MATPOWER

A MATLAB Power System Simulation Package

Version 3.1b1
August 1, 2006

Users Manual

Ray D. Zimmerman
[email protected]

Carlos E. Murillo-Snchez
[email protected]

Deqiang (David) Gan

[email protected]

1997-2005 Power Systems Engineering Research Center (PSERC)

School of Electrical Engineering, Cornell University, Ithaca, NY 14853

MATPOWER Users Manual

Version 3.1b1

Table of Contents

Table of Contents...................................................................................................................................2
1

Introduction................................................................................................................................3

2
2.1
2.2
2.3
2.4
2.5
3
3.1
3.2
3.3
3.4
3.4.1
3.4.2
3.4.3
3.5
3.6
3.7
4

Getting Started ...........................................................................................................................3

System Requirements....................................................................................................................3
Installation.....................................................................................................................................4
Running a Power Flow..................................................................................................................4
Running an Optimal Power Flow..................................................................................................4
Getting Help..................................................................................................................................4
Technical Reference ...................................................................................................................5
Data File Format ...........................................................................................................................5
Modeling.......................................................................................................................................8
Power Flow.................................................................................................................................10
Optimal Power Flow ...................................................................................................................11
Traditional AC OPF Formulation.............................................................................................12
Generalized AC OPF Formulation (fmincon and MINOPF)................................................16
DC OPF Formulation ..............................................................................................................23
Unit Decommitment Algorithm...................................................................................................23
MATPOWER Options ...............................................................................................................24
Summary of the Files..................................................................................................................28
Acknowledgments....................................................................................................................32

References..................................................................................................................................32

Appendix A: Notes on LP-Solvers for MATLAB...............................................................................33

Appendix B: Additional Notes..........................................................................................................33
Appendix C: Auction Code ...............................................................................................................34

MATPOWER Users Manual

Version 3.1b1

1 Introduction
What is MATPOWER?
MATPOWER is a package of MATLAB M-files for solving power flow and optimal power flow problems.
It is intended as a simulation tool for researchers and educators that is easy to use and modify.
MATPOWER is designed to give the best performance possible while keeping the code simple to understand and modify. The MATPOWER home page can be found at:
http://www.pserc.cornell.edu/matpower/
Where did it come from?
MATPOWER was developed by Ray D. Zimmerman, Carlos E. Murillo-Snchez and Deqiang Gan of
PSERC at Cornell University (http://www.pserc.cornell.edu/) under the direction of Robert Thomas. The
initial need for MATLAB based power flow and optimal power flow code was born out of the computational requirements of the PowerWeb project (see http://www.pserc.cornell.edu/powerweb/).
Who can use it?

MATPOWER is free. Anyone may use it.

We make no warranties, express or implied. Specifically, we make no guarantees regarding the
correctness MATPOWERs code or its fitness for any particular purpose.
Any publications derived from the use of MATPOWER must cite MATPOWER
http://www.pserc.cornell.edu/matpower/.
Anyone may modify MATPOWER for their own use as long as the original copyright notices
remain in place.
MATPOWER may not be redistributed without written permission.
Modified versions of MATPOWER, or works derived from MATPOWER, may not be distributed
without written permission.

2 Getting Started
2.1 System Requirements
To use MATPOWER you will need:
MATLAB version 5 or later1
MATLAB Optimization Toolbox (required only for some OPF algorithms)
Both are available from The MathWorks (see http://www.mathworks.com/).

MATPOWER 2.0 and earlier required only version 4 of Matlab.

MATPOWER Users Manual

Version 3.1b1

2.2 Installation
Step 1:
Step 2:
Step 3:

Go to the MATPOWER home page (http://www.pserc.cornell.edu/matpower/) and follow the

download instructions.
Unzip the downloaded file.
Place the files in a location in your MATLAB path.

2.3 Running a Power Flow

To run a simple Newton power flow on the 9-bus system specified in the file case9.m, with the default
algorithm options, at the MATLAB prompt, type:
>> runpf('case9')

2.4 Running an Optimal Power Flow

To run an optimal power flow on the 30-bus system whose data is in case30.m, with the default algorithm options, at the MATLAB prompt, type:
>> runopf('case30')

To run an optimal power flow on the same system, but with the option for MATPOWER to shut down
(decommit) expensive generators, type:
>> runuopf('case30')

