1. This
classification is based on the time scale of the phenomenon and the (math-
ematical) models used in analysis.
equations. One such example is fault current calculations, which often is done by using
algebraic models even if it is a true dynamic phenomenon, see Figure 8.1.
8.1. Dynamics in Power Systems 81

Power Flow
D.
Harmonic Distortion (Algebraic
equations)

C. Fault Currents

Long Term Stability

(Differential and
B. Algebraic Equations) Short Term Stability

Stator Transients

Saturation and
Resonances (Differential
Equations)
Switching Transients
A.
Lightning Transients (Travelling Wave Theory)

Time (s)
10-6 10-3 1 103

Figure 8.1. Dynamic phenomena in a power system. In the figure

approximate time scales are given and types of mathematical models
used. The different groups are called: A. Electro-magnetic transients.
B. Synchronous machine dynamics. C. Quasi steady state phenomena
D. Steady state phenomena.

8.1.2 Modelling
It is of course almost impossible to develop models that can describe all dy-
namics in a power system and still being of practical use. Often one has to
utilise a model that captures correctly the specific dynamic phenomenon or
interaction that is the aim of the particular investigation. Depending on the
purpose of the study the appropriate model of a given power system com-
ponent could vary significantly. It is obvious that if the aim is to study rela-
tively slow power oscillations between generators in the system, completely
different models are required as compared with if one wants to analyse the
influence of lightning impulses in the windings of the synchronous machine.
Even if it were theoretically possible to develop a complete model of all
the dynamics in the power system, it is questionable if such a model would be
particularly useful. Firstly, such a model would require an enormous amount
of parameter data to be uniquely specified. Secondly, the results obtained
from such a model would be very hard to analyse and interpret. Critical
review and understanding of obtained results is a necessary prerequisite
82 8. Classification and Definitions of Power System Stability

for sound engineering. When making simulations and computations, which

all are done with the help of computers nowadays in system sciences, it is
important to have an expectation of what are reasonable outputs. Thus
trivial errors due to wrong input data files, mistakes in modelling, etc. can
be eliminated to a large extent. The human factor is of utmost importance
in computer-based analysis and simulation.
Models can in principle be erroneous in two different ways. Firstly, it
can have the wrong structure. It can be too simple overlooking important
interactions and processes or modelling them incorrectly. This is of course
very serious and might give rise to detrimental consequences. But it is
also very serious if wrong parameter data is used in a model of the correct
structure. This latter shortcoming occurs not seldom in technical systems,
which might look surprising at first sight. Since technical systems are man
made, one should in principle have access to all design parameters defining
the system. But it turns out that many parameters, e.g. the gain in the
controller, could easily be changed after the system has been commissioned
and such changes are not always reported to system analysts. It is obvious
that the consequences could be very serious. In technical systems there are of
course parameters that are “genuinely” unknown, e.g. the ground resistivity
under a power line. In a large system like a power system, thousands of
parameters are needed to define the system completely. It is a very difficult,
but also very important, task to maintain and keep the data bases where
all these parameter values are stored and updated. This is now a special
activity usually referred to as data engineering.
In this compendium models needed for the problems to be analysed are
developed. Due to space limitations more detailed and elaborate derivation
cannot be presented, but the reader is advised to consult other sources, e.g.
books in electrical machines or the books listed at the end of this chapter.

8.2 Power System Stability

8.2.1 Definition of Stability
A dynamic phenomenon in a power system is, as said above, initiated by
a disturbance in the system. Such a disturbance could as an example be
that a line impedance is changed due to an external cause. The behaviour
of the system after this disturbance depends of course on a how “large”
this disturbance is. A small disturbance results usually in small transients
in the system that are quickly damped out, while a larger disturbance will
excite larger oscillations. We all have an intuitive feeling for what is meant
with stability in this context. Stability is associated with that the system
oscillations decay and that the operation of the power systems can continue
without any major impacts for any of the consumers. But, and this is
very important, as the power system is a nonlinear system (this will be
8.2. Power System Stability 83

elaborated on later) system stability depends on the kind and magnitude of

the disturbance.2 This distinguishes nonlinear systems from linear systems
that can be classified as stable or unstable independent of the disturbance,
i.e. stability is a property of the linear system as such. As will be shown
later, stability of a power system is strongly coupled to both the magnitude
and character of the disturbance as well as to the initial operating point.
Over the years many different definitions of power system stability have
been proposed. The most recent one, which will be adopted in these lecture
notes, was the result of a joint IEEE/CIGRE working group activity. In the
report in reference 7 of section 8.3, the following definition is given:
Definition 8.1. Power system stability is the ability of an electric power
system, for a given initial operating condition, to regain a state of operating
equilibrium after being subjected to a physical disturbance, with most system
variables bounded so that practically the entire system remains intact.
The following two comments elaborate on some important aspects of this
definition.
Comment 8.1. It is not necessary that the system regains the same steady
state operating equilibrium as prior to the disturbance. This would be the
case when e.g. the disturbance has caused any power system component (line,
generator, etc.) to trip. Voltages and power flows will not be the same after
the disturbance in such a case. Most disturbances that are considered in
stability analyses incur a change in system topology or structure.
Comment 8.2. It is important that the final steady state operating equi-
librium after the fault is steady state acceptable. Otherwise protections or
control actions could introduce new disturbances that might influence the sta-
bility of the system. Acceptable operating conditions must be clearly defined
for the power system under study.
As mentioned above there are also other definitions of power system sta-
bility in the literature. They are not all identical, but could differ in some
details, but most of them are in line with the definition proposed above.
Power system dynamics can be modelled by systems of differential and al-
gebraic equations, and the mathematics of those are studied in the theory
of dynamic systems. It is therefore desirable that the stability definitions
introduced in a more mathematically stringent way are compatible with the
more practically oriented definition above. In the report where the above
definition is proposed and motivated, these aspects are further discussed.
2
To give an example of what is meant with magnitude in this context one could consider
a three phase to earth fault on a power line that is cleared by disconnecting the line. The
magnitude of this disturbance increases with the fault clearing time. If the fault clearing
time is sufficiently small, the system will remain stable, while a longer fault clearing might
cause system instability. Similarly, a high impedance fault gives a smaller disturbance than
a solid fault.
84 8. Classification and Definitions of Power System Stability

8.2.2 Classification of Power System Stability

To achieve a better overview and structure of stability analyses of power

systems, it is of great help to classify possible power system stability. The
classification to be introduced here is based on the physical mechanism be-
ing the main driving force in the development of the associated instability.
It could be either the active or the reactive power that is the important
quantity.
A common characteristic of the instabilities to be discussed here is that
they have their origin in too large an imbalance of active or reactive power in
the system, locally or globally. This imbalance can then develop in different
ways and cause unstable behaviour depending on system characteristics.
Exactly what is meant with imbalance will be elaborated in the following.

Rotor Angular or Synchronous Stability

The total active electrical power fed into the power system by the generators
is always equal to the active power consumed by the loads including the
losses in the system. On the other hand, there is not always a similar
balance between the loads and the power fed into the generators by the
prime movers, e.g. the hydro and steam turbines. If such an imbalance
develops, the rotating parts of the generators and other rotating machines
will act as energy buffer, and the kinetic energy stored in these will decrease
or increase as a result of the imbalance. Rotor angle stability refers to the
ability of synchronous machines of a power system to remain in synchronism
after a disturbance.
If the disturbance is local and substantial, e.g. an earth fault close to a
generator, the generator can fall out of step since it has been accelerated
during the fault. As quite big currents will flow in the generator windings
in such a case, it must be disconnected to avoid that it is damaged. Typical
time scale for such an instability to develop is a second to a couple of seconds.
This kind of instability is called transient instability and instability appears
usually in form of aperiodic angular separation due to lack of synchronizing
torque. This form of instability is also referred to as large-disturbance rotor
angle instability.
Small-disturbance (or small-signal) rotor angle stability is concerned
with the ability of the power system to maintain synchronism under small
disturbances. These disturbances are considered to be sufficiently small that
linearization of the system equations is permissible for purposes of analysis.
Usually small-disturbance rotor angle stability is associated with insufficient
damping of oscillations.
8.2. Power System Stability 85

Frequency Stability

A third variety of active power imbalance, which is different from the ones
above, is when the imbalance is not local but global. In the preceding cases,
the sum of active power infeed was enough but there was an imbalance lo-
cally. But if the total power fed into the system by the prime movers is
less than what is consumed by the loads, including losses, this imbalance
will influence the frequency of the whole system. As explained above the
kinetic energy stored in rotating parts of the synchronous machines, and
other rotating electrical machines, will compensate for the imbalance result-
ing in a frequency deviation. If the imbalance is not too large the generators
participating in the frequency control will regulate the active power input
from their prime movers, and bring back the frequency deviation to accept-
able values. If the imbalance is too large, the frequency deviation will be
significant with possible serious consequences. Particularly thermal power
plants are sensitive to large frequency drops of long durations, since detri-
mental oscillations could be excited in the turbines. As a last resort the
generators are disconnected, making the situation even more serious. This
type of instability is called frequency instability and the time scale could be
from a few seconds up to several minutes. Since the involved mechanisms
could be quite different, one often distinguishes between short-term and
long-term frequency instability. In the latter, the control and protections
characteristics of turbines, boilers, and reactors play important roles.

Voltage Stability

When it comes to reactive power balance the situation is not as clear and
simple as concerning active power. There is always a balance between “pro-
duced” and “consumed” reactive power in every node of a network. This
is in fact a direct consequence of Kirchoff’s first current law. When one
talks about imbalance in this context we mean that the injected reactive
power is such, normally too small, that the voltage in the node cannot be
kept to acceptable values. (At low load the injected reactive power could
be high resulting in a too high voltage, possibly higher than the equipment
might be designed for. This is of course not desirable but it could usually
be controlled in such a way that no instabilities develop.) When we talk
about imbalance in this case we thus mean that the injected reactive power
differs from the desired injected reactive power, needed to keep the desired
voltage. If this imbalance gets too high, the voltages exceed the acceptable
range.
Reactive power is a more local quantity than active power since it can-
not be transported as easily in power system where normally X ≫ R. This
explains why voltage problems often are local, and often only occur in part
of the system. When the imbalances (voltage problems) develop into insta-
86 8. Classification and Definitions of Power System Stability

Power System
Stability

Rotor Angle Frequency Voltage

Stability Stability Stability

Small-Disturbance Transient Large- Small-

Angle Stability Stability Disturbance Disturbance
Voltage Stability Voltage Stability

Short Term Short Term Long Term

Short Term Long Term

Figure 8.2. Classification of power system stability.

bilities these are called voltage instabilities or voltage collapses. In the latter
case the instability develops into very low voltages in the system. In prin-
ciple too high voltages can also occur at a voltage instability. Low voltages
arise at high load conditions, while high voltages are associated with low
load conditions. Depending on the time scale the voltage instabilities are
classified as short-term, a couple of seconds, or long-term, tens of seconds to
minutes. The short-term voltage instability involves dynamics of fast acting
components such as induction motors, electronically controlled loads, and
HVDC converters, while the long-term voltage instability involves slower
acting equipment such as tap-changing transformers, thermostatically con-
trolled loads, and generator current limiters. As for rotor angle stability
one distinguishes between large-disturbance and small-disturbance voltage
stability.
The classification of power system instabilities is summarised as in Fig-
ure 8.2.

Connection between Instabilities and System Components

As explained above the generators, i.e. the synchronous machines, are very
important in angular instabilities, and it is sometimes said that these are
the driving force in this instability. A more detailed analysis shows that the
loads are very often the driving force when it comes to voltage instability,
which consequently sometimes is called load instability.
8.3. Literature on Power System Dynamics and Stability 87

Instabilities in Real Systems

The classification above is based on simplified and ideal conditions in the
system. In a real system these assumptions might not be valid. In real
systems it is not seldom a combination of active and reactive power imbal-
ances that trigger an instability. However, in many cases it is possible to
identify which is the dominating processes in the beginning of the instabil-
ity. During the course of the dynamics new consequential imbalances might
occur, resulting in a combined angular and voltage instability in the final
phase. There are examples of black outs in power systems that have started
as slow voltage instabilities, which through low voltages have reduced the
power transfer capability resulting in rotor angular instabilities causing the
final collapse of the system. On the other hand, rotor angular instabili-
ties can cause generators to trip, which most systems are designed to cope
with, but it can effect the reactive power balance in such a way that voltage
instabilities can develop.
The purpose of a classification is to define a structure for a complicated
problem and thereby better understanding it. Furthermore it often helps to
identify important and critical quantities, processes and components in the
system. Classifications of this kind should not be driven too far. Most im-
portant is always that useful and adequate results are obtained from realistic
models of the power system.

8.3 Literature on Power System Dynamics and Sta-

bility
These lecture notes should be seen as an introduction to power system dy-
namics and stability. For those who want to get deeper knowledge in the
subject there are a number of books that can be recommended. They have
all their strong and weaker sides, and one has often to consult several books
to get a complete view of a problem. Below are only references listed that
are fairly modern and focus on power system dynamics and stability.

1. Power System Stability and Control by Prabha Kundur. (McGraw-

Hill Inc., 1994, ISBN 0-07-035958-X, 1176 pages)
This is the most complete modern book on the subject and is already
a classic textbook. It covers most subtopics and its approach is rather
practical, but it contains a fair amount of theory also. The book
contains a lot of references to other books and published papers.

2. Power System Dynamics and Stability by Jan Machowski, Janusz W.

Bialek and James R. Bumby. (John Wiley & Sons Ltd, 1997, ISBN
0-471-97174-X, 461 pages)
This book does not contain as many applications as the previous one.
88 8. Classification and Definitions of Power System Stability

On the other hand, it contains rather detailed motivations and deriva-

tions of many of the basic assumptions in power system analysis. A
large part of the book is devoted to control and supervision of power
systems.

3. Power Systems Dynamics. Stability and Control by K. R. Padiyar.

(John Wiley & Sons Ltd, 1996, ISBN 0-471-19002-0, 629 pages)
The book gives a good overview with many solved problems. The
focus is on angular stability, while voltage stability is on briefly dealt
with.

4. Power System Dynamics and Stability by Peter W. Sauer and M. A.

Pai. (Prentice Hall, 1998, ISBN 0-13-678830-0, 357 pages) The mathe-
matical level in this book is higher than the previous ones and to fully
appreciate it knowledge in the theory of nonlinear systems is needed.
Only angular stability is included.

5. Power System Voltage Stability by Carson W. Taylor. (McGraw-Hill

Inc., 1994, ISBN 0-07-063184-0, 273 pages) This book deals only with
voltage stability. The approach is rather practical. Many examples
from voltage instabilities in real systems are reviewed and analysed.

6. Voltage Stability of Electric Power Systems by Thierry Van Cutsem

and Costas Vournas. (Kluwer Academic Publishers, 1998, ISBN 0-
7923-8139-4, 378 pages) Voltage stability is the subject of this book
also. However the approach is more mathematical than in the previous
one, and to fully appreciate it, knowledge about nonlinear systems is
required.

7. Definition and Classification of Power System Stability IEEE/CIGRE

Joint Task Force on Stability Terms and Definitions, 2002.
9
Synchronous Machine Models

This chapter starts with a short discussion that motivates why synchronous
machines are important when electro-mechanical oscillations are studied. A
brief overview of the operating principles and design of the synchronous ma-
chine is given. This forms the basis of simplified models that are derived –
first for stationary, then for dynamic operation. Even if the dynamic model
is very simple and neglects a number of features, it captures the most impor-
tant properties to model electro-mechanical oscillations in power systems.

S YNCHRONOUS MACHINES, i.e. practically all generators together with

synchronous motors and synchronous compensators, are the most impor-
tant power system components in the analysis of electro-mechanical oscilla-
tions in power systems. The oscillations are manifested in that the rotors
of the synchronous machines do not rotate with constant angular velocity
corresponding to system frequency, but superimposed are low frequency os-
cillations, typically 0.1 – 4 Hz. It is important that these superimposed
oscillations are not too large, because then the stability of the power system
can be endangered. A correct description of these oscillations often requires
detailed models of many different system components, but to get an un-
derstanding of and insight into the basic physical phenomena and processes
that determine the stability it is often sufficient to employ the simple model
that will be derived and motivated in this chapter. This simplified model of
the synchronous machine together with a simple model of the power trans-
mission system provide a description that will be used in chapter 11.
As the name electro-mechanical oscillations suggests, both electrical and
mechanical phenomena are involved, i.e. both currents in and voltages across
different windings in the machines but also the mechanical motion of the
rotor. Therefore, models of both electrical and mechanical parts of the
synchronous machine are needed.

