ECEN 667

Power System Stability

Lecture 2: Numeric Solution of Differential
Equations

Prof. Tom Overbye

Dept. of Electrical and Computer Engineering
Texas A&M University
[email protected]
 Announcements
• RSVP to Alex at [email protected] for the
TAMU ECE Energy and Power Group (EPG)
picnic. It starts at 5pm on September 27, 2019
• Be reading Chapters 1 and 2

2
 Power System Stability Terms
• Terms continue to evolve, but a good reference is [1]; image shows Figure 1
from this reference

[1] IEEE/CIGRE Joint Task Force on Stability Terms and Definitions, “Definitions and Classification of Power
System Stability,” IEEE Transactions Power Systems, May 2004, pp. 1387-1401 3
 Physical Structure
Power System Components

P. Sauer and M. Pai, Power System Dynamics and Stability

4
 Physical Structure
Power System Components
Mechanical System Electrical System
Stabilizer Load
Line
Exciter Relay Relay
Supply Pressure Speed Voltage Network Load
control control control Control control control
Fuel Furnace
Turbine Generator Network Loads
Source and Boiler
Machine Load
Fuel Steam Torque V, I P, Q Char.
Governor

5
 Differential Algebraic Equations
• Many problems, including many in the power area, can
be formulated as a set of differential, algebraic
equations (DAE) of the form
x  f (x, y )
0  g ( x, y )

• A power example is transient stability, in which f

represents (primarily) the generator dynamics, and g
(primarily) the bus power balance equations
• We'll initially consider the simpler problem of just
x  f (x)
6
Ordinary Differential Equations (ODEs)
• Assume we have a problem of the form
x  f (x) with x(t 0 )  x 0
• This is known as an initial value problem, since the
initial value of x is given at some time t0
– We need to determine x(t) for future time
– Initial value, x0, must be either be given or determined by
solving for an equilibrium point, f(x) = 0
– Higher-order systems can be put into this first order form
• Except for special cases, such as linear systems, an
analytic solution is usually not possible – numerical
methods must be used
7
 Equilibrium Points
• An equilibrium point x* satisfies
x  f (x*)  0
• An equilibrium point is stable if the response to a small
disturbance remains small
– This is known as Lyapunov stability
– Formally, if for every e > 0, there exists a d = d(e) > 0 such
that if x(0) – x* < d, then x(t) – x* < e for t  0
• An equilibrium point has asymptotic stability if there
exists a d > 0 such that if x(0) – x* < d, then
lim x(t )  x *  0
t 

8
 Power System Application
• A typical power system application is to assume
the power flow solution represents an equilibrium
point
• Back solve to determine the initial state variables,
x(0)
• At some point a contingency occurs, perturbing the
state away from the equilibrium point
• Time domain simulation is used to determine
whether the system returns to the equilibrium point

9
 Initial value Problem Examples
Example 1: Exponential Decay
A simple example with an analytic solution is
x   x with x(0)  x 0
This has a solution x(t)  x 0et
Example 2: Mass-Spring System
k x  gM  Mx  Dx
or
x 1  x2
1
x 2   k x1  g M  D x2 
M
10
 Numerical Solution Methods

• Numerical solution methods do not generate exact

solutions; they practically always introduce some error
– Methods assume time advances in discrete increments, called
a stepsize (or time step), Dt
– Speed accuracy tradeoff: a smaller Dt usually gives a better
solution, but it takes longer to compute
– Numeric roundoff error due to finite computer word size
• Key issue is the derivative of x, f(x) depends on x, the
value we are trying to determine
• A solution exists as long as f(x) is continuously
differentiable
11
 Numerical Solution Methods
• There are a wide variety of different solution
approaches, we will only touch on several
• One-step methods: require information about
solution just at one point, x(t)
– Forward Euler
– Runge-Kutta
• Multi-step methods: make use of information at
more than one point, x(t), x(t-D t), x(t-D 2 t)…
– Adams-Bashforth
• Predictor-Corrector Methods: implicit
– Backward Euler
12
 Error Propagation