2.5 Getting Help

As with MATLABs built-in functions and toolbox routines, you can type help followed by the name of a
command or M-file to get help on that particular function. Nearly all of MATPOWERs M-files have
such documentation. For example, the help for runopf looks like:
>> help runopf
RUNOPF Runs an optimal power flow.
[baseMVA, bus, gen, gencost, branch, f, success, et] = ...
runopf(casename, mpopt, fname, solvedcase)
Runs an optimal power flow and optionally returns the solved values in
the data matrices, the objective function value, a flag which is true if
the algorithm was successful in finding a solution, and the elapsed time
in seconds. All input arguments are optional. If casename is provided it
specifies the name of the input data file or struct (see also 'help
caseformat' and 'help loadcase') containing the opf data. The default
value is 'case9'. If the mpopt is provided it overrides the default
MATPOWER options vector and can be used to specify the solution
algorithm and output options among other things (see 'help mpoption' for
details). If the 3rd argument is given the pretty printed output will be
appended to the file whose name is given in fname. If solvedcase is
specified the solved case will be written to a case file in MATPOWER
format with the specified name. If solvedcase ends with '.mat' it saves
the case as a MAT-file otherwise it saves it as an M-file.

MATPOWER Users Manual

Version 3.1b1

MATPOWER also has many options which control the algorithms and the output. Type:
>> helpmpoption

and see Section 3.6 for more information on MATPOWER's options.

3 Technical Reference
3.1 Data File Format
The data files used by MATPOWER are simply MATLAB M-files or MAT-files which define and return
the variables baseMVA, bus, branch, gen, areas, and gencost. The baseMVA variable is a scalar and the
rest are matrices. Each row in the matrix corresponds to a single bus, branch, or generator. The columns
are similar to the columns in the standard IEEE and PTI formats. The details of the specification of the
MATPOWER case file can be found in the help for caseformat.m:
>> help caseformat
CASEFORMAT
Defines the MATPOWER case file format.
A MATPOWER case file is an M-file or MAT-file which defines the variables
baseMVA, bus, gen, branch, areas, and gencost. With the exception of
baseMVA, a scalar, each data variable is a matrix, where a row corresponds
to a single bus, branch, gen, etc. The format of the data is similar to
the PTI format described in
http://www.ee.washington.edu/research/pstca/formats/pti.txt
except where noted. An item marked with (+) indicates that it is included
in this data but is not part of the PTI format. An item marked with (-) is
one that is in the PTI format but is not included here. Those marked with
(2) were added for version 2 of the case file format. The columns for
each data matrix are given below.
MATPOWER Case Version Information:
A version 1 case file defined the data matrices directly. The last two,
areas and gencost, were optional since they were not needed for running
a simple power flow. In version 2, each of the data matrices are stored
as fields in a struct. It is this struct, rather than the individual
matrices that is returned by a version 2 M-casefile. Likewise a version 2
MAT-casefile stores a struct named 'mpc' (for MATPOWER case). The struct
also contains a 'version' field so MATPOWER knows how to interpret the
data. Any case file which does not return a struct, or any struct which
does not have a 'version' field is considered to be in version 1 format.
See also IDX_BUS, IDX_BRCH, IDX_GEN, IDX_AREA and IDX_COST regarding
constants which can be used as named column indices for the data matrices.
Also described in the first three are additional columns that are added
to the bus, branch and gen matrices by the power flow and OPF solvers.