9.1 Design and Operating Principle

In the synchronous machine a magnetized rotor creates a rotating magnetic
field in the air gap. If the rotor field is ideally sinusoidal and if the rotor
rotates at constant speed, this will induce ideally sinusoidal voltages in the

89
90 9. Synchronous Machine Models

R R
T S

T F S S T
F

D R D R

S T T S

S T
R R
Round rotor Salient pole rotor
p = 2, n = 3000 min−1 for f = 50 Hz p = 4, n = 1500 min−1 for f = 50 Hz

Figure 9.1. Cross-sections through different rotor types.

stator windings.
If the machine terminals are connected, the currents flowing in the stator
windings create a second rotating magnetic field which causes a torque on
the rotor. In a synchronous motor, this torque drives a mechanical load; in a
synchronous generator, the magnetic torque opposes the mechanical driving
torque of the prime mover (e.g. a turbine). Under balanced, steady-state
conditions the magnetic torque is equal to the mechanical torque, and so
the rotor continues to rotate at constant speed.

9.1.1 Rotor Types

The magnetic rotor field is generated by a field winding F on the rotor which
is fed with an adjustable direct current. In addition, the rotor has a short-
circuited damper winding D at the surface. This winding serves to dampen
electrical and mechanical oscillations and to shield the field winding from
inverse rotating fields in case of asymmetries or harmonics in the stator
currents. (In rotors without an explicitly realized damper winding, eddy
currents in the rotor iron can have a similar effect.)
Depending on the application of the generator, two different types of
rotors are used that are shown in Figure 9.1.

Round Rotor Round rotors are used with high-speed turbines such as
steam or gas turbines. For this reason, generators with round rotors are
also called turbo generators. They can have ratings as high as 1800 MVA
per unit. Due to the large centrifugal forces, the rotor consists of a long,
narrow, solid steel cylinder.
The field windings are mounted in slots that are mill-cut into about 2/3
of the perimeter. Because of the discrete distribution of the windings on the
9.1. Design and Operating Principle 91

T S
α

S T

Figure 9.2. Simplified arrangement of a stator with a solid, unwound rotor.

rotor surface, the magnetic flux density in the air gap always has a stair-step
form. Through proper distribution of the windings these stair-steps can be
made approximately sinusoidal!

Salient Pole Rotor Salient pole rotors are used with low-speed hydro tur-
bines with rated powers of up to 800 MVA per unit. In order to obtain
the appropriate electrical power frequency in spite of the low rotor speed,
salient pole rotors typically have multiple pole pairs. For run-of-river power
stations the number of poles can be as high as p = 200! Such rotors have
very large diameters (several meters) and short lengths.
The field windings are mounted on the individual poles. By properly
designing the geometric form of the poles, the magnetic flux density in the
air gap at the stator surface can also be made approximately sinusoidal!

9.1.2 Stator Field

The magnetic stator field is generated by an alternating three-phase current

flowing in the stator windings. The variation of this field in space and time
can best be studied using the simplified arrangement shown in Figure 9.2.
In this arrangement, the stator has only one pole pair and the slots are
assumed to be infinitely narrow. The rotor is replaced by a solid, unwound
cylinder. The stator field is derived here for a stator with one single pole
pair. The transfer to the case of a stator with p poles is left as an exercise
to the reader.
The magnetic field strength in the iron is neglected and the induction in
the air gap is assumed to be radial. Due to the π-symmetry of the set-up
92 9. Synchronous Machine Models

Ampère’s law gives for the dotted path shown in Figure 9.2

1 ~ d~s = B(α) · 2 · δ
ZZ I
Θ(α) = ~
i d~a = B (9.1)
µ0 µ0
A(α) S(α)

where A(α) is the area enclosed by the path, δ is the width of the air gap,
B(α) is the flux density in the air gap at the angle α, and Θ(α) is the total
flux linkage (or number of ampere windings) enclosed by the path.

One single-phase winding If only phase R is fed by an alternating current,

the total enclosed flux linkage is either +w · iR (t) or −w · iR (t), depending
on whether the path encloses the upper slot (0 < α < π) or the lower one
(π < α < 2π). w is the number of windings in one slot.
The left side of Figure 9.3 shows the flux linkage of a single-phase winding
plotted against the angle α for different instants of time. Since B(α) is
directly proportional to Θ(α) as shown in eq. (9.1), the magnetic field in the
air gap is a rectangular, standing wave with a variable amplitude.

Three-phase windings If the three phases R, S and T are fed by an al-

ternating three-phase current, the integration path in Figure 9.2 always
encloses three slots that carry windings with different momentary currents.
The resulting total flux as a function of the angle α is thus a superposition
of three rectangular curves that are shifted by 2/3 · π and have different
amplitudes.
This is shown on the right side of Figure 9.3. It is evident that the
fundamental component of the total flux is a traveling wave with a constant
amplitude which moves with the angular velocity ω. This angular velocity
of the stator field is the same as the angular frequency of the alternating
currents. In a real stator, the windings are of course not concentrated in
infinitely narrow slots. Instead, they are distributed among several slots
of finite extent – possibly with a different number of windings in different
slots. Through proper design the magnetic field in the air gap can be made
approximately sinusoidal.

It should be noted that this simplified treatment of the stator field is only
applicable for generators with round rotors where the air gap has a constant
width. In the case of a salient pole rotor, the width of the air gap is variable,
and therefore the magnetic stator field depends not only on the momentary
winding currents but also on the momentary rotational angle of the rotor.
In some cases, the differences between round rotor and salient pole rotor can
be neglected (as will be done frequently in the remainder of this chapter).
However, it should be kept in mind that this is a simplification which needs
to be justified from case to case.
9.1. Design and Operating Principle 93

ωt=0

ΘN ΘN

−ΘN 2π α −ΘN 2π α

2ΘN 2ΘN

2π α 2π α
−2ΘN −2ΘN

ω t = π/6

ΘN ΘN

−ΘN 2π α −ΘN 2π α

2ΘN 2ΘN

2π α 2π α
−2ΘN −2ΘN

ω t = π/3

ΘN ΘN

−ΘN 2π α −ΘN 2π α

2ΘN 2ΘN

2π α 2π α
−2ΘN −2ΘN

Figure 9.3. Magnetic field in the air gap of a stator with a solid,
unwound rotor at different instants of time. Left: one single-phase
stator winding; right: three-phase stator windings. The top plots show
the ampere windings in the individual slots; the bottom plots show the
total linkage.
94 9. Synchronous Machine Models

dα

δ
D

Figure 9.4. Geometric quantities used to calculate the magnetic en-

ergy stored in the air gap. The width of the air gap is assumed to be
much smaller than the rotor diameter: δ ≪ D. Rotor and stator have
the length l.

9.1.3 Magnetic Torque

The magnetic fields generated by the stator and rotor currents are super-
posed to a total field
B = Bs + Br (9.2)

with
p
Bs = B̂s sin( (α + ωt))
2 (9.3)
p
Br = B̂r sin( (α + ωt + θ))
2
Both fields are assumed to be ideally sinusoidal and to rotate with the same
angular velocity ω; the rotor field (together with the rotor itself) leads the
stator field by the angle θ. p is the number of poles.
Using the geometry shown in Figure 9.4, the energy stored in the mag-
netic field of the air gap is given by

Z2π
B 2 (α, t, θ) 1 D
Z
Wm = dV = B 2 (α, t, θ) · lδ dα (9.4)
2µ0 2µ0 2
V 0

The torque on the rotor is obtained by calculating the derivative of the

energy with respect to the displacement angle θ between the rotor and the
stator field
Z2π
∂Wm lδD ∂B 2 (α, t, θ)
M= = dα (9.5)
∂θ 4µ0 ∂θ
0

Using several trigonometric identities together with the fact that only Br
depends on θ, this can be simplified to
p
M = Mmax · sin(− θ) (9.6)
2
9.2. Stationary Operation 95

−1.2

−1

−0.8
Rotor torque

−0.6

−0.4

−0.2

−0
20 40 60 80 100 120 140 160 180
0.2

Figure 9.5. Rotor torques in function of the rotor angle θ. Solid line:
round rotor torque; dotted line: reluctance torque; dashed line: salient
pole rotor torque.

where
lδDπ
Mmax = B̂s B̂r (9.7)
4µ0
is the maximum torque that the machine can exert on the rotor. If the
mechanical torque from the turbine exceeds this limit, the generator will
fall out of step. If no protective measures are taken to reduce the turbine
power, the rotor will accelerate until the centrifugal forces eventually destroy
the generator! This is particularly dangerous in the case of turbo generators,
that have a high speed to begin with.
As noted before, the stator field can only be regarded as sinusoidal for
a turbo generator. In the salient pole generator, the inhomogeneous air gap
gives rise to a reluctance torque Mr ∼ sin(pθ) which is added to the torque
given in eq. (9.6). This is shown in Figure 9.5 for p = 2.

9.2 Stationary Operation

9.2.1 Stationary Single Phase Equivalent Circuit
If the rotor rotates at constant speed the rotor field will induce sinusoidal
voltages in the stator windings that are shifted by 120◦ with respect to
each other. The frequency of the voltages is equal to the rotor frequency
(multiplied by the number of pole pairs). The magnitude of the voltages
depends on both the rotor speed and the magnitude of the rotor field; it can
therefore be controlled via the rotor field current. Due to the symmetry of
the voltages, the generator can be represented by the single phase equivalent
circuit shown in Figure 9.6. In this circuit the induced voltages are modeled
96 9. Synchronous Machine Models

jX d I

E U

Figure 9.6. Stationary single phase equivalent circuit.

by the voltage source E; I is the stator current, and U is the terminal voltage
of the generator.
With the losses neglected, the total impedance of the generator as seen
from the stator can be modeled with a single lumped inductance Xd . This is
strictly only valid for the turbo generator with its homogeneous air gap. In
the salient pole generator, the same effects that give rise to the reluctance
torque mentioned earlier also cause the active part of the stator current (with
respect to the induced voltage E), to “see” a smaller reactance Xq than the
reactive part, for which the reactance Xd is effective. This is illustrated in
Figure 9.7. Typical values for Xd and Xq are listed in Table 9.1. For our
purposes in this lecture we will neglect these differences (but keep in mind
that they exist).
If the terminal voltage U is assumed to lie on the real axis, the phasor
equation is
E = U + jXd I (9.8)
or – in polar form –
E · (cos θ + j sin θ) = U + Xd I · (j cos ϕ + sin ϕ) (9.9)
where θ is the angle between E and the real axis counted counter-clockwise,
and ϕ the angle between I and the real axis counted clockwise. Looking
only at the imaginary part gives
E
· sin θ = I · cos ϕ (9.10)
Xd
The active power of the machine (which is directly proportional to the
mechanical torque) is given by
P = U I cos ϕ (9.11)
Using eq. (9.10) this can be written as
UE
P = sin θ (9.12)
Xd
A comparison with eq. (9.6) reveals that – for a machine with a single pole
pair – the mechanical angle between the magnetic fields of rotor and stator
is identical to the electrical angle between the phasors of the induced voltage
E and the terminal voltage U .
9.2. Stationary Operation 97

d−axis
Iq E
q−axis
θ
jXqIq
I ϕ U
d jXdId

Figure 9.7. Phasor diagram of voltages and currents of the salient pole machine.

Round Rotor Salient Pole Rotor

Xd (p.u.) 1.0 – 2.3 0.6 – 1.5

Xq (p.u.) 1.0 – 2.3 0.4 – 1.0

Table 9.1. Typical values of reactances for synchronous machines.

9.2.2 Phasor diagram

Figure 9.8 illustrates the relation between the three phasors for different
modes of operation. They can be categorized as described below. It is
assumed that the terminal voltage phasor, U , is the reference phasor, i.e.
parallel with the real axis. The angle between the induced voltage phasor
and the terminal voltage phasor is θ, and the angle between the terminal
voltage phasor and the stator current phasor is ϕ. It should be noted that
the statements concerning leading and lagging currents are based on the
current reference definition of Figure 9.7.

Generator operation (right column): |ϕ| < 90◦ , θ > 0

The machine delivers active power. The stator current phasor lies in
the right half-plane; the induced voltage phasor leads the terminal
voltage phasor.

Motor operation (left column): |ϕ| > 90◦ , θ < 0

The machine consumes active power. The stator current phasor lies
in the left half-plane; the induced voltage phasor lags the terminal
voltage phasor.

Overexcitation (top row): |E| · cos θ > |U |

The machine delivers reactive power (i.e. it acts like a capacitor). The
stator current phasor lags the terminal voltage phasor; the real part
of the induced voltage phasor is larger than the terminal voltage.
98 9. Synchronous Machine Models

Motor Generator

Im Im

E
jXdI
overexcited

U U
θ
I ϕ Re I Re
jX I
d
E

Im Im

E jXdI
underexcited

I
U I
U
Re Re
E jXdI

Figure 9.8. Phasor diagrams for different modes of operation of a

synchronous machine.

Underexcitation (bottom row): |E| · cos θ < |U |

The machine consumes reactive power (i.e. it acts like an inductor).
The stator current phasor leads the terminal voltage phasor; the real
part of the induced voltage phasor is smaller than the terminal voltage.
As a special case consider a synchronous machine where the rotor is
running idle. In steady state the mechanical torque is zero and therefore
the machine neither consumes nor delivers active power (θ = 0). Neverthe-
less, depending on the excitation it can either consume (underexcitation)
or deliver (overexcitation) reactive power, i.e. ϕ = ± 90◦ . Such a set-up
is called synchronous compensator and is used for variable reactive power
compensation.

9.2.3 Operational Limits

Solving eq. (9.8) for I gives
U E
I=− + (9.13)
jXd jXd
If the terminal voltage U is considered constant, the stator current is
determined by two parameters. According to eq. (9.11), the real part of
9.2. Stationary Operation 99

−Im(I)
Q
← Stator current thermal limit

← Rated operating point

I ← Field current thermal limit

E/jX Re(I)
d
−U/jX P
d

← Practical stability margin (20°)

θ

← Safety margin against 10% voltage drops

← Theoretical stability limit

Figure 9.9. Operational limits of the synchronous generator.

the stator current is given by the turbine power. The imaginary part can
then be controlled by adjusting the magnitude of the induced voltage via
the rotor field current.
Eq. (9.13) is illustrated in Figure 9.9 in the P-Q plane. The limits of the
operating area can be constructed as follows.

Stator current thermal limit: |I| < Imax

The stator current limit is a circle around the origin with radius Imax .
f (If,max )
Field current thermal limit: |E| = f (If ) ⇒ |E| Xd < Xd
U
The field current limit is a circle around − jXd .

Rotor angle stability limit:

The theoretical rotor angle stability limit is determined by θ < 90◦ . It
U
is a parallel of the real axis that passes through − jX d
. In practice, a
◦
safety margin of 20 is kept to this limit.

Safety margin against voltage drops:

If the generator delivers little or no active power and operates close to
the stability limit (underexcitation), even small voltage drops at the
machine terminals could cause the rotor angle to exceed 90◦ . To allow
100 9. Synchronous Machine Models

for 10 % voltage drops, the operating point must stay above a parallel
U
of the real axis passing through −0.9 · jX d
.

Depending on the design of the generator, there may be other limiting factors
(e.g. for turbo generators the stator end region heating limit mentioned in
section 2.5).

9.3 Dynamic Operation

9.3.1 Transient Single Phase Equivalent Circuit
During network transients, the reactance of the synchronous generator is not
constant. For symmetrical transients, as discussed in section 7.2 on short
circuits of synchronous machines, the machine reactance itself undergoes
transient changes as the machine passes through the sub-transient, transient,
and steady-state stages. In the fault analysis in chapter 7, only the first few
cycles after a fault were of interest. The generator was therefore modeled
with the sub-transient reactance Xd′′ .
Compared to the time scale of short-circuit overcurrents, electro-mechan-
ical oscillations are much slower (0.1 – 4 Hz as mentioned in the introduction
to this chapter). For the study of these phenomena, the sub-transient phase
is therefore neglected and the generator is modeled with the transient reac-
tance Xd′ . Similarly, during transients, the magnitude of the induced voltage
drops to the value E ′ which can be calculated as follows

Xd′ Xd − Xd′
E′ = · E0 + · U 0 · cos θ0 (9.14)
Xd Xd
or
E ′ = E 0 − (Xd − Xd′ ) · I 0 · sin ϕ0 (9.15)
where the superscript 0 denotes the steady-state quantities prior to the tran-
sient.
Thus, the phasor equation valid during network transients is

E ′ = U + jXd′ I (9.16)

9.3.2 Simplified Mechanical Model

As stated above the rotor of a synchronous machine rotates with synchronous
speed in steady state. If the electrical frequency of the system is ωe , the
mechanical angular velocity ωm of the rotor is given by
ωe
ωm = (9.17)
p/2

where p is the number of poles of the machine.