• At each time step the total round-off error is the sum of

the local round-off at time and the propagated error
from steps 1, 2 , … , k − 1
• An algorithm with the desirable property that local
round-off error decays with increasing number of steps
is said to be numerically stable
• Otherwise, the algorithm is numerically unstable
• Numerically unstable algorithms can nevertheless give
quite good performance if appropriate time steps are
used
– This is particularly true when coupled with algebraic
equations 13
 Forward Euler’s Method
• The simplest technique for numerically integrating
such equations is known as the Euler's Method
(sometimes the Forward Euler's Method)
• Key idea is to approximate
dx x
x  f (x(t ))  as
dt t
Then
x(t  t )  x(t )  t f (x(t ))
• In general, the smaller the Dt, the more accurate the
solution, but it also takes more time steps
14
 Euler’s Method Algorithm
Set t = t 0 (usually 0)
x(t 0 ) = x0
Pick the time step t, which is problem specific
While t  t end Do
x(t  t )  x(t )  t f (x(t ))
t  t  t
End While

15
 Euler’s Method Example 1
Consider the Exponential Decay Example
x   x with x(0)  x 0
This has a solution x(t)  x 0e t
Since we know the solution we can compare the accuracy
of Euler's method for different time steps

16
 Euler’s Method Example 1, cont’d
t xactual(t) x(t) Dt=0.1 x(t) Dt=0.05
0 10 10 10
0.1 9.048 9 9.02
0.2 8.187 8.10 8.15
0.3 7.408 7.29 7.35
… … … …
1.0 3.678 3.49 3.58
… … … …
2.0 1.353 1.22 1.29
17
 Euler’s Method Example 2
Consider the equations describing the horizontal
position of a cart attached to a lossless spring:
x 1  x2
x 2   x1
Assuming initial conditions of x1(0)  1 and x 2 (0)  0,
the analytic solution is x1(t )  cos t.

We can again compare the results of the analytic and

numerical solutions

18
 Euler's Method Example 2, cont'd
Starting from the initial conditions at t =0 we next
calculate the value of x(t) at time t = 0.25.
x1 (0.25)  x1 (0)  0.25 x2 (0)  1.0
x2 (0.25)  x2 (0)  0.25 x1(0)  0.25
Then we continue on to the next time step, t = 0.50
x1 (0.50)  x1 (0.25)  0.25 x2 (0.25) 
 1.0  0.25  (0.25)  0.9375
x2 (0.50)  x2 (0.25)  0.25 x1(0.25) 
 0.25  0.25  (1.0)  0.50

19
Euler's Method Example 2, cont'd
t x1actual(t) x1(t) Dt=0.25
0 1 1
Since we
know from
0.25 0.9689 1 the exact
0.50 0.8776 0.9375 solution that
x1 is bounded
0.75 0.7317 0.8125
between
1.00 0.5403 0.6289 -1 and 1,
… … … clearly the
method is
10.0 -0.8391 -3.129
numerically
100.0 0.8623 -151,983 unstable
20
 Euler's Method Example 2, cont'd

Below is a comparison of the solution values for x1(t)

at time t = 10 seconds
Dt x1(10)

actual -0.8391
0.25 -3.129
0.10 -1.4088
0.01 -0.8823
0.001 -0.8423

21
 Second Order Runge-Kutta Method
• Runge-Kutta methods improve on Euler's method by
evaluating f(x) at selected points over the time step
• Simplest method is the second order method in which
1
 x  t  t      x  t      k 1   k 2  
2
where   
k1    t f  x  t  
 k 2    t f  x  t   k 1  

• That is, k1 is what we get from Euler's; k2 improves on

this by reevaluating at the estimated end of the time
step 22
 Second Order Runge-Kutta Algorithm

t = 0, x(0) = x0, Dt = step size

While t  tfinal Do
k1 = Dt f(x(t))
k2 = Dt f(x(t) + k1)
x(t+Dt) = x(t) + ( k1 + k2)/2
t = t + Dt
End While