MATPOWER Users Manual

Version 3.1b1

Bus Data Format

1
bus number (1 to 29997)
2
bus type
PQ bus
= 1
PV bus
= 2
reference bus
= 3
isolated bus
= 4
3
Pd, real power demand (MW)
4
Qd, reactive power demand (MVAr)
5
Gs, shunt conductance (MW (demanded) at V = 1.0 p.u.)
6
Bs, shunt susceptance (MVAr (injected) at V = 1.0 p.u.)
7
area number, 1-100
8
Vm, voltage magnitude (p.u.)
9
Va, voltage angle (degrees)
(-)
(bus name)
10 baseKV, base voltage (kV)
11 zone, loss zone (1-999)
(+) 12 maxVm, maximum voltage magnitude (p.u.)
(+) 13 minVm, minimum voltage magnitude (p.u.)
Generator Data Format
1
bus number
(-)
(machine identifier, 0-9, A-Z)
2
Pg, real power output (MW)
3
Qg, reactive power output (MVAr)
4
Qmax, maximum reactive power output (MVAr)
5
Qmin, minimum reactive power output (MVAr)
6
Vg, voltage magnitude setpoint (p.u.)
(-)
(remote controlled bus index)
7
mBase, total MVA base of this machine, defaults to baseMVA
(-)
(machine impedance, p.u. on mBase)
(-)
(step up transformer impedance, p.u. on mBase)
(-)
(step up transformer off nominal turns ratio)
8
status, > 0 - machine in service
<= 0 - machine out of service
(-)
(% of total VAr's to come from this gen in order to hold V at
remote bus controlled by several generators)
9
Pmax, maximum real power output (MW)
10 Pmin, minimum real power output (MW)
(2) 11 Pc1, lower real power output of PQ capability curve (MW)
(2) 12 Pc2, upper real power output of PQ capability curve (MW)
(2) 13 Qc1min, minimum reactive power output at Pc1 (MVAr)
(2) 14 Qc1max, maximum reactive power output at Pc1 (MVAr)
(2) 15 Qc2min, minimum reactive power output at Pc2 (MVAr)
(2) 16 Qc2max, maximum reactive power output at Pc2 (MVAr)
(2) 17 ramp rate for load following/AGC (MW/min)
(2) 18 ramp rate for 10 minute reserves (MW)
(2) 19 ramp rate for 30 minute reserves (MW)
(2) 20 ramp rate for reactive power (2 sec timescale) (MVAr/min)
(2) 21 APF, area participation factor

MATPOWER Users Manual

Version 3.1b1

Branch Data Format

1
f, from bus number
2
t, to bus number
(-)
(circuit identifier)
3
r, resistance (p.u.)
4
x, reactance (p.u.)
5
b, total line charging susceptance (p.u.)
6
rateA, MVA rating A (long term rating)
7
rateB, MVA rating B (short term rating)
8
rateC, MVA rating C (emergency rating)
9
ratio, transformer off nominal turns ratio ( = 0 for lines )
(taps at 'from' bus, impedance at 'to' bus, i.e. ratio = Vf / Vt)
10 angle, transformer phase shift angle (degrees)
(-)
(Gf, shunt conductance at from bus p.u.)
(-)
(Bf, shunt susceptance at from bus p.u.)
(-)
(Gt, shunt conductance at to bus p.u.)
(-)
(Bt, shunt susceptance at to bus p.u.)
11 initial branch status, 1 - in service, 0 - out of service
(2) 12 minimum angle difference, angle(Vf) - angle(Vt) (degrees)
(2) 13 maximum angle difference, angle(Vf) - angle(Vt) (degrees)
(+) Area Data Format
1
i, area number
2
price_ref_bus, reference bus for that area
(+) Generator Cost Data Format
NOTE: If gen has n rows, then the first n rows of gencost contain
the cost for active power produced by the corresponding generators.
If gencost has 2*n rows then rows n+1 to 2*n contain the reactive
power costs in the same format.
1
model, 1 - piecewise linear, 2 - polynomial
2
startup, startup cost in US dollars
3
shutdown, shutdown cost in US dollars
4
n, number of cost coefficients to follow for polynomial
cost function, or number of data points for piecewise linear
5 and following, cost data defining total cost function
For polynomial cost:
c2, c1, c0
where the polynomial is c0 + c1*P + c2*P^2
For piecewise linear cost:
x0, y0, x1, y1, x2, y2, ...
where x0 < x1 < x2 < ... and the points (x0,y0), (x1,y1),
(x2,y2), ... are the end- and break-points of the cost function.

Some columns are added to the bus, branch and gen matrices by the solvers. See the help for idx_bus,
idx_brch, and idx_gen for more details.

MATPOWER Users Manual

Version 3.1b1

3.2 Modeling
AC Formulation
Fixed loads are modeled as constant real and reactive power injections, Pd and Qd specified in columns 3
and 4, respectively, of the bus matrix. The shunt admittance of any constant impedance shunt elements at
a bus are specified by Gsh and Bsh in columns 5 and 6, respectively, of the bus matrix
G + jBsh
Ysh = sh

baseMVA
transmission line, transformer or phase shifter, is modeled as a standard -model
Each branch,whether
transmission line, with series resistance R and reactance X and total line charging capacitance Bc , in series
with an ideal transformer and phase
shifter, at the from end, with tap ratio and phase shift angle shift .
The parameters R, X, Bc , and shift , are found in columns 3, 4, 5, 9 and 10 of the branch matrix, respectively. The branch voltages and currents at the from and to ends of the branch are related by the branch

admittance matrix Ybr as follows

If
Vf
(1)
= Ybr
It
Vt

Bc 1
Ys + j 2
2
where Ybr =
Ys 1
j

e shift

e and Y = 1 .
s
B
R + jX
Ys + j c
2
1

Ys

j shift

The elements of the individual branch admittance matrices and the bus shunt admittances are combined
by MATPOWER to form a complexbus admittance matrix Ybus , relating the vector of complex bus voltages Vbus with the vector of complex bus current injections Ibus
Ibus = Ybus Vbus

Similarly, admittance matrices Yf and Yt , are formed to compute the vector of complex current injections
at the from and to ends of each line, given

the bus voltages Vbus .