9.3. Dynamic Operation 101

Type of Synchronous Machine Inertia Constant H (s)

Thermal Power
• Steam Turbine 4–9
• Gas Turbine 7 – 10
Hydro Power
• Slow (< 200 min−1 ) 2–3
• Fast (≥ 200 min−1 ) 2–4
Synchronous Compensators 1 – 1.5
Synchronous Motors ≈2

Table 9.2. Typical values of H for different types of synchronous machines.

An important parameter in the analysis of rotor oscillations is the total

moment of inertia of the synchronous machine J. This is the sum of all
moments of inertia of all rotating parts of the synchronous machine, i.e.
the sum of the moments of inertia of the rotor, turbines, shafts and other
devices on the shaft system, e.g. generator feeding the field winding. As for
electrical quantities it is practical to express J in a suitable p.u. base and
therefore the inertia constant of the synchronous machine H is defined as
2
0.5Jωmo
H= (9.18)
S
where S is the MVA rating of the machine. In eq. (9.18) the numerator is an
expression for the total kinetic energy stored in the synchronous machine in
steady state and the unit for H is thus seconds. (If there is a gear box in the
system, it is of course difficult to define one mechanical angular velocity. In
such a case H is calculated as the ratio between total stored kinetic energy
and the MVA rating of the machine.) The inertia constant states how much
time it would take to bring the machine from synchronous speed to standstill
if rated power is extracted from it while no mechanical power is fed into it.
The value of the inertia constant will vary within a much smaller range than
the value of J for different machines. Table 9.2 shows typical values of H
for different types of synchronous machines. It can be concluded that the
value is higher for thermal units as compared with hydro units. Typically
30 – 60% of the total moment of inertia comes from the turbines and shafts
for thermal units, while the corresponding value for hydro units is 5 – 15%.
For synchronous motors the inertia constant depends to a high degree of
what kind of load that is connected.
102 9. Synchronous Machine Models
10
The Swing Equation

This chapter contains a derivation and a discussion of the swing equation,

which is the basic mathematical relation describing how the rotor of a syn-
chronous machine will move (swing) when there is an unbalance between
mechanical power fed into the machine and the electrical power extracted
from it. Even if the basic physical relations are very simple, a number of
simplifications must be done to make the equation useful in practice.

T HE SWING EQUATION is of fundamental importance in the study of

power oscillations in power systems. Electro-mechanical oscillations is
an important phenomenon that must be considered in the analysis of most
power systems, particularly those containing long transmission lines, as e.g.
the Swedish system. In normal steady state operation all synchronous ma-
chines in the system rotate with the same electrical angular velocity, but
as a consequence of disturbances one or more generators could be acceler-
ated or decelerated and there is risk that they can fall out of step, i.e. lose
synchronism. This could have a large impact on system stability and gen-
erators losing synchronism must be disconnected otherwise they could be
severely damaged. A short description of power system protections is given
in Appendix B.

10.1 Derivation of the Swing Equation

The synchronous machine models developed in the previous chapter will
form the basis for the derivation of the swing equation describing the electro-

Pe Pe
Te Tm

Pm Generator Pm Motor

Tm ωm ωm Te

Figure 10.1. Schematic description of powers and torques in syn-

chronous machines.

103
104 10. The Swing Equation

mechanical oscillations in a power system. Schematically the different torques

and powers of a synchronous machine can be depicted as in Figure 10.1. In-
dex m denotes in the following a mechanical quantity, and index e denotes an
electrical quantity. The differential equation describing the rotor dynamics
is
d 2 θm
J = Tm − Te (10.1)
dt2
The quantities in eq. (10.1) are:
J: The total moment of inertia of the synchronous machine (kg·m2 )
θm : The mechanical angle of the rotor (rad)
Tm : Mechanical torque from turbine or load (N·m). Positive Tm corresponds
to mechanical power fed into the machine, i.e. normal generator operation
in steady state.
Te : Electrical torque on the rotor (N·m). Positive Te in normal generator
operation.
If eq. (10.1) is multiplied with the mechanical angular velocity ωm one
gets
d 2 θm
ωm J = Pm − Pe (10.2)
dt2
where
Pm = Tm ωm = mechanical power acting on the rotor (W)
Pe = Te ωm = electrical power acting on the rotor (W)

If the angular acceleration should be expressed in electrical angle instead,

eq. (9.17) is used to give

2 d 2 θe
ωm J 2 = Pm − Pe (10.3)
p dt
where the left hand side can be re-arranged:
2 1 2 d 2 θe
2 ( ω J) = Pm − Pe (10.4)
pωm 2 m dt2
If eq. (10.4) is divided by the rating of the machine S, and eq. (9.17) is
utilised again, the result is

2 ( 21 ωm
2 J) 2
d θe Pm − Pe
2
= (10.5)
ωe S dt S
Observations and experiences from real power systems show that during
disturbances, the angular velocity of the rotor will not deviate significantly
from the nominal values, i.e. from ωm0 and ωe0 , respectively. This implies
that eq. (10.5) together with the definition eq. (9.18) can be written as

2H d2 θe pu
= Pm − Pepu (10.6)
ωe0 dt2
10.2. Analysis of the Swing Equation 105

where superscript pu indicates that the mechanical and electrical powers

should be expressed in p.u. of the rating of the synchronous machine. It is
of course possible to use another base power than the rating of the machine in
eq. (10.6), but this must then be made consistently regarding the definition
of H and when calculating the power values in the right hand side. If
another base power than the rating is used, the physical interpretation of
H made in section 9.3.2 will not be valid anymore. Furthermore the typical
values of H in Table 9.2 do not apply. In the following it will be assumed, if
not otherwise explicitly stated, that electrical angles and electrical angular
velocities are considered, and consequently the index e in the left hand side
of eq. (10.6) can be omitted. It is also assumed that powers are expressed
on the same base power as H, and the superscript pu can also be omitted
in eq. (10.6). We are thus going to use the following form of the swing
equation:
2H d2 θ
= Pm − Pe (10.7)
ω0 dt2

10.2 Analysis of the Swing Equation

Before applying the swing equation to a specific system, it could be of value
to briefly discuss the different terms of eq. (10.7) and their influence. This
will give an insight into the fundamental relations governing the dynamics
during rotor oscillations.
The difference between mechanical power fed into the machine and the
electrical output power will cause a motion of the rotor relative to a rotation
with constant angular velocity ω0 .1 Therefore it is of interest to discuss in
some detail how Pe and Pm will vary during a rotor swing.
The mechanical power Pm is in most cases provided by a hydro, steam
or gas turbine. This power is determined by the gate opening of the tur-
bine, and the time constant for changing the mechanical power is in most
cases several seconds. There are different reasons why the mechanical power
should be changed. For each generator there is a dispatch plan made for each
hour. This plan is determined by the expected load in the power system.
These plans are often the result of extensive optimisations of the resources
available with due consideration to load, prices, cost, etc. The mechanical
input powers are changed in accordance with these dispatch plans. But it
is not possible to exactly forecast the power consumption in the system.
Furthermore, generators could suddenly be disconnected due to faults, re-
sulting in a shortage of power in the system. These unbalances will cause a
1
In the following it is assumed that the synchronous machine is a generator. The same
discussion could be applied to a synchronous motor, but then the signs of the electrical and
mechanical powers must be changed. But since the dynamics of the system is dominated
by the generators, it is natural to use these as the basis for type discussion. However, there
is no principal difference in the dynamics between a synchronous motor and generator.
106 10. The Swing Equation

frequency deviation in the system. This frequency deviation can be used as

an input signal to selected generators, which will change their mechanical
input power and thereby their electrical output so that the frequency devia-
tion is brought down to acceptable levels, typically ±0.1 Hz in interconnected
systems. The frequency controller could be rather fast, but since it takes
several seconds to change the mechanical power Pm it will take some time
for the frequency control to act. Generally it can be said that Pm = Pm (θ̇)
for those generators that are equipped with frequency controllers, but if time
scales up to some seconds are considered it is a fair approximation to say
that Pm = constant, at least for studies and discussion of more principal
nature.
In section 9.2 an expression for the active power of a synchronous ma-
chine connected to an infinite bus was derived, eq. (9.12). As can be seen,
the power depends on the angle θ, and the rotor angle occurs thus on both
sides of the swing equation. This applies for a synchronous machine con-
nected to an infinite bus, but it can be shown to be the case also for more
complex systems with more synchronous machines. In this latter case, the
electrical power of a given machine will not only depend on the value of its
own rotor angle, but also on the rotor angles of other machines. The result
is thus a system of coupled differential equations.
A more detailed analysis of the electrical power shows that it depends not
only on θ but also on θ̇, that is on the relative angular velocity as compared
with a synchronously rotating system. This contribution, which only occurs
during transients, is due to currents induced in the rotor circuits, and it
tends to damp out the oscillations. A voltage controller might also give a
contribution to the electrical power depending on θ̇.
With fairly good accuracy the electrical power can be written as
Pe = Pe (θ, θ̇) = Ps (θ) + Pd (θ̇) (10.8)
with Ps being the synchronising power and Pd being the damping power.
This is further explained in section 11.4.

10.3 Swing Equation as System of First Order Differ-

ential Equations
The swing equation (10.7) is an ODE of the second order.2 ODEs of higher
orders can be written as a system of first order ODEs, which in many cases
is practical. Particularly for multi-machine systems this turns out to be
an attractive approach. In most cases it is not possible to solve the swing
equation analytically, but one has to use numerical integration in solving it.
Numerical integration usually requires that the ODE is in the form of ODEs
of first order.
2
The abbreviation ODE for Ordinary Differential Equation is here introduced.
10.3. Swing Equation as System of First Order Differential Equations 107

By introducing ω = θ̇ eq. (10.7) can be written as

 " #
ω

θ̇ ω0
= (10.9)
ω̇ (Pm − Pe )
2H
This form of the swing equation will also be used in the following. The
quantities θ and ω in eq. (10.9) are named state variables, or just states,
and eq. (10.9) describes the system in state form or standard form. The
vector [θ ω]T is the state vector of the system.
Some clarifications will be made concerning the angular velocity ω in-
troduced in eq. (10.9) above. This angular velocity denotes the frequency
with which the rotor oscillates relative to a system rotating with the syn-
chronous and constant angular velocity ω0 . The absolute angle of the rotor
θabs relative to a reference system at time t is given by

θabs = ω0 t + θ + θ0 (10.10)

where θ0 is the angle of the rotor at t = 0. Differentiation of eq. (10.10)

with respect to time gives
θ̇abs = ω0 + θ̇ (10.11)
from which it can be concluded that the absolute angular velocity of the
rotor ωabs is given by
ωabs = θ̇abs (10.12)
and the angular velocity relative the synchronously rotating system ω is
given by
ω = θ̇ (10.13)
It is this latter relative angular velocity that is of interest when rotor oscil-
lations are studied, while the absolute angular velocity is of interest when
studying frequency stability.
108 10. The Swing Equation
11
Power Swings in a Simple System

In this chapter the models and equations derived earlier are applied to the
simple system of a synchronous machine connected to an infinite bus. De-
spite the simplicity of the system, useful and general information can be
gained from this analysis. Through qualitative discussions the features of
the equilibrium points of the system can be determined. Time domain sim-
ulations of the system show the principal behaviour of stable and unstable
solutions. The equal area criterion is derived. This is a very powerful method
for determining the stability of a system. Another powerful tool is provided
by small signal analysis, or linear analysis, which gives valuable informa-
tion about the local behaviour of the system. The chapter is concluded by
a discussion of different ways of improving the angular stability of a power
system.

A COMPLETE STABILITY analysis of a power system is an extensive

and complicated task. However, it turns out that many of the most
important phenomena and mechanisms can be found in very simple systems,
where they can be seen very clearly. In large and complicated systems it is
often hard to distinguish the fundamental and decisive phenomena from the
more irrelevant ones. It is therefore of importance to study simple systems
to get an insight into and understanding of the basics, that can be used
in the analysis of more complex systems. This chapter focuses on power
oscillations in the simple system of Figure 11.1. This system can be a model
of a synchronous machine or group of synchronous machines connected to
a larger system through one or more power lines. The reactance Xe in
Figure 11.1 is an equivalent reactance including transformers and parallel
lines. Even if this system is simple, a number of simplifications are required
to get simple solutions to the system.
The reader is encouraged to solve problems from the problem set to get
a better understanding of the issues discussed here.

11.1 The Swing Equation and its Solutions

The solutions to the swing equations of the simple system in Figure 11.1
will be analysed and discussed in this sub-section. Despite its simplicity, a

109
110 11. Power Swings in a Simple System

number of important conclusions concerning the angular stability in large

systems can be drawn from this example. Earlier a number of simplifications
and assumptions have been made, more or less explicitly, and for the sake
of completeness and clarity they are summarised here:

1. The synchronous machine is modelled as a constant electro-magnetic

field behind the transient reactance Xd′ . The angle of the electro-
magnetic field is assumed to coincide with the rotor angle.

2. Resistances in lines, transformers, and synchronous machines are ne-

glected.

3. Voltages and currents are assumed to be perfectly symmetrical, i.e.

pure positive sequence.

4. The angular velocity is close to the nominal one.

5. Static models for lines are used.

6. The mechanical power Pm , i.e. the power from the prime mover, is
constant during the transient under study.

Furthermore, it is assumed that the damping power of the system can writ-
ten as
Pd = Dθ̇ (11.1)
An electrical equivalent to Figure 11.1 is shown in Figure 11.2. Together
with the assumptions above the swing equation can now be formulated as

2H d2 θ
= Pm − Pe (11.2)
ω0 dt2

with
Eq′ UN
Pe = sin θ + Dθ̇ (11.3)
Xd′ + Xe
and
Pm = Pm0 = constant (11.4)

Xe
~ Infinite bus

U N ∠0

Figure 11.1. Synchronous machine connected to infinite bus.

11.1. The Swing Equation and its Solutions 111

X d′ Xe

E q′ ∠θ
~ U N ∠0

Figure 11.2. Equivalent electric circuit of a synchronous machine

connected to an infinite bus.

Now we introduce
Eq′ UN
Pe,max = (11.5)
Xd′ + Xe
and eq. (11.2) can be written as

2H d2 θ
= Pm0 − Pe,max sin θ − Dθ̇ (11.6)
ω0 dt2
which is the complete swing equation of the simplified system with the in-
troduced assumptions.

11.1.1 Qualitative Analysis

In order to make a qualitative analysis of the solutions to eq. (11.6) a further
simplification will be made, i.e. the damping is neglected and D is thus set
to zero :
2H d2 θ
= Pm0 − Pe,max sin θ (11.7)
ω0 dt2
Of fundamental importance to a non-linear system are its equilibrium
points, i.e. the points in state space where all time derivatives vanish. For
the equilibrium points the right hand side of eq. (11.7) is thus zero resulting
in the following equation for θ0

Pm0 = Pe,max sin θ0 (11.8)

Figure 11.3 shows how Pe and Pm vary with rotor angle θ, and the following
conclusions about the equilibrium points can be drawn from this figure:

1. If Pm0 < Pe,max there are two equilibrium points, i.e. θ0 and π − θ0
for 0 ≤ θ ≤ π.

2. If Pm0 = Pe,max there is exactly one equilibrium point θ0 = π/2 for

0 ≤ θ ≤ π.

3. If Pm0 > Pe,max there is no equilibrium point.

112 11. Power Swings in a Simple System

Pe, max sin θ

P(θ)
e

Pm0

0
θ0 π−θ0 π
0

Figure 11.3. Diagram showing the variation of electric and mechan-

ical power for the system in eq. (11.7).

(The angle θ0 is given by θ0 = arcsin(Pm0 /Pe,max ).)