23
 RK2 Oscillating Cart
• Consider the same example from before the
position of a cart attached to a lossless spring.
Again, with initial conditions of x1(0) =1 and x2(0)
= 0, the analytic solution is x1(t) = cos(t)
x 1  x2
x 2   x1
0  0 
• With Dt=0.25 k1  (0.25)      
1
   0.25 
at t = 0
1   0   1 
x(0)  k1        
0
   0.25   0.25 
24
 RK2 Oscillating Cart
 0.0625
k 2  (0.25)  f  x(0)  k1    
 0.25 
1  1 0.96875
x(0.25)      k1  k 2    
0
  2  0.25 

25
 Comparison
• The below table compares the numeric and exact
solutions for x1(t) using the RK2 algorithm
time actual x1(t) x1(t) with RK2
Dt=0.25
0 1 1
0.25 0.9689 0.969
0.50 0.8776 0.876
0.75 0.7317 0.728
1.00 0.5403 0.533
10.0 -0.8391 -0.795
100.0 0.8623 1.072

26
 Comparison of x1(10) for varying Dt

• The below table compares the x1(10) values for

different values of Dt; recall with Euler's with Dt=0.1
was -1.41 and with 0.01 was -0.8823
Dt x1(10)
actual -0.8391
0.25 -0.7946
0.10 -0.8310
0.01 -0.8390
0.001 -0.8391

27
 RK2 Versus Euler's
• RK2 requires twice the function evaluations per
iteration, but gives much better results
• With RK2 the error tends to vary with the cube of
the step size, compared with the square of the step
size for Euler's
• The smaller error allows for larger step sizes
compared to Eulers

28
 Fourth Order Runge-Kutta
• Other Runge-Kutta algorithms are possible, including
the fourth order
1
 x  t  t      x  t      k 1   2k 2  2k 3  k 4  
6
where   
k1    t f  x  t  
 1 
 k 2    t f  x  t   k 1  
 2 
 1 
 k 3    t f  x  t   k 2  
 2 
k 4    t f  x  t   k 2  
29
 RK4 Oscillating Cart Example
• RK4 gives much better results, with error varying with
the time step to the fifth power
time actual x1(t) x1(t) with RK4
Dt=0.25
0 1 1
0.25 0.9689 0.9689
0.50 0.8776 0.8776
0.75 0.7317 0.7317
1.00 0.5403 0.5403
10.0 -0.8391 -0.8392
100.0 0.8623 0.8601

30
 Multistep Methods
• Euler's and Runge-Kutta methods are single step
approaches, in that they only use information at
x(t) to determine its value at the next time step
• Multistep methods take advantage of the fact that
using we have information about previous time
steps x(t-Dt), x(t-2Dt), etc
• These methods can be explicit or implicit
(dependent on x(t+Dt) values; we'll just consider
the explicit Adams-Bashforth approach

31
 Multistep Motivation
• In determining x(t+Dt) we could use a Taylor series
expansion about x(t)
t 2
x(t  t )  x(t )  t x (t )  x(t )  O (t 3 )
2
t 2  f (x(t ))  f ( x(t  t )) 
x(t  t )  x(t )  t f (t )    O( t ) 
2  t 
3 1
x(t  t )  x(t )  t  f  x(t )   f  x(t  t )    O(t 3 )

2 2 

(note Euler's is just the first two terms on the right-

hand side)
32
 Adams-Bashforth
• What we derived is the second order Adams-Bashforth
approach. Higher order methods are also possible, by
approximating subsequent derivatives. Here we also
present the third order Adams-Bashforth
Second Order
t
x(t  t )  x(t )   3f (x(t ))  f ( x(t  t ))   O(t 3 )
2
Third Order
t
x(t  t )  x(t )   23f (x(t ))  16f ( x(t  t ))  5f ( x(t  2t ))   O( t 4 )
12

33
 Adams-Bashforth Versus
Runge-Kutta
• The key Adams-Bashforth advantage is the
approach only requires one function evaluation per
time step while the RK methods require multiple
evaluations
• A key disadvantage is when discontinuities are
encountered, such as with limit violations;
• Another method needs to be used until there are
sufficient past solutions
• They also have difficulties if variable time steps are
used