I f = Yf Vbus
It = Yt Vbus

The vectors of complex bus power injections, and branch power injections can be expressed as

Sbus = diag(Vbus )I*bus

S f = diag(Vf )I*f
St = diag(Vt )I*t

MATPOWER Users Manual

Version 3.1b1

where Vf and Vt are vectors of the complex bus voltages at the from and to ends, respectively, of all
branches, and diag() converts a vector into a diagonal matrix with the specified vector on the diagonal.

DC Formulation

For the DC formulation, the same parameters are used, with the exception that the following assumptions
are made:

Branch resistances R and charging capacitances Bc are negligible (i.e. branches are lossless).

All bus voltage magnitudes are close to 1 p.u.

Voltage angle differences are small enough that sin ij ij .

Combining these assumptions and equation (1) with the fact that S = VI * , the relationship between the
real power flows and voltage angles for an individual branch can be written as

Pf
f Pf ,shift
(2)

= Bbr +
Pt
t Pt,shift

where

1 1 1

X 1 1

(3)

Pf ,shift shift 1

=
.
Pt,shift X 1

(4)

Bbr =

The elements of the individual branch shift injections and Bbr matrices are combined by MATPOWER to
form a bus Bbus matrix and Pbus,shift
shift injection vector, which can be used to compute bus real power
injections from bus voltage angles
Pbus =Bbus bus + Pbus,shift

Similarly, MATPOWER builds the matrix B f and the vector P f,shift which can be used to compute the
vector s Pf and Pt of branch realpower injections

Pf = B f bus + Pf,shift

Pt = Pf

MATPOWER Users Manual

Version 3.1b1

3.3 Power Flow

MATPOWER has five power flow solvers, which can be accessed via the runpf function. In addition to
printing output to the screen, which it does by default, runpf optionally returns the solution in output arguments:
>> [baseMVA, bus, gen, branch, success, et] = runpf(casename);

The solution values are stored as follows:

bus voltage magnitudes
bus voltage angles
generator real power injections
generator reactive power injections
real power injected into from end of branch
real power injected into to end of branch
reactive power injected into from end of branch
reactive power injected into to end of branch
1 = solved successfully, 0 = unable to solve
computation time required for solution
The default power flow solver is based on a standard Newtons method [12] using a full Jacobian, updated at each iteration. This method is described in detail in many textbooks. Algorithms 2 and 3 are
variations of the fast-decoupled method [10]. MATPOWER implements the XB and BX variations as described in [1]. Algorithm 4 is the standard Gauss-Seidel method from Glimm and Stagg [5], based on
code contributed by Alberto Borghetti, from the University of Bologna, Italy. To use one of the power
flow solvers other than the default Newton method, the PF_ALG option must be set explicitly. For example, for the XB fast-decoupled method:
bus(:, VM)
bus(:, VA)
gen(:, PG)
gen(:, QG)
branch(:, PF)
branch(:, PT)
branch(:, QF)
branch(:, QT)
success
et

>> mpopt = mpoption('PF_ALG', 2);

>> runpf(casename, mpopt);

The last method is a DC power flow [13], which is obtained by executing runpf with the PF_DC option
set to 1, or equivalently by executing rundcpf directly. The DC power flow is obtained by a direct, noniterative solution of the bus voltage angles from the specified bus real power injections, based on equations (2), (3) and (4).
For the AC power flow solvers, if the ENFORCE_Q_LIMS option is set to true (default is false), then if any
generator reactive power limit is violated after running the AC power flow, the corresponding bus is converted to a PQ bus, with the reactive output set to the limit, and the case is re-run. The voltage magnitude
at the bus will deviate from the specified value in order to satisfy the reactive power limit. If the generator
at the reference bus is reaches a reactive power limit and the bus is converted to a PQ bus, the first remaining PV bus will be used as the slack bus for the next iteration. This may result in the real power output at this generator being slightly off from the specified values.
Currently, none of MATPOWERs power flow solvers include any transformer tap changing or handling
of disconnected or de-energized sections of the network.
Performance of the power flow solvers, with the exception of Gauss-Seidel, should be excellent even on
very large-scale power systems, since the algorithms and implementation take advantage of MATLABs
built-in sparse matrix handling.