It is clear that if Pm0 > Pe,max the system is unstable and further analysis
is superfluous. In this case the rotor will be accelerated until the protections
trip the generator and turbine. A necessary condition for stability is that
at least one equilibrium point exists.
In the case of Pm0 < Pe,max the equilibrium point θ = θ0 is stable for
(sufficiently) small disturbances, which implies that if the system is moved
away from θ0 the dynamics of the system tends to bring it back to θ0 . This
is verified by the following qualitative reasoning:
If the rotor has been accelerated so that θ > θ0 , the right hand side of
eq. (11.7) will be negative and the system is decelerated and starts to move
back to θ0 (provided ω is sufficiently small). Correspondingly, a deviation
θ < θ0 will give an acceleration so that θ is brought back to θ0 . (As seen from
Figure 11.3 these arguments are valid only for sufficiently small deviations
from θ0 and ω = 0.) It is seen that if θ > π − θ0 and if ω ≥ 0 then the
system will never return to θ = θ0 .
For a state vector [θ ω]T in the two dimensional state space, or phase
space as it is sometimes referred to, that initially is close to [θ ω]T = [θ0 0]T ,
will be close to this point for all t > 0. As in a real system there is posi-
11.1. The Swing Equation and its Solutions 113

tive (hopefully) damping, so the solution will converge (asymptotically) to

[θ0 0]T . The region that has been described above as close to [θ0 0]T is
called the region of attraction of this equilibrium point, and the curve that
separates this region from points that will not converge to [θ0 0]T , is called
the separatrix. This curve separates the stable and unstable solutions from
each other.
A similar discussion concerning the equilibrium point θ = π − θ0 shows
that this point is unstable.
A stability criterion for the system in Figure 11.3 can now be formulated:

Stability Criterion
The system in Figure 11.3 is stable, i.e. the synchronous machine will not
fall out of step, if it is after a disturbance in the region of attraction of the
equilibrium point [θ0 0]T . 1

The criterion above implies that the system must have decelerated so that
ω = 0 before the point θ = π − θ0 is reached, otherwise the generator will
fall out of step. The discussion above has disclosed a number of important
features of the system without that any equation has been solved. The
solutions to eq. (11.7) are not easy to obtain analytically. It is seen that
eq. (11.7) is the same as the equation describing a mathematical pendulum,
and the exact solutions are given by elliptic integrals. Usually a mathemat-
ical pendulum is studied by its linearised equations, but one cannot get all
information regarding stability from these equations. The linearised equa-
tions can however give other kinds of useful information, and we will come
back to that in section 11.4.
In the next section the behaviour of the stable and unstable solutions
will be discussed.

11.1.2 Stable and Unstable Solutions

In order to verify the conclusions drawn from the qualitative discussion
above simulated solutions to eq. (11.7), i.e. for the system in Figure 11.1,
will be shown for a few cases.
The disturbance considered here is a three phase to earth fault on one
of the lines close to the generator. That the fault is close to the generator
implies that the electric power during the fault is zero. The faulty line is
disconnected by the distance protections, see Appendix B.2, causing that
the reactance Xe is changed when the fault is cleared. By varying the
1
It is clear that θ = θ0 is the only possible stable equilibrium point. P (θ) is periodic
with the period 2π, but if the system has moved so that θ is close to another “stable”
equilibrium point, ω will be so large that it is outside the region of attraction of this
equilibrium point.
114 11. Power Swings in a Simple System

Electrical Power Before and After the Fault

2

1.8

1.6

1.4
Electric Power (p.u.)

1.2

0.8

0.6

0.4

0.2

0
0 20 40 60 80 100 120 140 160 180
Rotor Angle (degrees)

Figure 11.4. The electric power before the fault, Pe = 1.86 sin θ,
after the fault, Pe = 1.26 sin θ, and the mechanical power, Pm = 1, as
function of the rotor angle, θ.

fault clearing time different types of solutions to the swing equation can be
obtained.
The system is thus described by the following equation

2H d2 θ
= Pm0 − Pe,max sin θ − Dθ̇ (11.9)
ω0 dt2
The electric power is as follows: Pre-fault Pe,max = 1.86 p.u., post fault
Pe,max = 1.26 p.u. and during the fault Pe = 0. Power curves before and
after fault are shown in Figure 11.4 where also the constant power from the
turbine Pm = 1 p.u. is drawn. Before the fault θ = 32.5◦ , which corresponds
to the left intersection between Pe = 1.86 sin θ and Pm = 1. During the fault
the rotor will accelerate since Pe = 0 and consequently Pm − Pe > 0. When
the fault is cleared the electric power will follow the curve Pe = 1.24 sin θ.
If the system is stable it will settle down to the point θ = 53.5◦ which is
the left intersection between Pe = 1.24 sin θ and Pm = 1. When the rotor
angle is between 53.5◦ and 180 − 53.5 = 126.5◦ , which corresponds to the
right intersection between Pe = 1.24 sin θ and Pm = 1, then Pm − Pe < 0
11.1. The Swing Equation and its Solutions 115

and the rotor will decelerate. Should the rotor move beyond this point, then
Pm − Pe > 0 and the rotor will accelerate and the system becomes unstable.
It is often said that the generator falls out of step or loses synchronism.
The longer the duration of the fault, the more the rotor will be accel-
erated and the larger the rotor angle. If the duration of the fault is too
long there is a risk that the rotor swing is so large that it passes the right
equilibrium point and the synchronous machine loses synchronism. To ver-
ify and demonstrate this the system has been simulated for fault clearing
times equal to 4, 6, 6.5, and 8 cycles.2 Results from these simulations are
shown in Figures 11.5 to 11.8 where both the swing curves and the phase
portraits of the system are shown. The phase portrait, or phase trajectory,
is the trajectory of the system in the (ω θ)-plane during the transient. The
damping D is set to 0.02. The value of the damping determines how fast
a stable solution will converge to its equilibrium point, and this value has
minor influence on the first swing directly after the fault.3
With a fault clearing time of 4 cycles the system is clearly stable, see
Figure 11.5. The maximal rotor angle is ≈ 90◦ which gives an ample margin
to the critical value of 126.5◦ . The system will eventually converge to the
point θ = 53.5◦ .
If the fault clearing time is increased to 6 cycles, see Figure 11.6, the
stability margin is much smaller. The rotor swings out to ≈ 120◦ and
it is very close to the critical point where it will start to accelerate. If
the fault clearing time is increased to 6.5 cycles the curves of Figure 11.7
are obtained, and here the system is unstable. The generator cannot be
decelerated enough before it reaches the critical point, but it passes this point
and is further accelerated and loses synchronism. This type of instability
is called first swing instability. The phase portrait in this case shows that
rotor is close to be decelerated before reaching the critical point, but it never
reaches a zero value of ω. The previous fault clearing time, i.e. 6 cycles, is
called critical fault clearing time.
The last case shows a simulation with the fault clearing time equal to 8
cycles, see Figure 11.8. In this case the rotor is accelerated for such a long
time that it passes the critical point very soon. The deceleration period
could hardly be seen in the swing curve. The phase portrait shows also that
2
Fault clearing times are often given in cycles of the power frequency. The reason for
this is that part of the clearing time comes from the time needed for the breakers to clear
the fault. Since the fault current can only be intercepted at a zero crossing, it is natural
to measure the breaker time in cycles. Modern high voltage breakers can break a fault
current in 2 – 3 cycles. Should still faster breaker times be required, special breakers could
be installed, but these are more expensive than standard breakers. Protections are also
described in Appendix B.
3
This is the case for realistic values of D. If the damping is very large, and positive,
this will increase the stability significantly in the system. The value used here is typical
in a system where no special equipment has been installed to increase the damping, e.g.
Power System Stabilisers on the voltage regulators.
116 11. Power Swings in a Simple System

the deceleration just after fault clearing is clearly insufficient to stabilise the
system.
11.1. The Swing Equation and its Solutions 117

Swing Curve for Generator Against Infinite Bus

150

100
Rotor Angle (degrees)

50

0
0 0.5 1 1.5 2 2.5 3
Time(s)
Phase Portrait of Generator Against Infinite Bus

4
Relative Angular Velocity of Rotor(rad/s)

−2

−4

−6

0 50 100 150
Rotor Angle (degrees)

Figure 11.5. Swing curve and phase portrait for generator swinging
against an infinite bus. Fault clearing time = 4 cycles. Critical fault
clearing time = 6 cycles.
118 11. Power Swings in a Simple System

Swing Curve for Generator Against Infinite Bus

150

100
Rotor Angle (degrees)

50

0
0 0.5 1 1.5 2 2.5 3
Time(s)
Phase Portrait of Generator Against Infinite Bus

4
Relative Angular Velocity of Rotor(rad/s)

−2

−4

−6

0 50 100 150
Rotor Angle (degrees)

Figure 11.6. Swing curve and phase portrait for generator swinging
against an infinite bus. Fault clearing time = 6 cycles. Critical fault
clearing time = 6 cycles.
11.1. The Swing Equation and its Solutions 119

Swing Curve for Generator Against Infinite Bus

150

100
Rotor Angle (degrees)

50

0
0 0.5 1 1.5 2 2.5 3
Time(s)
Phase Portrait of Generator Against Infinite Bus

4
Relative Angular Velocity of Rotor(rad/s)

−2

−4

−6

0 50 100 150
Rotor Angle (degrees)

Figure 11.7. Swing curve and phase portrait for generator swinging
against an infinite bus. Fault clearing time = 6.5 cycles. Critical fault
clearing time = 6 cycles.
120 11. Power Swings in a Simple System

Swing Curve for Generator Against Infinite Bus

150

100
Rotor Angle (degrees)

50

0
0 0.5 1 1.5 2 2.5 3
Time(s)
Phase Portrait of Generator Against Infinite Bus

4
Relative Angular Velocity of Rotor(rad/s)

−2

−4

−6

0 50 100 150
Rotor Angle (degrees)

Figure 11.8. Swing curve and phase portrait for generator swinging
against an infinite bus. Fault clearing time = 8 cycles. Critical fault
clearing time = 6 cycles.
11.2. Equal Area Criterion 121

11.2 Equal Area Criterion

It was concluded above that even for quite simple systems, like the one in
Figure 11.1 it was hard to get analytical solutions to the system equations.
But quite often the exact behaviour of the solutions is not of interest, but
most important is to determine if the system is stable or not after a given
contingency. By using the equal area criterion that will be derived in this
section, it is possible with rather simple calculations to investigate the sta-
bility of the system in Figure 11.1 for different disturbances. The equal area
criterion can be extended to more complicated systems, so that preliminary
stability analyses could be done very fast.
The swing equation of the system in Figure 11.1 can be written as
d2 θ ω0
= (Pm − Pe,max sin θ) (11.10)
dt2 2H
or
d2 θ ω0
2
= Pa (11.11)
dt 2H
with the accelerating power Pa defined by
Pa = Pm − Pe,max sin θ (11.12)
As the interest here is focused on the dynamics during the first swing, the
damping can be neglected, i.e. D = 0. From the Figures 11.5 to 11.8 it can
be concluded that a necessary, but not sufficient, condition for stability is
that there exists a moment in time tm during the swing such that θ̇(tm ) = 0.
The corresponding angle is θm . Some formal manipulations with eq. (11.11)
together with the condition θ̇(tm ) = 0 will provide a stability criterion. If
eq. (11.11) is multiplied with θ̇ one gets
d2 θ ω0
θ̇ 2
= Pa θ̇ (11.13)
dt 2H
which can be written as
 2  θ 
1 d dθ d ω0
Z
= Pa dθ′ (11.14)
2 dt dt dt 2H θi

that can be integrated to give

s
θ
dθ ω0
Z
= Pa dθ′ + C (11.15)
dt H θi

with C being a constant of integration, which is 0, since θ̇ = 0 when θ = θi .

(θi is the pre-fault rotor angle.) Thus a necessary condition for stability is
that there is an angle θm such that
ω0 θm
Z
Pa dθ′ = 0 (11.16)
H θi
122 11. Power Swings in a Simple System

or Z θm
Pa dθ′ = 0 (11.17)
θi

The stability criterion can thus be formulated as:

The system is stable if there exists an angle θm such that the area below the
accelerating power Pa in the θ–P (θ) diagram between θi and θm vanishes.

In most cases of practical interest the dynamics could be divided into two
different phases: A first phase when the rotor is accelerated, and a second
phase when it is decelerated. Assume that the rotor is accelerated up to
θ = θc and that it is decelerated when θc ≤ θ ≤ θm . Then two different
areas in the θ–P (θ) plane can be defined as
Z θc
Aa = (Pm − Pe (θ′ ))dθ′ (11.18)
θi
Z θm
Ar = (Pe (θ′ ) − Pm )dθ′ (11.19)
θc

Indices a and r denote here accelerating and decelerating (retarding) areas,

respectively. The angle θc is most conveniently chosen as the angle when
the fault is cleared. The stability criterion can now be formulated as

The system is stable if there exists an angle θm such that the areas Aa and
Ar are equal, i.e. Aa = Ar .

This latter formulation gave rise to the name equal area criterion for
this stability criterion. This is illustrated in Figure 11.9. Note that Pe is
reduced to zero during the fault, hence the shape of the area Aa .
The advantage of the equal area criterion is that the stability of a system
can be investigated without any excessive computational efforts. But the
price paid is that the time t was eliminated from the equations. Since the
time does not appear explicitly in the equations, actions to improve the
stability must be formulated in the angle space, which is usually not so
practical. It is of course of more interest to have these actions formulated in
terms of times, e.g. critical fault clearing times. By the equal area criterion
critical fault clearing angles could be formulated, and these must then be
transformed to critical fault clearing times. Sometimes this can be done
easily, sometimes it is not straightforward.
In the second formulation of the equal area criterion, an angle θc was
introduced. This angle defines for which values that Pa should be calculated
as positive and negative, respectively. The choice of θc has no influence on
the result from the equal area criterion, which is obvious from the first
formulation. When applying the equal area criterion in a practical case it is
11.3. Lyapunov Stability Criterion 123

Pe, max sin θ

P(θ)
e
Ar
Pm0

Aa

0
θi θc θm π
0

Figure 11.9. Application of the equal area criterion after a disturbance.

natural to chose θc as the angle when the fault is cleared. At this angle the
topology of the system is often changed, and thus the expression for Pe (θ),
resulting in different expressions for Pe in the computation of Aa and Ar .
Different stability related problems can be solved by using the equal
area criterion. One common application is the calculation of how fast a
fault must be cleared to ensure stability of the system. The maximum fault
clearing time for which the system remains stable is called critical fault
clearing time, as mentioned above, and the corresponding angle is called
critical fault clearing angle. The equal area criterion can also be used to
calculate the maximum power that can be transmitted for a given fault
scenario. A third application is to determine if a system is stable or not
for given data and disturbances. It is strongly recommended that problems
from the problem set are solved to get an understanding of the application
of the equal area criterion.

11.3 Lyapunov Stability Criterion

The theory of the Russian mathematician A. Lyapunov (1857 -1918) offers
powerful tools for stability analysis of dynamical systems. In this theory the
following theorem, first formulated in 1899, forms the basis.
124 11. Power Swings in a Simple System

Theorem. Consider the dynamical system ẋ = f (x), where x0 is an equi-

librium point. If in a region around the point x0 there exists a real valued
and continuously differentiable function V such that

1. V(x) > 0 and V(x0 ) = 0,

2. V̇(x) = grad(V(x)) · f (x) ≤ 0,

then x0 is a stable equilibrium point. If V̇(x) < 0, then x0 is asymptotically

stable.

The existence of the function V is also a necessary condition for stability.

A problem with Lyapunov’s theorem is that it does not give any informa-
tion on how to find the function V. It is obvious from the theorem that
V can be regarded as a form of energy, and the theorem says that if the
system is stable, it will stay in the region around x = x0 where V is positive
definite provided the “energy” introduced to the system by a disturbance is
sufficiently small.
For the system studied above the following function can be shown to
fulfill the requirements of Lyapunov’s theorem

H 2
V(ω, θ) = ω + Pm0 (θ0 − θ) + Pe,max (cos θ0 − cos θ) = Vk + Vp (11.20)
ω0
with
H 2
Vk = ω (11.21)
ω0
and
Vp = Pm0 (θ0 − θ) + Pe,max (cos θ0 − cos θ) (11.22)
Vk can here be interpreted as kinetic energy and Vp as potential energy.
It is straightforward to show that V̇ = −Dω 2 by using eq. (11.6), and
consequently the second requirement of the above theorem is fulfilled if D ≥
0. It is clear that V is positive definite in ω. In Figure 11.10 the potential
energy, Vp is shown for a system with Pm0 = 0.3 and Pe,max = 1 and it
can be seen that it is positive definite in a neighbourhood of the equilibrium
point θ0 = arcsin(Pm0 /Pe,max ). Thus, if the energy injected into the system
during the fault will be less than the potential energy at the local maximum
of Vp , the system will be stable. It can be shown that the above Lyapunov
function, V, gives the same stability criterion as the equal area criterion.

11.4 Small Signal Analysis

The equal area criterion offers a tool to make stability analyses of the system
in Figure 11.1 by rather simple computations. For this system it is also
possible to do other types of analyses that do not require any more extensive
11.4. Small Signal Analysis 125

Potential energy part of Lyapunov function

1.8

1.6

1.4

1.2

1
Vp

0.8

0.6

0.4

0.2

−0.2
−100 −50 0 50 100 150 200
theta (degrees)

Figure 11.10. The function Vp of eq. (11.22).

computations. One method often used to analyse non-linear systems is to

linearise the system and study this latter system. Linear analysis is usually
not so computationally demanding, and there exist many different tools for
linear analysis. If the non-linearities of the system are “small”, it is clear
that the linear system is a good approximation of the non-linear one. But
it can be shown that for non-linear systems in general, provided the system
equations are locally differentiable, the linear system would give valuable
information about the non-linear system, e.g. regarding its stability. Small
signal, or linear, analysis will now be applied to the system in Figure 11.1.
The system in Figure 11.1 is described by
 " #
x2

ẋ1
= ω0 (11.23)
ẋ2 (Pm0 − Pe,max sin x1 − Dx2 )
2H
with
[x1 x2 ]T = [θ ω]T (11.24)
It is now assumed that the system is in an equilibrium point

[x10 x20 ]T = [θ0 0]T (11.25)

126 11. Power Swings in a Simple System

K
M

Figure 11.11. Mass and spring in analogy with the system in Figure 11.1.