34
 Numerical Instability
• All explicit methods can suffer from numerical
instability if the time step is not correctly chosen for
the problem eigenvalues
Values are scaled by the
time step; the shape
for RK2 has similar
dimensions but is closer
to a square. Key point
is to make sure the time
step is small enough
relative to the eigenvalues

Image source: http://www.staff.science.uu.nl/~frank011/Classes/numwisk/ch10.pdf

35
 Stiff Differential Equations
• Stiff differential equations are ones in which the
desired solution has components the vary quite rapidly
relative to the solution
• Stiffness is associated with solution efficiency: in order
to account for these fast dynamics we need to take
quite small time steps
x 1  x2
x 2  1000 x1  1001x2
 0 1 
x    x
 1000 1000 
x1 (t )  Ae t  Be 1000t
36
 Implicit Methods
• Implicit solution methods have the advantage of being
numerically stable over the entire left half plane
• Only methods considered here are the is the Backward
Euler and Trapezoidal
x  f (x(t ))  Ax(t ))
Then using backward Euler
x(t  t )  x(t )  t A (x(t  t ))
 I  t A  x(t  t )  x(t )
1
x(t  t )   I  t A  x(t )
37
 Implicit Methods
• The obvious difficulty associated with these methods is
x(t) appears on both sides of the equation
• Easiest to show the solution for the linear case:
x  f (x(t ))  Ax(t ))
Then using backward Euler
x(t  t )  x(t )  t A( x(t  t ))
 I  t A  x(t  t )  x(t )
1
x(t  t )   I  t A  x(t )

38
 Backward Euler Cart Example
• Returning to the cart example
 0 1
x    x(t ))
 1 0 
Then using backward Euler with t  0.25
1
1  1 0.25
x(t  t )   I  t A  x(t )    x(t )
 0.25 1 

39
 Backward Euler Cart Example
• Results with Dt = 0.25 and 0.05
time actual x1(t) with x1(t) with
x1(t) Dt=0.25 Dt=0.05
0 1 1 1
0.25 0.9689 0.9411 0.9629
0.50 0.8776 0.8304 0.8700
0.75 0.7317 0.6774 0.7185
1.00 0.5403 0.4935 0.5277
2.00 -0.416 -0.298 -0.3944

Note: Just because the method is numerically stable

doesn't mean it is error free! RK2 is more
accurate than backward Euler. 40
 Trapezoidal Linear Case
• For the trapezoidal with a linear system we have
x  f (x(t ))  Ax(t ))
t
x(t  t )  x(t )   A(x(t ))  A(x(t  t )) 
2
 I  t A  x(t  t )   I  t A  x(t )
 2   2 
1  t 
x(t  t )   I  t A   I  A  x(t )
 2 

41
 Trapezoidal Cart Example
• Results with Dt = 0.25, comparing between backward
Euler and trapezoidal
time actual Backward Trapezoidal
x1(t) Euler
0 1 1 1
0.25 0.9689 0.9411 0.9692
0.50 0.8776 0.8304 0.8788
0.75 0.7317 0.6774 0.7343
1.00 0.5403 0.4935 0.5446
2.00 -0.416 -0.298 -0.4067

42
 Example Transient Stability Results
• Figure shows simulated generator frequencies after a
large generator outage contingency

43
Spatial Variation of Frequency
Response: EI Model

44
 Electromagnetic Transients
• The modeling of very fast power system dynamics
(much less than one cycle) is known as electromagnetics
transients program (EMTP) analysis
– Covers issues such as lightning propagation and switching
surges
• Concept originally developed by Prof. Hermann
Dommel for his PhD in the 1960's (now emeritus at
Univ. British Columbia)
– After his PhD work Dr. Dommel worked at BPA where he was
joined by Scott Meyer in the early 1970's
– Alternative Transients Program (ATP) developed in response to
commercialization of the BPA code
45