10

MATPOWER Users Manual

Version 3.1b1

3.4 Optimal Power Flow

MATPOWER includes several solvers for the optimal power flow (OPF) problem, which can be accessed
via the runopf function. In addition to printing output to the screen, which it does by default, runopf optionally returns the solution in output arguments:
>> [baseMVA, bus, gen, gencost, branch, f, success, et] = runopf(casename);

In addition to the values listed for the power flow solvers, the OPF solution also includes the following
values:
bus(:, LAM_P)
bus(:, LAM_Q)
bus(:, MU_VMAX)
bus(:, MU_VMIN)
gen(:, MU_PMAX)
gen(:, MU_PMIN)
gen(:, MU_QMAX)
gen(:, MU_QMIN)
branch(:, MU_SF)
branch(:, MU_ST)
f

Lagrange multiplier on bus real power mismatch

Lagrange multiplier on bus reactive power mismatch
Kuhn-Tucker multiplier on upper bus voltage limit
Kuhn-Tucker multiplier on lower bus voltage limit
Kuhn-Tucker multiplier on upper generator real power limit
Kuhn-Tucker multiplier on lower generator real power limit
Kuhn-Tucker multiplier on upper generator reactive power limit
Kuhn-Tucker multiplier on lower generator reactive power limit
Kuhn-Tucker multiplier on MVA limit at "from" end of branch
Kuhn-Tucker multiplier on MVA limit at "to" end of branch
final objective function value

The (chronologically) first of the OPF solvers in MATPOWER is based on the constr function included
in earlier versions of MATLABs Optimization Toolbox, which uses a successive quadratic programming
technique with a quasi-Newton approximation for the Hessian matrix. The second approach is based on
linear programming. It can use the LP solver in the Optimization Toolbox or other MATLAB LP solvers
available from third parties. Version 3 of MATPOWER has a new generalized OPF formulation that allows general linear constraints on the optimization variables, but requires fmincon.m found in MATLABs
Optimization Toolbox 2.0 or later, or the MINOS [14] based MEX file available separately as part of the
optional MINOPF package (see http://www.pserc.cornell.edu/minopf/). MINOPF is distributed separately because it has a more restrictive license than MATPOWER.
The performance of MATPOWERs OPF solvers depends on several factors. First, the constr function
uses an algorithm which does not exploit or preserve sparsity, so it is inherently limited to small power
systems. The same is still true for the combination of parameters required to be able to employ the newer
fmincon function. The LP-based algorithm, on the other hand, does preserve sparsity. However, the LPsolver included in the older Optimization Toolbox does not exploit this sparsity. In fact, the LP-based
method with the old LP solver performs worse than the constr-based method, even on small systems.
Fortunately, there are LP-solvers available from third parties which do exploit sparsity. In general, these
yield much higher performance. One in particular, called BPMPD [8] (actually a QP-solver), has proven
to be robust and efficient. Even the constr or fmincon-based methods, when tricked into calling
BPMPD with full matrix data instead of the older qp.m, become much faster.
It should be noted, however, that even with a good LP-solver, MATPOWERs LP-based OPF solver, unlike its power flow solver, is not suitable for very-large scale problems. Substantial improvements in performance may still be possible, though they may require significantly more complicated coding and possibly a custom LP-solver. However, when speed is of the essence, the preferred choice is the MINOSbased MEX file solver; assuming that its licensing requirements can be met. It is coded in FORTRAN
and evaluates the required Jacobians using an optimized structure that follows the order of evaluation imposed by the compressed-column sparse format which is employed by MINOS. In fact, the new generalized OPF formulation included in MATPOWER 3.0 and later is inspired by the data format used by
MINOS.

11

MATPOWER Users Manual

Version 3.1b1

MATPOWERs OPF implementation is not currently able to handle unconnected or de-energized sections of the network.
Piecewise linear costs using constrained cost variables (CCV)
The OPF formulations in MATPOWER allow for the specification of convex piecewise linear cost functions for active or reactive generator output. An example of such a cost curve is shown below.