The linearised system describes the behaviour of small oscillations around

the equilibrium point eq. (11.25). A first order Taylor expansion of eq. (11.23)
about the equilibrium point gives
 " #
0 1
 
∆ẋ1 ∆x1
= ω 0 ω 0 (11.26)
∆ẋ2 − Pe,max cos x10 − D ∆x2
2H 2H
From eq. (11.26) the frequency of small oscillations around the equilibrium
point can be calculated and they are given by the eigenvalues of the 2 × 2
matrix, the Jacobian matrix, in the right hand side of eq. (11.26).
First it is assumed that there is no damping in the system, i.e. D = 0.
The eigenvalues of eq. (11.26) are then purely imaginary, which corresponds
to (undamped) oscillations if 0 ≤ θ0 ≤ π/2,
r
ω0
λ=± − Pe,max cos θ0 (11.27)
2H

and the frequency of oscillation is given by

r
ω0
ωp = Pe,max cos θ0 (11.28)
2H

Eq. (11.28) is similar to the equation describing the motion for a mass con-
nected to a wall with a spring according to Figure 11.11. The frequency of
oscillation for this system is given by
r
K
ωKM = (11.29)
M
By comparing eq. (11.28) with eq. (11.29) it is seen that the spring
constant K in the mechanical system corresponds to Pe,max cos θ0 , which
sometimes is called the stiffness of the electrical system or its synchronising
power. Furthermore, M corresponds to 2H/ω0 .
From eq. (11.27) it is also seen that if π/2 < θ0 < π, the eigenvalues will
be real, and one of them will be positive corresponding to an unstable mode.
This was earlier also concluded by the qualitative discussion in section 11.1.1,
and this is now verified by the mathematical analysis done here.
11.5. Methods to Improve System Stability 127

It was pointed out earlier that the damping is not significant for large
disturbances in the system. For large disturbances it is the ability of the
system to stay within the region of attraction of the equilibrium point that
is important. This is depending on the stiffness of the system. The damping
is more important for how the system settles down to the equilibrium point
when it has survived the first swing. The linearised system could thus be
used for studies of the damping in power systems. For the simple system
studied here the eigenvalues with D 6= 0 become
s  ′ 2
D ′ D
λ=− ± j (ωp )2 − (11.30)
2 2

with ωp given by eq. (11.28) and

ω0
D′ = D (11.31)
2H
As the damping in a power system usually is quite small, eq. (11.30) can be
approximated by
D′
λ≈− ± jωp (11.32)
2
The frequency of oscillations can also be approximated in this case with good
accuracy by eq. (11.28), and it is thus determined mainly by the stiffness of
the system and the inertia constant of the machine.

11.5 Methods to Improve System Stability

It is obvious that it is desirable that a power system can withstand as many
disturbances as possible without becoming unstable. However, it is realised
that it is not possible to design a system that can cope with all conceivable
contingencies, so one has to restrict the considered disturbances in the de-
sign of the system. Usually, one considers only the most frequent faults in
the system. It is also important to consider the consequence of a fault or
disturbance when designing a system. Generally it can be said that the more
investments that can be made in a system, the more robust the system can
be made. One is thus faced with a technical-economical optimisation prob-
lem that is very complex. Furthermore, there are a number of parameters
that are both very important in the optimisation process and at the same
time very hard to quantify. Such a parameter is the value that consumers
put on uninterruptable supply of electric power of good quality. This opti-
misation process has gained a lot of interest during the last few years due to
the liberalisation (de-regulation) of the power market taking place all over
the world. Theoretically the optimum value of reliability of power supply is
when the marginal increase in the value perceived by the customers is equal
to the cost of the investment made to achieve this increase in reliability.
128 11. Power Swings in a Simple System

Practically it is of course impossible to deduce this optimum, and one has

to judge by different methods if the consumers are satisfied with the relia-
bility of the system, and if not if they are willing to pay more to improve
it.
To have some more clear and practical rules for system design a number
of deterministic rules have been used over the years. One such rule is the
so called (N − 1)-rule, which says that a system should be able to supply
customer loads with any component, generator, line, etc., out of service.
The disconnection of the component should preceded by a fault and the
transients triggered by this fault should be considered. A more conservative
approach would be to apply an (N − 2)-criterion, which would result in a
more reliable but more expensive system.
Modern methods apply probabilistic tools. In many of these approaches
one considers both the probability of a fault and its consequences. By as-
signing a measure, i.e. a number, for this combination of fault probability
and consequence a risk based security analysis can be made. Some of these
aspects are also discussed in chapter 14.
This section will be concluded by a discussion of different ways to improve
the angular stability in power systems. The stabilising effect of the different
methods can be verified by studying Figures 11.3 and 11.9. Which method to
apply in a given situation depends on a large number of parameters and must
be determined from case to case, usually after extensive studies. In some
cases, one or several methods can be excluded, e.g. the construction of new
lines since the needed permission cannot be granted from the authorities.
The most common methods to increase system angular stability are:

• Increase of the inertia constant of the generators. This makes the

rotors more difficult to accelerate in connection with faults and the
risk for losing synchronism is reduced. In most cases this is a very
expensive means and only in special cases it can be applied, e.g. by
installing a flywheel on a small hydro unit.

• Increase of system voltage. This increases Pe,max and for given power
Pm the stability margin is increased.

• Reduction of the transfer reactance Xe . This will also increase Pe,max

as in the previous case. This can be achieved by constructing parallel
lines, or by installing series capacitors on existing lines or new lines.
By installing series capacitors the effective reactance of the line is
reduced. This method has been used extensively over the years, e.g.
in the Swedish system.

• Installation of fast protections and fast breakers. In this way the time
with a fault connected can be reduced and thereby the time during
which the generator rotors are accelerated. The ability for the system
11.5. Methods to Improve System Stability 129

to decelerate the rotor swings is increased. Another way is to use

automatic re-closure after the fault is cleared, see Appendix A.

• Implementation of fast valving in steam turbines. By fast control of

the mechanical power during and after a fault, the acceleration of the
rotors can be decreased. It cannot be applied to nuclear power plants
by security reasons. The method has not been used to any larger
extent, since it is claimed to impose large thermal and mechanical
stresses on valves, turbines, etc.

• Installation of braking resistors. These are shunt resistors that are

connected by breakers fast after a fault close to a critical generator.
The electric load of the critical machine increases and the risk for
losing synchronism is reduced.

• Stability control of controllable devices such as High Voltage Direct

Current (HVDC), controllable series capacitors, controllable reactive
shunt compensation (SVC), etc. These devices are usually too ex-
pensive to install just for stabilising the system, but when they are
installed, the cost to use their controllability for stabilisation is usu-
ally marginal.
130 11. Power Swings in a Simple System
12
Oscillations in Multi-Machine Systems

This chapter contains a very brief overview of modelling of multi-machine

power systems. Many of the issues mentioned here will be dealt with in the
advanced course on power system analysis.

12.1 Simplified Model

The simplified model considered in the previous section has of course a
limited application to realistic systems. The analysis of such a simple model
is primary motivated by the insight it gives, and by that the model can be
used for simplified, preliminary calculations. In this section we will expand
the model to a system with more than one machine. The same assumptions
for modelling are established, i.e.:

• The damping is neglected.

• The mechanical power is constant.

• The synchronous machines are modelled as constant voltage sources

behind the transient reactance. The phase angle for the voltage source
coincides with the rotor angle.

• Loads are represented by impedances (admittances).

A model based on these assumptions is usually referred to as the classical

model for electro-mechanical oscillations.
Since the loads are represented by impedances the load buses can be elim-
inated and included in the node-admittance matrix, and the power which
flows out from the generators can be written as a function of the voltage
magnitudes and phase angles of the different machines.

Pei = Pei (E, θ) (12.1)

where E is a vector of voltages, of the N generators, and θ is a vector of

their rotor angles
 T
θ = θ1 θ2 . . . θN (12.2)

131
132 12. Oscillations in Multi-Machine Systems

and analogously for E. The swing equation for machine i can now be written
as
  " ωi
#
θ̇i ω0
= (12.3)
ω̇i (Pmi − Pei )
2Hi
To get a compact form of this system of equations we introduce
 T
x = θ1 ω1 . . . θN ωN (12.4)

and the N coupled equations in (12.3) can be written as

ẋ = f (x) (12.5)

where f is a vector valued function containing the swing equations for the
different machines. This system of swing equations is not easily analysed in
the general case. In most cases numerical methods must be used for studies
of the solutions. It can also be done by studying the linearised system
according to the method in section 11.4. It is found that the system can
be compared to a mechanical system with N masses connected with springs
and with spring constants determined by the stiffnesses of the lines. The
eigenvalues and eigenvectors of the Jacobian matrix define the swing modes
of the system, which often give a very good picture of how the real system
behaves in the case of disturbances. Actual oscillations in the system often
can be approximated with good accuracy with linear combinations of these
swing modes. Another way to analyse the system is by solving the non-linear
equations with numerical integration (time-simulations).

12.2 Detailed Model

A number of simplifications were earlier introduced to facilitate the analysis
of the equations by utilising well-known mathematical methods. Unfortu-
nately, the used model has some important shortcomings which makes it less
appropriate for more accurate studies of the dynamics of the power system.
The most serious shortcomings are:

• The load model is too simple. A large part of the loads consists of
asynchronous machines and other drives and their characteristics do
not follow that of an impedance at voltage oscillations. In principle,
the loads consume as much power as the generators produce. Therefore
it is equally motivated to develop realistic load models as to make
detailed models of the synchronous machines.
• The dynamics of the synchronous machines is too simple. Here it is
only determined by two states, θi and ωi , while a realistic model needs
more states for a satisfactory description. Different fluxes in the syn-
chronous machines will vary with time and different regulators, e.g. for
12.2. Detailed Model 133

voltage and frequency, will have influence on the process. Saturation

phenomena in generators can also have influence.

• Modern controllable equipment like static compensators (SVC) and

high-voltage direct current transmissions (HVDC), will in many cases
have a great influence on the stability of the system. Therefore these
should be modelled in an adequate way.

The shortcomings originate essentially from a simplified modelling of

important components in the system. A more detailed description gives
naturally a better description, but the overview and handiness of the prob-
lem are at the same time lost. More complicated models imply that more
state variables must be introduced. In the classical model every generator
is modelled by two states, θi and ωi , while a more detailed model demands
more states. To model exciters, voltage regulators, and turbine regulations,
more states also need to be introduced. Components like SVC and HVDC,
with associated control equipment generally also require dynamic modelling,
which imply that new state variables must be introduced. For more real-
istic load models, the load buses cannot be eliminated, as was done in the
classical model. This implies that the equations for loads will formulate a
constraint that the state variables must fulfill. The above outlined general
model for electro-mechanical oscillations has the following properties, which
are worthwhile to comment on:

• It is differential-algebraic. The classical model resulted in a pure dif-

ferential equation. Differential-algebraic systems are much more com-
plicated to analyse than systems which can be modelled by a ”pure”
system of differential equations.

• The system is non-linear.

• The system is non-time-invariant, i.e. the time occurs explicitly in the

equations. This depends on events in the system, e.g. disturbances
and connections, but even on protective actions of different kind.

For realistic power systems the number of state variables is very large.
As an example, the Nordic system contains hundreds of generators and the
total number of states at simulations can be up to several thousands. The
used analysis methods are based either on the linearised system or on time-
simulations of the complete model.
134 12. Oscillations in Multi-Machine Systems
13
Voltage Stability

Voltage stability has been a limiting factor in many power systems during the
last years. Due to a gradual increase in load the reactive power balance can
be critical at peak load conditions. This chapter summarises the basic con-
cepts and provides basic analyses for two simple test systems. A number of
stability indicators are introduced and applied to one of the test systems. The
importance of load characteristics and load dynamics are demonstrated.

I N THE PREVIOUS chapters it was shown that an active power imbal-

ance could give rise to rotor oscillations, which can cause instability if the
disturbances are large enough. If all generators in a system rotate with the
same electrical angular velocity (in average over a certain time period) the
system is said to be synchronously stable. (Sometimes it is said to be in
angle stability). This type of stability is strongly connected to the active
power balance in the system. As has been seen from the static analysis the
reactive power and the voltage are strongly coupled to each other. As a rule
of thumb it can be said that if there is net production of reactive power
in a node, the voltage is high in the node, while a deficit of reactive power
implies that the voltage is low. It could now be asked whether the interplay
between Q and U can also cause instabilities similar to the angular insta-
bility discussed earlier. It turns out that this is the case. The underlying
mechanisms for this voltage instability will be dealt with in this chapter
together with some methods of analysis.

13.1 Mechanisms of Voltage Instability

In chapter 8 voltage instabilities were classified into slow and fast ones. Even
if the physical mechanism is the same, i.e. a lack of reactive power, the
power system contingencies and power system structures causing them are
different. The final break down, collapse, in the system could be quite fast
also in a slow voltage instability, and the actual instability can sometimes
be disguised, which makes this kind of instability troublesome to detect in
a real system. Consequently it can be hard to get reliable indicators to
activate preventive actions if the time needed is available.

135
136 13. Voltage Stability

13.1.1 Long Term Voltage Instability

The time scale of the long term, or slow, voltage instability could be tens
of seconds to hours. It develops from a gradual lack of reactive power at a
node or in a part of the system. Even if the voltage stability phenomenon is
closely coupled to the reactive power balance the active power plays a very
important role indirectly. Many loads have an inductive power factor that
is not unity, and thus an increase in active power will cause an increase in
reactive power consumption. Furthermore, the reactive losses in transmis-
sion lines are proportional to the square of the active power transmitted, so
heavily loaded lines will influence the reactive power balance significantly. A
very deceptive role is sometimes played by synchronous machines and other
sources of reactive power. These devices can often supply more reactive
power than their steady state rating for a limited time, sometimes up to
several minutes, and during this period of time the reactive power balance
and voltage could be kept at acceptable values. However, when protections
in these devices act to avoid that the equipment is damaged, the reactive
power output is drastically reduced with a very fast voltage collapse as a
result.
The contributing factors to a slow voltage collapse are thus:

• Load increase

• Load recovery after faults

• Reactive losses in line due to high power transfer (This could be caused
by line outages due to contingencies.)

• Loss of reactive supply (Capacitors, synchronous machines, etc.)

For the second bullet above, the control and dynamics of load tap-
changers are very important. A voltage reduction in the high voltage system
will propagate to the lower voltage levels. This will usually result in a re-
duction of the load powers. However, load tap-changers on the transformers
will try to restore the voltage on the low voltage side and thereby increasing
the load power. This has in most cases a de-stabilising effect on the voltage
stability.

13.1.2 Short Term Voltage Instability

For the short term, or fast, voltage instability the actual voltage collapse
comes very soon after the disturbance, just a few seconds or less. In these
instabilities, loads or other devices with special reactive power characteristics
are often involved. Such loads are e.g. induction motors that just after fault
consume a lot of reactive power. Other loads that could be very detrimental
in this context are some types of power electronics based equipment. As an
13.2. Simple Systems for Analysis of Voltage Stability 137

Ze
Infinite bus UlÐ-f

Load
UNÐ0 Pl + jQl

Figure 13.1. Simple system for analysis of voltage stability.

Ðf
example could a line commutated HVDC converter could be de-stabilising
in a weak system, and if not properly accounted for in the system design it
could lead to system collapse for critical contingencies.
Induction motors areÐtroublesome also in the slow voltage instability.
This is particularly the case in countries where the peak load contains a
large fraction of air conditioning, which is driven by induction motors. If
the voltage gets low, these could stall and hence imply a higher reactive
power consumption.

13.2 Simple Systems for Analysis of Voltage Stability

The simple model in Figure 11.1 turned out to include all essential prop-
erties needed to get an understanding of electro-mechanical oscillations in
a power system and of the physical mechanisms that govern the dynamic
process. A corresponding simple system for voltage stability is shown in
Figure 13.1, which shows how a load is supplied via a line. It shall be
emphasised that a detailed analysis of voltage stability in a power system
is considerably more complicated than what shall be shown here, but the
fundamental mechanisms can be illustrated by the simple system in Figure
13.1.
If it is assumed that the line is lossless, i.e. Ze = jXe , the relations
below hold:
Ul UN
Pl = sin φ (13.1)
Xe
Ul UN cos φ − Ul 2
Ql = (13.2)
Xe
The angle φ is not of interest and can be eliminated by moving the non
cosine term to the left hand side in (13.2), thereafter squaring both sides of
the new equation and of (13.1). After adding the new equations one gets:
2 
Ul2 Ul UN 2
 
2
Pl + Ql + = (13.3)
Xe Xe
138 13. Voltage Stability

Nose Curve
1.2

0.8
P
l,max
Load Voltage (Ul)

0.6

0.4

0.2

0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Load Power (P )
l

Figure 13.2. Relation between active power and voltage magnitude

in the load node. The source voltage UN is chosen so that nominal
voltage gives nominal load power. (UN = 1.044, Xe = 0.3 p.u., and
power factor of the load = 1.)

It is now assumed that the reactive load varies in a given way with Pl and
Ul , e.g. so that the load has constant power factor:
Ql
=k (13.4)
Pl
If the relation between Pl and Ul is plotted in a diagram, Figure 13.2 is
obtained, where it is assumed that k in eq. (13.4) is zero, i.e. the power
factor = 1.
The type of diagram shown in Figure 13.2 is often called a “nose curve”
by obvious reasons, or a PU-curve. From this figure three observations can
be made concerning the number of (theoretically) possible operating points:

1. For Pl < Pl,max there are two solutions.

2. There is exactly one solution for Pl = Pl,max .

13.2. Simple Systems for Analysis of Voltage Stability 139

3. There is no solution for Pl > Pl,max

These observations will now be discussed briefly. For values on Pl less than
Pl,max it was concluded that there are two solutions. Physically the upper
solution corresponds to “high” voltage and “low” current, while the lower
one corresponds to “low” voltage and “high” current. In the power system
the upper solution is of interest, since it yields lower losses. Furthermore in
most situations this solution has, which will be shown in section 13.3, other
properties which makes it attractive. For Pl = Pl,max there is exactly one
solution. The value on Pl,max depends on Xe and UN , but also on the power
factor for the load. The more reactive power the load draws the less will
Pl,max be. For loads larger than Pl,max there is obviously no solution,1 , and
it is consequently clear that the system is unstable if a load greater than
Pl,max is connected.
The analysis made above was quite simplified. It was assumed that the
load power Pl could be varied as an independent parameter. However, for a
realistic physical load the load power will be a function of the load voltage
Ul , i.e. Pl = Pl (Ul ). This function is often referred to as the load (voltage)
characteristics. The behaviour of the function Pl (Ul ) depends on the time
scale considered. In a shorter time scale it is the physical processes involved
in the load device that determine the characteristics, while in a longer time
scale various control systems must also be taken into account. It turns out
that in a the shorter time scale many loads could be modelled as
  kp
Ul
Pl = Pl0 (13.5)
Ul0
for not too large variations around the operation point Ul = Ul0 , with kp
typically ≈ 1 − 2. In the longer time scale control systems strive for keeping
the load power constant for many load devices. Load modelling is very
important in voltage stability analysis, but the detailed knowledge about
load models in real systems is often very rudimentary. Much work is being
done in the field, but still more work is needed.
Another simple system that could exhibit voltage unstable behaviour is
shown in Figure 13.3. In this system there are two fairly strong systems, 1
and 2, interconnected by long transmission lines. Between the two areas 1
and 2 there is a small system connected, denoted by index m in Figure 13.3.
1
Mathematically the point Pl = Pl,max is called a bifurcation point with the active load
as parameter. At this point the two solutions coalesce to one, which will disappear when
Pl is further increased. It can be worth mentioning that it is not always true that the
maximum available load power is a bifurcation point, but this depends on the independent
parameter in the system. In the studied case Pl is the parameter, but sometimes it can
be the load impedance or another parameter, e.g. numbers of connected asynchronous
machines. In these cases maximum available load power and the bifurcation point do not
necessarily coincide. See also the discussion in section 13.3
140 13. Voltage Stability

X1 X2
~
~
U1 Um U2

Figure 13.3. Simple system equivalent for studying voltage instability.

Um
U1 U2

Figure 13.4. Voltage phasors for the system in Figure 13.3.

For simplicity it is assumed that X1 = X2 , U1 = U2 , and that no load is

consumed at the intermediate bus. The voltage Um will then be

Um = U1 cos(φ/2) (13.6)

where φ is the angle between the voltages U1 and U2 , see Figure 13.4. If
now the angle φ is increased due to an increase in the transmitted power or
in a fault increasing the reactance between areas 1 and 2, the voltage Um
will decrease. If there is no reactive power resources at this point it might
lead to a voltage collapse in this area, that could in worst case influence the
whole system.

13.3 Analysis of Voltage Stability

The simple system of Figure 13.1 will now be analysed for a few cases. In the
analysis of voltage stability a few indicators have been introduced to simplify
13.3. Analysis of Voltage Stability 141

the stability analysis. These indicators are based on small signal properties
of the system and are consequently derived from linearised models. As
concluded from the analysis of synchronous stability, linear analysis is often
very powerful, but must be used with care since it only captures the local
behaviour of the system around an operating point. To determine the global
behaviour the full non-linear model must be employed.

13.3.1 Stability Indicators

In this section voltage stability indicators of the system in Figure 13.1 will
be discussed. Similar indicators with similar properties exist also for larger
systems and these are often used in voltage stability studies of real systems.

Voltage Sensitivity Factor

The Voltage Sensitivity Factor VSF is one of the most used indicators or
indices of voltage stability. It is defined for a given node i as
∆Ui
V SFi = (13.7)
∆Qi
The physical interpretation of V SF is that it measures the change in voltage
magnitude at a given node as a consequence of a reactive power injection
in that node. Since all voltage control in a power system is based on that a
reactive power injection results in voltage increase, it is clear that a stability
criterion is
V SFi > 0 (13.8)
Furthermore, it is reasonable to require that an injection of reactive power in
any node will not decrease the voltage in any node in the system. This prop-
erty is often called voltage regularity. The V SF -indicator will be demon-
strated on the system in Figure 13.1 later on.

∆U/∆E-indicator
Another indicator that is more closely related to the voltage control in a
power system is given by ∆U/∆E, where U is the voltage magnitude of a
given bus and E is a voltage that is controllable, e.g. at the terminals of a
synchronous machine. Also in this case the requirement of stability is given
by
∆U
>0 (13.9)
∆E

∆Qg /∆Ql-criterion
This criterion relates the reactive power consumed at the load(s) to the
reactive power generated by the synchronous machines. In Figure 13.1 it
142 13. Voltage Stability

thus relates the reactive power injected at the infinite bus, here called Qg ,
to the reactive power consumed by the load Ql . The stability criterion is
∆Qg
>0 (13.10)
∆Ql
It can be shown that all the stability criteria above are equivalent in
most realistic systems. Which one to use is then dependent on which system
variables that are known and controllable in a given situation. The three
criteria have also that in common that at the transition point to instability
they tend to infinity before going negative.

13.3.2 Analysis of Simple System

To gain a better understanding of the mechanisms governing voltage stability
a few different cases will be analysed. The system in Figure 13.1 will be used
in all cases.

Influence of System Parameters

Here the influence of a few system parameters will be investigated. The nose
curve in Figure 13.2 was drawn for a lossless system, i.e. θ = 90◦ , and the
power factor of the load being equal to unity. In Figure 13.5 the power factor
has been varied, and in Figure 13.6 nose curves for a number of different
values of the impedance angle is plotted. In all the cases the magnitude of
the source voltage UN has been adjusted so that when the load is 1 p.u.
the load voltage Ul = 1 p.u. From Figure 13.5 it can be concluded that
the power factor has a significant influence on the nose curve, and thereby
on the voltage stability. If the reactive consumption of the load could be
compensated at the load point, this results in higher Pl,max but also in a
better voltage regulation around the nominal operating point.
The conclusion drawn from Figure 13.6 is that the impedance angle θ
does not influence the nose curve significantly for realistic values of high
voltage systems. A X/R ratio of 10 results in θ ≈ 84◦ and the deviation
form the curve cooresponding to θ = 90◦ is insignificant. Thus, in the
following it will be assumed that θ = 90◦ .

Influence of Load Characteristics

The voltage characteristics of the loads have a very large impact on the
voltage stability as will be shown in the following. It should be pointed
out that the actual dynamics occurring in a real system during the course
of events to be discussed here could be very complicated and that the de-
scription given here is simplified to emphasise the fundamental dynamics in
the system. The very detailed interplay between load dynamics and voltage
control might cause that the actual course of events turn out to be different.
13.3. Analysis of Voltage Stability 143

Nose Curve
1.2

0.8
1
Load Voltage (Ul)

0.6

0.4

0.2

0
0 0.5 1 1.5 2
Load Power (Pl)

Figure 13.5. Nose curves for different values of the power factor of
the load of the system in Figure 13.1. Curve 1: Power factor = 1.
Curve 2: Power factor = 0.95 (inductive). Curve 3: Power factor =
0.95 (capacitive)

Load Admittance Influence on Load Power

In Figure 13.7 the load characteristics of an impedance load, i.e. with kp = 2

in eq. (13.5) has been plotted together with two different nose curves. A
very important property of a load is the load power controllability, i.e. the
possibility to increase or decrease a load power by controlling suitable pa-
rameters. If it is assumed that the physical load in Figure 13.7 is a heater,
this control is achieved by changing the resistance (conductance) of the load.
Assuming that the voltage is constant the load power can then be increased
by decreasing the resistance of the load, i.e. to increase the conductance of
the load. In a more general case this is usually expressed as an increase in
load power which is achieved by increasing the load admittance. Increasing
the load admittance in a general case, when the load is more complicated
than a simple resistor as in the heater case, should be interpreted as con-
necting more load devices of the same kind at the load node. Actually this
144 13. Voltage Stability

Nose Curve
1.2

1
0.8

2
Load Voltage (Ul)

0.6

0.4

0.2

0
0 0.5 1 1.5 2
Load Power (Pl)

Figure 13.6. Nose curves for different values of the impedance angle
θ of Ze of the system in Figure 13.1. Curve 1: θ = 90◦ . Curve 2:
θ = 80◦ . Curve 3: θ = 70◦

has formed the basis for one definition of voltage stability:

A system is said to be voltage stable if an increase in load admittance results

in an increase in load power.

Let us now revert to Figure 13.7. If the nose curve of the system is according
to curve 1 it is seen that an increase in load admittance, dashed load curve,
gives an increase in load power. Correspondingly a decrease in load admit-
tance, dotted curve, for the operating point on nose curve 1, gives a decrease
in load power. Thus the system is voltage stable according to the definition
above for the undisturbed (nominal system). If a disturbance occurs increas-
ing the line impedance drastically, the nose curve will be according to curve
2 in Figure 13.7. The new operating point will be the intersection between
the load curve and the nose curve resulting a lower load power and a lower
voltage. If there is a controller trying to restore the load power to nominal
it will increase the load admittance (dashed load curve), which will result
in a voltage decrease and a load power decrease. This could be explained
in the following way. The load admittance increase results in an increase
13.3. Analysis of Voltage Stability 145

Nose Curve and Load Characteristics

1.2

0.8
Load Voltage (Ul)

0.6

0.4

0.2

0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Load Power (Pl)

Figure 13.7. Nose curve for undisturbed system, curve 1, and for a
system after a contingency resulting in increase of Ze , curve 2. Load
characteristics for impedance load at nominal operating point, solid
curve. Load curve with 10% increase in load admittance, dashed curve.
Load curve with 10% reduction in load curve, dotted curve.

of the load current. This current increase is however so large that the load
voltage drops so much that the load power decreases. In the nominal oper-
ating point of nose curve 1, the voltage drop due to the current increase is
however so small that the net effect is an increase in the load power. The
situation is clearly unstable for nose curve 2, and the system will collapse.
In Figure 13.7 the voltage after the disturbance, nose curve 2, is very low
and in a real system protections would have been activated. It should also
be noted that since Pl,max of nose curve 2 is much lower than the nominal
power, the load power could never recover to the pre-fault value. Generally
it can be seen that an increase in load admittance will give an increase in
load power if
dPld dPl
> (13.11)
dU dU
where Pld refers to the load device, i.e. the load characteristics, and Pl to
the load power of the nose curve.
146 13. Voltage Stability

VSF Criterion
The V SF criterion was introduced above and that will now be analysed
for different load characteristics. This criterion is closely related to how
the voltage control will act in the system. It is rather straightforward to
calculate V SF for different load characteristics and operating points on the
nose curve. From such an analysis the following can be concluded:

• For constant power loads, i.e. kp = 0 in eq. (13.5), operating points

on the upper half of the nose curve are stable, i.e. V SF > 0, while
those on the lower part are unstable or V SF < 0.

• For impedance loads V SF > 0 for all operating points, indicating

voltage regularity.

• For 0 < kp < 2 operating points on the upper part of the nose curve
are stable. In addition a part of the lower part is also stable ranging
from Pl,max to lower Pl values. This interval increases with kp with
boundaries according to the two previous conclusions.

Since in a longer time scale most loads could be approximated by con-

stant power loads, permanent operation is feasible only at the upper part
of the nose curve. During transients operation on the lower part could be
acceptable.

Load Dynamics
The discussion in this paragraph relates to the discussion concerning load
admittance, but deals with a somewhat different disturbance. The system
in Figure 13.1 is considered again. It is now assumed that a voltage increase
of 5% occurs in the sending end system, i.e. UN is increased by 5%. In
Figure 13.8 the nose curves and load characteristics corresponding to an
impedance load are depicted. The load is controlled in such a way that the
load power should be constant, which e.g. could be the case for a thermostat
controlled heater. The voltage increase resulted in a load power increase and
consequently the load power control orders a decrease in load admittance.
This decrease in load admittance results in the dashed load characteristics
in Figure 13.8, and the system can settle to a new operating point at the
desired load power, i.e. 1 p.u., but at a slightly higher voltage. An analysis
of the behaviour of the system shows that, if the sending end voltage is
slightly decreased, the system will settle at 1 p.u. load power at a slightly
lower voltage.
Now assume that the system is operating on the lower part of the nose
curve, see Figure 13.9. According to the V SF criterion this operating point
is stable. The same disturbance as in the previous case is considered. From
Figures 13.8 and 13.9 it can be concluded that both systems are stable
13.3. Analysis of Voltage Stability 147

Nose Curve and Load Characteristics

1.4

1.2

2
Load Voltage (Ul)

0.8
1

0.6

0.4

0.2

0
0 0.5 1 1.5
Load Power (Pl)

Figure 13.8. Nose curve for undisturbed system, curve 1, and for
a system where the sending end voltage has been increased by 5%,
curve 2. Solid load characteristics corresponds to load admittance
before voltage increase. Dashed load characteristics corresponds to
decreased load admittance to give load power = 1 p.u. for system after
disturbance.

according to the ∆U/∆E criterion since an increase in the sending voltage

causes an increase in the load voltage. It is seen that voltage increase in
Figures 13.9 implies a load power increase. The same load power controller
as in the previous case would order a decrease in the load admittance to
decrease the load power. Such a load admittance decrease is represented by
the dashed load characteristics in Figures 13.9. But, the load admittance
decrease results in a load power increase. If there are no restrictions on
the action of the controller it will continue to decrease the load admittance
until the point “A” of Figure 13.9 is reached. A similar analysis shows that
a voltage decrease in the sending end system will result in that the load
admittance will increase until (theoretically) the point (0,0) is reached. It
is thus clear that the initial operating point of Figures 13.9 is unstable.
148 13. Voltage Stability

Nose Curve and Load Characteristics

1.4

1.2

1
2

1
Load Voltage (Ul)

0.8

0.6

0.4

0.2

0
0 0.5 1 1.5
Load Power (Pl)

Figure 13.9. Nose curve for undisturbed system, curve 1, and for a
system where the sending end voltage has been increased by 5%, curve
2. Solid load characteristics corresponds to load admittance before
voltage increase. Dashed load characteristics corresponds to decreased
load admittance.

QU curves
Finally another method sometimes used to analyse voltage stability will be
demonstrated. In the system of Figure 13.1 one can calculate the amount of
reactive power that must be provided to the network for a given load power
for different load voltages. If this is done for the system of Figure 13.1
assuming Pl = 1 p.u., a curve according to Figure 13.10 is obtained. The
same data as in Figure 13.2 have been used. Since Figure 13.2 was calculated
with unity power factor of the load, the load voltages corresponding to
Q = 0 of Figure 13.10 should correspond to the voltages giving Pl = 1 p.u.
in Figure 13.2. This is easily verified. Furthermore, the V SF criterion
for stability says that an reactive power injection should result in a higher
voltage. In the QU curve this corresponds to a positive derivative, which
are for load voltages higher than ≈ 0.6 p.u.
13.3. Analysis of Voltage Stability 149

QU Curve

0.3

0.2

0.1
Reactive Power Demand of Network

−0.1

−0.2

−0.3

−0.4

−0.5

−0.6

−0.7

−0.8
0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1.1
Load Voltage (Ul)

Figure 13.10. QU curve for the system of Figure 13.1. (UN = 1.044,
Xe = 0.3 p.u., Pl = 1.0 p.u.)
150 13. Voltage Stability
14
Control of Electric Power Systems

This chapter gives an introduction to control of power systems. Both the

faster distributed controls and slower centralised controls are described.

A POWER SYSTEM is planned, designed and built to supply the con-

sumers with electrical energy considering:

• Economy

• Quality

• Supply security (Reliability)

• Environmental impact

The last point has become more and more important during the latest
years. This can be seen from the fact that most of the larger companies
nowadays have special departments responsible for that the companies fol-
low laws and regulations within the environment area. The three first points
above constitute the basis of the optimisation that the power companies
must make regarding their investments and daily operation. It is evident
that economical considerations must be regarded when deciding issues about
quality and supply security for the system. The higher quality and supply
security that is required, the more expensive the electrical energy becomes.
Methods based on mathematical analysis and decision making theory (oper-
ations research) have been worked out to handle this optimisation problem,
which contains several considerations/adjustments that are hard to formu-
late in a stringent mathematical form. As cheap and safe access to electrical
energy is of great importance for almost all activities in a modern society,
it is not surprising that the decisions makers in the power industry were
pioneers to utilise advanced optimisation methods and other analysis tools
for expansion and investment planning.
In this compendium we will limit ourselves to discussing how a power
system can be controlled to fulfill the constrains regarding quality. A very
short introduction is also given about the overall control in a power system,
which besides its aim to fulfill the quality even plays an important role in
the supply security.

151
152 14. Control of Electric Power Systems

Three important factors that define power quality are:

• Frequency variations

• Voltage variations

• Waveform of voltage and current

The two first factors will be dealt with in this course, but the last point
is getting greater importance because more nonlinear components are con-
nected to the electrical system, e.g. power electronics. These can give rise
to harmonics.
It has been shown earlier that the frequency is a very good indicator
on the active power balance in a power system. The frequency is constant
when the same amount of electrical power is produced as consumed by the
loads, including system losses. If this is not the case frequency changes will
occur. It can also be noted that the frequency is the same in the whole
system at steady state. This depends on that the active power easily can be
transported in the system (that is why it has been possible to build the power
systems of today’s size). It is often said that the active power is a global
quantity, unlike the reactive power which is a local quantity, since reactive
power cannot be transmitted over longer (electrical) distances. (This is
because of that X normally is much greater than R in a power system, at
least for transmission and sub-transmission networks.)
The frequency of the system is reduced when a load increase is not
compensated for by a corresponding increase of the turbine power of the
connected generators. The power deficit decelerates the generator rotors
and consequently the frequency is reduced.
Frequency reductions also arise when production is lost, e.g. as a conse-
quence of failures in the system which lead to that protections disconnect
the failed equipment. Too large reductions of the frequency could lead to
system collapse, since a lot of equipment in the power stations, e.g. power
supply systems, do not tolerate too low frequencies. A load reduction in the
system which is not compensated for by a reduction of turbine power leads
to a frequency increase.
The permitted stationary frequency deviation in interconnected power
systems is at normal operation typically 0.2 % or 0.1 Hz. There are also
sometimes requirements on how large the difference between actual, physical
time and the time corresponding to the integrated frequency from the system
can be. This time difference is normally not permitted to be larger than 10s.
It is also important that the voltage deviations in the system is limited.
This is of importance for the connected loads, but a “good” voltage profile is
also essential for keeping the losses low and for utilising the reactive reserves
to establish a secure operation of the system. Voltage control is, as been
pointed out earlier, a more local control than the frequency control. If the
14.1. Control of Active Power and Frequency 153

Figure 14.1. Typical variation of the load in a power system. The

upper curve shows the highest hourly average values over the year and
the lower curve the lowest hourly average value.

voltage deviates from the set value in a node, the control action must be
made in this or a nearby node.

14.1 Control of Active Power and Frequency

Since the stored energy in the system is relatively small, the constant of
inertia for a typical generator is ≈ 4 s, the electrical energy must be produced
in the same moment as it is consumed by the loads. Since the load varies, a
certain power reserve must constantly be available. In addition for the daily
variation, see Figure 14.1, there are continuously spontaneous load variations
up to ≈ 2 % of the total load during a minute. The generation reserves
are generally divided into different groups according to their properties:
spinning, supplementary and back-up, see Figure 14.2. The reserves are at
operation planning divided into normal operation and disturbance reserve
after the cause of the needed reserve.
At the operation planning of the power system forecasts over the ex-
pected load are carried out continuously. Forecasted values can never ex-
154 14. Control of Electric Power Systems

MW
Operation
Unavailable Planning Operation

Total Installed Capacity

Generation
Disturbance Back-up
Reserves Reserve
Supple-
Generation = mentary
= Normal
Loads Operation
+ Reserve Spinning
Losses

Figure 14.2. The different generation reserves in a power system.

actly coincide with real values, therefore a reserve which can compensate
the difference is required. Therefore frequency variations arise, which must
be compensated for. Both power imbalances due to incorrect load forecasts
and to occasional load variations are controlled with the normal operation
reserve.
Events which can lead to utilisation of the disturbance reserve are: gen-
erator trips or line trips. Disturbances of that kind can lead to frequency
reductions and reserves must often be put into operation.
In the operation the reserves are divided after needed time for activation,
see Figure 14.2. To keep reserves is expensive and therefore it is of interest
to minimise the needed effort of the reserves for maintaining the wanted
reliability and security.

14.1.1 Spinning reserve

The spinning reserve is co-ordinated in the Nordic system and is above all
located in hydro units with turbine control, see Figure 14.3. The turbine
control is activated within some seconds if the frequency deviates from the
normal and changes then the turbine setting for the regulating generators by
changing the guide vane opening to the turbines. When the system frequency
is changed the power demand of certain loads is also changed, specially for
motors, in such a way that a frequency increase leads to increased power
consumption and a frequency decrease gives lower power consumption. This
frequency dependence of the load stabilises the frequency. Components for
heat- and light production are fairly insensitive to frequency variations.
The resulting change in active power, ∆P , at a frequency change in the
system is:
∆P = ∆Pg − ∆Pl (14.1)
where: ∆Pg = the total change in output power from the generators par-
14.1. Control of Active Power and Frequency 155

Uset
∆U f Ugen

Ufield

Turbine
P mech
~ Power System

∆P mech ∆f f

fset

Figure 14.3. Generator unit with voltage regulator and turbine reg-
ulator for frequency control.

ticipating in the frequency control, ∆Pl = the total frequency dependent

load power change. The frequency dependence of the load can often be
approximated as
∆Pl = D∆f (14.2)
where D is a constant. The value on D is typically 0 − 4 %/Hz, but large
variations occur depending on the kind of load.
The frequency control is performed so that, at least for small frequency
deviations, the total change in output power of the frequency controlling
generators are (in steady state) proportional to the frequency change, ∆f .
The relation between changes in active power and frequency deviations can
then be written
∆Pg = −R∆f (14.3)
The constant R is determined by the setting of the turbine control of the
frequency controlling generators, and is named the systems frequency regu-
lation constant or frequency droop and is normally given in MW/Hz. The
frequency regulation for a system is established so that the frequency does
not fall outside given limits at certain dimensional disturbances which can
occur, e.g. trip of the largest generator in the system.
It shall be noted that the variation of frequency, computed from the
above relations, are the steady state values which are obtained after the
frequency controllers have responded. The transient frequency deviation
during the process following a large disturbance can be considerably larger
than the stationary one, up to some Hz. In Figure 14.4 it is shown how
the frequency can vary after a disturbance in the system. As seen the
maximal frequency deviation is considerably greater than the stationary
156 14. Control of Electric Power Systems

Figure 14.4. The transient frequency response after a production

loss. The diagram is a recording from the Nordic system at a loss of a
1000 MW unit on the 24 November 1983.

one. The transient behaviour of the frequency is highly dependent on the

characteristics of the turbine regulators of the generators, and in this respect
the hydro turbines and steam turbines are totally different. Shortly it can
be said that the transient frequency deviation is considerably larger if the
frequency control is performed by hydro power. This is due to that the
hydro turbine is a non-minimum-phase system.

14.1.2 Supplementary Reserves

Supplementary reserves must also be available to cope with fast changes
in the generation plan at large forecast deviations or at permanent out-
ages when the spinning reserve has been used and must be restored. From
the remaining deviation of frequency and the knowledge about the system
frequency droop it is straightforward for the system operators to compute
how much “new” power production must be connected to restore the sys-
tem within the prescribed limits. This can either be done by connecting new
production or by increasing the turbine power at generators in operation. In
some systems this is handled manually, but in some countries, e.g. in conti-
nental Europe and USA, this is at certain power companies accomplished by
special slow automatic controllers, so-called Automatic Generation Control
(AGC). This regulation is considerably slower than frequency control and
must be designed not to interact with the frequency control in an adverse
14.2. Control of Reactive Power and Voltage 157

way.

14.1.3 Back-Up Reserves

The slow reserve shall be available within about 2 h. It is needed for security
reasons to restore the fast reserve when this has been used. The slow reserve
can consist of thermal power which is kept at stand by. This implies that
pressure and temperature is kept on a level which permits synchronisation
and load recovery within 2 hours. Co-generation back-up units for which it
is possible to increase the electrical power can also be used.

14.2 Control of Reactive Power and Voltage

14.2.1 Reactive Power Control
In steady state operation both active power balance and reactive power
balance must be maintained. The reactive power generated by synchronous
machines and capacitances must be equal to the reactive power of the loads
plus the reactive transmission losses. If the active power balance is not
kept, the frequency in the system will be influenced, while an imbalance in
reactive power will result in that the voltages in the system differ from the
desired ones.
If the power system is operated in the correct way, the voltage drops on
the lines are usually small. The voltages in the nodes of the system will
then almost be the same (flat voltage profile). In this case the transmission
system is effectively used, i.e. primarily for transmission of active power,
and not for transmission of reactive power.
As known from the Static Analysis the voltage in a system is strongly af-
fected by the reactive power flow. Consequently the voltage can be controlled
to desired values, by control of the reactive power. Increased production of
reactive power gives higher voltage nearby the production source, while an
increased consumption of reactive power gives lower voltage. Therefore it
is of great interest to study which components and devices which can be
used to regulate the reactive power in a power system. While the active
power is entirely produced in the generators of the system, there are several
sources of reactive power. In the other hand the reactive power cannot be
transported over long distances in the system, since normally X ≫ R in a
power system.
Important producers of reactive power are:

• Overexcited synchronous machines

• Capacitor banks

• The capacitance of overhead lines and cables

158 14. Control of Electric Power Systems

Important consumers of reactive power are:

• Inductive static loads

• Under-excited synchronous machines

• Induction motors

• Shunt reactors

• The inductance of overhead lines and cables

• Transformer inductances

• Line commutated static converters

For some of these the reactive power is easy to control, while for others it
is practically impossible. The reactive power of the synchronous machines is
easily controlled by means of the excitation. Switching of shunt capacitors
and reactors can also control the reactive power. If thyristors are used to
switch capacitors and/or thyristors are used to control the current through
shunt reactors, a fast and step-less control of the reactive power can be
obtained. Such a device is called SVC (Static Var Compensator).
As has been shown earlier it is most effective to compensate the reactive
power as close as possible to the reactive load. There are certain high
voltage tariffs to encourage large consumers, e.g. industries, and electrical
distributions companies to compensate their loads in an effective way. These
tariffs are generally designed so that the reactive power is only allowed to
reach a certain percentage of the active power. If this percentage is exceeded,
the consumer has to pay for the reactive power. The high voltage network
is in that way primarily used for transmission of active power.
The reactive losses of power lines and transformers depend on the size of
the reactance. In overhead-transmission lines the reactance can be slightly
reduced by the use of multiple conductors. The only possibility to radi-
cally reduce the total reactance of a transmission line is to connect a series
capacitor, see 6.5.1 in the part ”Static Analysis”.

14.2.2 Voltage Control

The following factors influence primarily the voltages in a power system:

• Terminal voltages of synchronous machines

• Impedances of lines

• Transmitted reactive power

• Turns ratio of transformers

14.2. Control of Reactive Power and Voltage 159

U1 zk τU 2 U2

N1 N2

Figure 14.5. Transformer with variable turns-ratio (tap changer).

A suitable use of these leads to the desired voltage profile.

The generators are often operated at constant voltage, by using an auto-
matic voltage regulator (AVR). The output from this controls the excitation
of the machine via the electric field exciter so that the voltage is equal to the
set value, see Figure 14.3. The voltage drop caused by the generator trans-
former is sometimes compensated totally or partly for, and the voltage can
consequently be kept constant on the high voltage side of the transformer.
Synchronous compensators are installed for voltage control. These are syn-
chronous machines without turbine or mechanical load, which can produce
and consume reactive power by controlling the excitation. Nowadays new
installations of synchronous compensators are very rare.
The impact of the impedances of the lines on the reactive power balance,
and thereby the voltage, have been analysed in the Static Analysis. These
are generally not used for control of the reactive power. Series capacitors
are generally installed to increase the active transmission capacity of a line.
From the static analysis it is also known that the reactive power trans-
mitted has a great impact on the voltage profile. Large reactive transmis-
sions cause large voltage drops, thus these should be avoided. Instead, the
production of reactive power should be as close as possible to the reactive
loads. This can be achieved by the excitation of the synchronous machines,
which have been described above. However, there are often no synchronous
machines close to the load, so the most cost-effective way is to use shunt
capacitors which are switched according to the load variations. An SVC can
be economically motivated if fast response or accuracy in the regulation is
required. Shunt reactors must sometimes be installed to limit the voltages
to reasonable levels. In networks which contain a lot of cables this is also
necessary, since the reactive generation from these is much larger than from
overhead lines. (C is larger and X is smaller.)
An important method for controlling the voltage in power systems is
by changing the turns ratio of a transformer. Certain transformers are
160 14. Control of Electric Power Systems

equipped with a number of taps on one of the windings. Voltage control can
be obtained by switching between these taps, see Figure 14.5. If switching
during operation can be made by means of tap changers, this possibility of
voltage control is very effective and useful. Normally the taps are placed
on the high voltage winding, the upper side, since then the lowest current
needs to be switched.
If N1 is the number of turns on the upper side and N2 is the number of
turns on the lower side, the turns ratio of transformer is defined as

N1
τ= (14.4)
N2
Then the relation between the voltage on the high voltage side, U1 , and on
the low voltage side, U2 , at no load is

U1
U2 = (14.5)
τ
If the voltage decreases on the high voltage side, the voltage on the lower
side can be kept constant by decreasing τ , i.e. by switching off a number of
windings on the high voltage side. When the transformer is loaded eq. (14.5)
is of course not correct, since the load current gives a voltage drop over the
leakage reactance of the transformer, zk , but the same principle can still be
applied at voltage control.
Transformers with automatic tap changer control are often used for volt-
age control in distribution networks. The voltage at the consumers can
therefore be kept fairly constant even though voltage variations occur at the
high voltage network. Time constants in these regulators are typically some
ten seconds.
Sometimes the turns ratio cannot be changed during operation, but just
manually when the transformer is off load. In this case one can only change
the voltage level in large but not control the voltage variations in the net-
work.

14.3 Supervisory Control of Electric Power Systems

The frequency- and voltage control described above are performed by local
controllers, but overall and central control is also needed to secure safe and
economical operation of the power system. This control is performed from
control centres which generally are hierarchically organised. Often there is a
national centre on the top, which supervises the national transmission sys-
tem and co-ordinates the frequency control. In certain cases this centre can
also co-ordinate the operation of different power companies, which in that
way can optimise their operation. The next level contains a number of re-
gional control centres which co-ordinate and supervise the regional networks
14.3. Supervisory Control of Electric Power Systems 161

Normal
State

Restor- Alert
ation State

System Extremis Emergency

Collapse State State

= Planned (Intended) Changes

= Unplanned Changes (Disturbances)

Figure 14.6. The different operating states of a power system.

and control the generation within that region. The next level of operation
centre has the task of controlling the operation for one or several power
stations. The information flow between these different levels is very large
and there is an extensive communication system parallel with the electrical
power system to handle this. Computers are extensively used to perform the
overall control of the power system, but also the operators make important
decisions in the control process.
The different states that a power system can be found in is often de-
scribed according to Figure 14.6. Normal operation is the state which is
desired for a system. Besides that all quantities in the system are within
the limits given by the equipment and the system, there are margins so
that the system can tolerate certain predetermined disturbances, e.g. loss
of lines and generation, without endangering the system security. When
the system is in alert operation all quantities are still within the allowed
limits, but the margins which existed for normal operation have been lost
as a consequence of some disturbance(s). In alert operation power can be
supplied as usual, and the consumers do not notice that any disturbance
has affected the system. In this state the goal for the control is to bring the
system back to normal operation through connecting equipment to establish
margins to cope with new disturbances. Emergency operation can occur if
further disturbances occur or if a large disturbance strikes the system. Some
quantities, e.g. voltages or power flows on the lines, are now outside the per-
mitted limits. Still there is no shortage of power, but fast actions must be
taken otherwise further equipment will be tripped by different protections.
If it is not succeeded to bring the system to alert operation and gradually
162 14. Control of Electric Power Systems

Operating Rating Pgen = Pload + Ploss

Margins Limits
Normal State OK OK OK
Alert State No OK OK
Emergency State No No OK
Extremis State No No No

Table 14.1. Characteristics of the different operating states of a power system.

to normal operation, there is a risk to end up in extremis operation. In this

state there is not enough power to supply the connected loads, and the fre-
quency is decreasing. Load shedding is usually needed to save the system.
First of all such load is shed that easily can be connected again without any
larger damages to the consumers, for example electrical heaters or domes-
tic loads. As a last resort, loads which demand long time for restoration,
e.g. process industry, are shed.
If this defence is not successful, the system, or often parts of the system,
will collapse, i.e. be de-energised, which is a state which must be avoided by
all means. If this state should be reached the system must as fast as possible
be brought back, often part by part, to normal operation. This process is
called system restoration. The transitions which are marked with unbroken
lines in the Figure 14.6 are initiated and performed from different control
centres. (The broken lines denote undesired disturbances.) The central
control plays an important role to get the system into normal operation.
In Table 14.1 the characteristics of the different operating states of a
power system are summarized.
Appendix A
Phase-Shifting Transformers

In three-phase transformers it is possible to arrange the windings in such way

that not only the voltage magnitudes between the primary and secondary
sides are different but also the phase angles. In this appendix a number
of winding arrangements are described that could accomplish such a phase
shift.

F OR SINGLE-PHASE TRANSFORMERS the relation between the pri-

mary and secondary side voltages is a real number, known as the turns-
ratio, during sinusoidal steady state conditions. In three phase systems the
windings can be arranged in different ways, and in some of these configu-
rations a phase-shift is also introduced between the primary and secondary
side. Consider the two three-phase transformers in Figure A.1. The trans-
former at the top has both the primary and secondary windings connected
in a Y, called a Y-Y or wye-wye connection and denoted Yy. Obviously
there is only a transformation of the magnitude of the voltages between the
two sides. In the transformer at the bottom however, the primary side is
delta connected and the secondary side is connected in Y, often denoted Dy
or ∆y. In addition a phase shift is here introduced between the primary and
secondary windings, in this case 30◦ . Obviously other multiples of 30◦ phase
shifts can also be obtained.
An obvious way to incorporate the phase-shift in the transformer model
would be to allow for a complex turns-ratio, i.e. t = a · ejϕ , as explained
in subsection 2.2.2. As also shown there, a phase shift of a transformer
has an influence on the power flow, predominantly the active power flow,
if one or more parallel paths exist. It is also shown that the magnitude of
the turns-ratio, i.e. a in the expression above, influences mostly the reactive
power flow. A phase-shifting transformer would thus be a tool to control
the active power flows on parallel paths.
However, the transformers introducing multiples of 30◦ phase shift would
in most cases be too crude for a meaningful power flow control. In realistic
systems phase-shifts up to maximum 10◦ − 20◦ or so would be very power-
ful. In addition it would be desirable to be able to control the phase-shift
depending on the loading conditions in the network. Thus a transformer
according to Figure A.2 where the ∆Ur etc. can be controlled, both in mag-

163
164 A. Phase-Shifting Transformers

R S T
r
R r
s
t

T S t s

R S T
r
s R r
t

T S
t
n

Figure A.1. Two different configurations of three-winding transformer.

nitude and in phase, would be of value. This is the so called regulating

transformer.
The regulating function of these transformers is achieved through a tap
changer that can change the turns on one or more windings and thereby
changing the turns-ratio. Depending on the design both the magnitude
and the phase of the turns-ratio could be changed. It is obvious that a
tap-changer in the transformer connection in Figure A.1 would only change
the magnitude of the turns-ratio. (It is assumed that the tap-changers on
the three phases are identical and operated identically.) Such arrangements
could thus only be used for voltage control. This is one of the main methods
to control the voltage in the distribution systems.
In order to introduce a controllable phase-shift to control the active
power flow more complicated winding configurations must be introduced.
There are several ways of achieving this, and the solution chosen depends
on a number of parameters, such as phase-shift range to be controlled, volt-
age difference between the primary and secondary sides, power rating, etc.
Phase-shifting transformers, or phase-shifters, are quite rare in comparison
to non phase-shifting transformers 1 . However, phase-shifters play an im-
1
With non phase-shifting transformers are here meant transformers where a possible
 165

DUr
UR
Ur
Ur Ur+DUr=UR

Us Us+DUs=US UT
Us
Ut Ut+DUt=UT DUt Ut
DUs
US

Figure A.2. Basic principle of a regulating transformer.

portant role particularly in highly meshed systems where power transfers

over long distances take place. Phase-shifters can then re-direct the power
flows to circuits that are not so highly loaded. Another function can be to
direct the power flow to high voltage lines from highly meshed networks at
lower voltage levels. In Europe and North America several phase-shifters
are installed for power flow control. Since the traditional tap-changers are
mechanical devices, with a typical minimum time interval between subse-
quent switchings of several tens of seconds, these controllers can only be used
for steady state power flow control. Tap-changers with thyristor switches
are today available, and with these much faster switchings can be achieved.
Thereby also dynamic power swings could be damped. For this purpose
also completely power electronics based controllable devices are developed,
so called FACTS devices (FACTS = Flexible AC Transmission Systems),
which normally offer a much higher degree of controllability.
To obtain a phase-shift, a voltage in quadrature to the phase voltage
must be added. Therefore, phase-shifters are also called quadrature trans-
formers or quadrature boosters. Figures A.3 and A.4 show two examples
of how phase-shifters can be designed. In Figure A.3 the voltage ∆U is in-
serted by a series transformer and the phase-shift is always accompanied by
a certain change in magnitude. For the transformer in Figure A.4 an almost
pure phase-shift can be obtained for small phase-shifts.

phase-shift does not influence the power flow. Transformers connected Yd are commonly
used as generator step-up transformers and as transformers feeding distribution networks.
However, since no parallel paths exists, or the parallel transformers are identical, the
phase-shift does not play any role for the power flow.
166 A. Phase-Shifting Transformers

Ur Ur+DUr DUr

Us Us+DUs
Ur a

Ut Ut+DUt
Us
Booster DUt Ut
DUs

Figure A.3. Phase-shift obtained by a series boost transformer.

DUr
Ur
Ur+DUr
aU
r

Ut
Us DUs
DUt
Us Us+DUs

Ut Ut+DUt

Figure A.4. Phase-shift obtained by a quadrature voltage.

Appendix B
Protections in Electric Power Systems

In this Appendix a brief summary of how protections are designed and how
they function is given. The very important distance protections and their
operating principles are discussed. Some special protections and system
wide protections that are of relevance for power system stability is briefly
reviewed.

D IFFERENT TYPES of protections are installed to protect the equip-

ment in an electric power system. Their task is to disconnect failed or
overloaded equipment or parts of the system to avoid unnecessary damages
on equipment and personnel. The purpose is also to limit the impact of
failures on the parts of the system that have not failed. Special types of
protection are the “system protections”. Their task is to prevent collapse
(black out) of the system or parts of the system.
An intensive development of protections based on modern information
technology is going on both regarding hardware and software. On the hard-
ware side microprocessors have been used over a long time to implement
different functions in the protections, and with the recent developments
more and more complicated functions can be implemented in a reliable way.
Powerful methods like signal processing, state estimation, and “artificial
intelligence”, are being integrated into the protections. In general the func-
tions which earlier were handled with separate relays are increasingly being
integrated with other functional units for control and supervision. Further-
more, more complicated criteria for activation of protections can be applied.
The interested reader is referred to the literature for further information.
The summary here is concentrated on general principles for protections.

B.1 Design of Protections

A protection for an electric power system comprises the following parts:

• Measurement device with current- and/or voltage transformers and

other sensors measuring the relevant quantities.
• Relay which when certain conditions are fulfilled sends signals to a
circuit breaker or another switching device. This relay was earlier a

167
168 B. Protections in Electric Power Systems

separate unit, but can in modern protections be a part of a larger unit

for protection, supervision and control.
• Circuit breakers which execute the given instruction(s) from the relay.
• Telecommunication system is mainly used at distance (line) protec-
tions to get a faster and more reliable performance.
• Power supply systems which shall secure the power supply to the pro-
tection system, even with faults in the system.
The requirements on a protection system are that they should be de-
pendable, secure, selective, sensitive, and fast.
• Dependability means that the protection should react and do its action
when a fault occurs for which it is designed to react for. To achieve
desired dependability double or even triple sets of certain parts of the
protection or of signal paths might be needed. Malfunctions can be
divided into ”not occurring” operations (which are actions that were
supposed to happen but did not) and ”unwanted” operations (which
are actions that happened although they should not have). Normally
not occurring operations are more serious malfunctions than unwanted
ones.
• Security means that the protection should not react when no fault
occurs or when a fault for which it is not intended to react occurs.
• Selectivity implies that not more than necessary pieces of equipment
and apparatuses are disconnected to isolate a fault.
• Sensitivity is needed to detect failures which cause small fault currents,
e.g. high impedance faults. This implies that the risk for misoperations
increases at “small” disturbances, e.g. at energisation of transformers,
or at high load operation but normal operation.
• The protection should react fast to secure that damages on persons
and equipment are prevented or limited.
The protections are often classified according to the object that they
protect. An example is shown in Figure B.1. If a failure occurs within an
indicated area in Figure B.1 this area should be isolated from the rest of the
network.
Many of the protections which protect separate pieces of equipment or
parts of a system which occupy a limited physical area are so called current
differential protections. These protections measure the difference between
two currents, which in normal operation should be equal, and the protection
is activated if this deviation exceeds a predetermined value. Both differences
in amplitude and phase can trigger the relay. The principle for a current
differential protection is shown in Figure B.2.
B.2. Distance Protections 169

Generator Protection

Distance Protection
~
~

Bus Bar Protection

Transformer Protection

Figure B.1. The different protection zones in a power systems.

B.2 Distance Protections

B.2.1 General Principles
So called distance protections are important protections concerning stability
and dynamics in a power system. Their task is to disconnect faulted lines
or cables. Since large parts of the power system consist physically of lines
and these are exposed to different disturbances, e.g. lightning strokes, down
falling trees etc., it is important that those faults can be isolated to minimise
the impact on the rest of the system. The most common faults are ground
(earth) faults, i.e. short circuits between two or more phases and ground
(shunt faults). Also interruptions in the lines can occur (series faults). The
operating principle of the distance protection is shown in Figure B.3.
Current and voltage are measured in both ends of the line and from these
an apparent impedance can be calculated: Z = U/I. In normal operation
this impedance varies within a certain area (large and almost resistive values
on Z), but if a fault occurs, it will drastically change. The given value
depends on where on the line the failure occurs, and from system parameters
as line data and short circuit capacity, it can be calculated where the fault
has occurred.
For each distance protection there are several protection zones defined
in the Z plane according to Figure B.4. A low value on Z implies that the
fault is close to the measurement. From line data and short circuit capacity
170 B. Protections in Electric Power Systems

Protected Object

IA + − IB
Σ
∆I

Relay Logics Signal(s) to Breaker(s)

Figure B.2. Principles of a current differential protection.

Instrument Transformers

Breaker #1 Breaker #2

U, I U, I
Telecommunication
Relay Logic Relay Logic

Figure B.3. The operating principle of a distance protection.

it can then be decided if the fault is in the protected line, within Zone 1,
or not. If that is the case, a trip order is given to the breaker at the same
station within some milliseconds, typically 10 ms, after Z has reached Zone
1. At the same time a trip order is given to the breaker in the other end of
the line. This latter trip order is not needed for isolation of the fault, if the
protection system in the other end works as it should, but this trip order
(transfer trip) increases the security in the system.
If the measured value on Z is in Zone 2 or 3, it implies that the fault is
outside the actual line. This implies that neither breaker 1 nor 2 in Figure
B.3 shall be opened. If the breakers, which according to the protection plane
should isolate the fault, are not operated by some reason, other breakers
which are further away from the fault must isolate it. These secondary
B.2. Distance Protections 171

X Zone 3

Zone 2

Zone 1

Normal Value of Z

Figure B.4. Different zones in a distance protection.

breakers will be used first after it is clear that the primary breakers have
not isolated the fault. Therefore if, Z is in Zone 2, the breaker does not get
the trip order until typically some hundred milliseconds have passed.
To coordinate and tune the settings of the protections to give a fast,
reliable, sensitive and selective protection system is a complicated and an
important task in an electric power system. In modern protection systems
different areas can be defined according to Figure B.4 with in principal ar-
bitrary geometric shapes, which facilitates the work. A plan comprising the
different areas of protections and time settings is usually called a selectivity
plan. The work to establish a selectivity plan is often very time consuming
because it should be appropriate for every feasible state of operation, i.e. for
different numbers of generators and lines connected and also at different
load levels. Often trade-offs must be made to reach acceptable results.

B.2.2 Automatic Re-Closure

When a ground fault or a short circuit occurs an ionised plasma (arc) that
carries the fault current is often formed. This arc remains ionised as long as
a the current flows through it. If the fault current is extinguished it is usually
sufficient for the plasma to cool down during some hundred milliseconds to
rebuild the isolation so that the line can be re-connected. This is used in
many systems and the line is automatically re-energised after a given time
period after fault clearing. This is the case in many systems, where failed
400 kV lines, or lines at higher voltage levels, are automatically re-closed
after 400 ms disconnection. If the failure would still remain, the line is kept
disconnected during a larger time period, typically 800 ms, before the next
re-closure is done. If the failure would remain after the second re-closure
172 B. Protections in Electric Power Systems

attempt, more attempts to re-close will not be made. In this latter case the
isolation of the line has probably been permanently damaged and it must
be repaired before the line can be put in operation again. (This scenario
is typical. Due to specific conditions in different countries, deviations can
occur.)
Automatic re-closure must be made with certain carefulness when it is
made close to large thermal power plants, e.g. nuclear power plants. The
connection can in this case cause large stresses on the generator shaft if it
occurs at certain phase positions. The re-closure can in this case be done in
the remote end of the line seen from the thermal power plant, and then a
synchronised re-closure at the other end is made. (With synchronised closure
is meant that the voltage over the breaker is zero when it is closed.) In that
way the transients in the system are drastically reduced, but of course the
re-closure takes longer time.
In some systems all fault clearings on the high voltage grid are made on
all three phases. In other systems one phase clearing is used. This means
that only the faulty phase(s) is (are) disconnected at fault clearing and re-
closure.

B.3 Out of Step Protections

A synchronous machine which has fallen out of step, i.e. its angular velocity
does not coincide with the angular velocity of the net, has lost the synchro-
nism with the system, and the machine must be disconnected. In order to
supply electrical power to the system it must be phased in to the system
later on. During the time period when the synchronous machine falls out
of step large current pulses will pass through the generator, and if these
become too many they can damage the generator and limit its life time.
Furthermore, vibrations that can jeopardise the generator can arise. To
protect a synchronous machine which has fallen out of step the synchronous
machine is equipped with a out of step relay.
Also in this case an impedance is defined from voltage and current. If the
operation state of the generator is in the critical impedance zone a clock is
started. If the generator then comes into the critical zone repeatedly times,
this is a criterion that the generator is falling out of step, and the protection
gives a trip order to the generator breakers.
Generators are also equipped with other protections against overload
(stator current protection), over-excitation, etc.

B.4 System Protections

System protections are special types of protection, the primary task of which
is not to isolate failed equipment, but to prevent that the total system or
B.4. System Protections 173

large parts of it collapse. System protections often use information from

several different points in the system or quantities which can give a reliable
diagnosis of the state of the system. These systems often work in a time
scale which is considerably longer than the more device oriented protections
which were considered earlier, typically several seconds or minutes. An
example of a system protection is load shedding. This is used to avoid that
the frequency in the system falls below acceptable values if the generation
capacity has dropped in the system. The load shedding then disconnects
predetermined loads depending on how much and how fast the frequency is
falling. Voltage collapse protection is another system protection, the task of
which is to prevent voltage collapse in the system. Load shedding uses only
the frequency as input signal, while the voltage collapse protection often
uses several different quantities as input signals.