This non-differentiable cost is modeled using an extra helper cost variable for each such cost curve and
additional constraints on this variable and Pg, one for each segment of the curve. The constraints build a
convex basin equivalent to requiring the cost variable to lie in the epigraph of the cost curve. When the
cost is minimized, the cost variable will be pushed against this basin. If the helper cost variable is y, then
the contribution of the generators cost to the total cost is exactly y. In the above case, the two additional
required constraints are

1) y m1 (Pg x0 ) + c0

(y must lie above the first segment)

2) y m2 (Pg x1 ) + c1

(y must lie above the second segment)

where m1 and m2 are the slopes of the two segments. Also needed, of course, are the box restrictions on
Pg :PminPg Pmax. The additive part of the cost contributed by this generator is y.
This constrained cost variable (CCV) formulation is used by all of the MATPOWER OPF solvers for
handling piecewise linear cost functions.
3.4.1 Traditional AC OPF Formulation
The AC optimal power flow problem solved by MATPOWER is a smooth OPF with no discrete variables or controls. The objective function is the total cost of real and/or reactive generation. These costs
may be defined as polynomials or as piecewise-linear functions of generator output. The problem is formulated as follows:

12

MATPOWER Users Manual

min

,V ,Pg ,Qg

1i

Version 3.1b1

(Pgi ) + f2i (Qgi )

subject to
Pi ( ,V ) Pgi + Pdi = 0

(active power balance equations)

Qi ( ,V ) Qgi + Qdi = 0

(reactive power balance equations)

Sijf ( ,V ) Sijmax

(apparent power flow limit of lines, from end)

Sijt ( ,V ) Sijmax

(apparent power flow limit of lines, to end)

Vi min Vi Vi max

(bus voltage limits)

Pgimin Pgi Pgimax

(active power generation limits)

Qgimin Qgi Qgimax

(reactive power generation limits)

Here f1i and f2i are the costs of active and reactive power generation, respectively, for generator i at a given
dispatch point. Both f1i and f2i are assumed to be polynomial or piecewise-linear functions. By defining
the variable x as


V
x =
Pg

Qg
the problem can be expressed compactly as follows:
min f (x)
x

subject to

g1 (x) = 0

(power balance equations)

g2 (x) 0

(branch flow limits)

xmin x xmax
(variable limits)

Optimization Toolbox
Based OPF Solver (constr)
two original OPF solvers in MATPOWER is based on the constr non-linear constrained
The first of the
optimization function in MATLABs Optimization Toolbox. The constr function and the algorithms it
uses are covered in the older Optimization Toolbox manual [6]. MATPOWER provides constr with two
M-files which it uses during for the optimization. One computes the objective function, f, and the constraint violations, g, at a given point, x, and the other computes their gradients f x and g x .
MATPOWER has two versions of these M-files. One set is used to solve systems with polynomial cost
functions. In this formulation, the cost functions are included in a straightforward way into the objective
function. The other set is used to solve systems with piecewise-linear costs. Piecewise-linear cost functions are handled by introducing a cost variable for each piecewise-linear cost function. The objective
function is simply the sum of these cost variables which are then constrained to lie above each of the linear functions which make up the piecewise-linear cost function. Clearly, this method works only for con13

MATPOWER Users Manual

Version 3.1b1

vex cost functions. In the MATPOWER documentation this will be referred to as a constrained cost variable (CCV) formulation.
The algorithm codes 100 and 200, respectively, are used to identify the constr-based solver for polynomial and piecewise-linear cost functions. If algorithm 200 is chosen for a system with polynomial cost
function, the cost function will be approximated by a piecewise-linear function by evaluating the polynomial at a fixed number of points determined by the options vector (see Section3.6 for more details on the
MATPOWER options).
It should be noted that the constr-based method can also benefit from a superior QP-solver such as
bpmpd. See Appendix A for more information on LP and QP-solvers.
LP-Based OPF Solver (LPconstr)
Linear programming based OPF methods are in wide use today in the industry. However, the LP-based
algorithm included in MATPOWER is much simpler than the algorithms used in production-grade software.
The LP-based methods in MATPOWER use the same problem formulation as the constr-based methods, including the CCV formulation for the case of piecewise-linear costs. The compact form of the OPF
problem can be rewritten to partition g into equality and inequality constraints, and to partition the variable x as follows:
min f (x 2 )
x

subject to
g1 (x 1 ,x 2 ) = 0
g2 (x 1 , x2 ) 0

(equality constraints)
(inequality constraints)

where x1 contains the system voltage magnitudes and angles, and x2 contains the generator real and reactive power outputs (and corresponding cost variables for the CCV formulation). This is a general nonlinear programming problem, with the additional assumption that the equality constraints can be used to
solve for x1 , given a value for x2 .
The LP-based OPF solver is implemented with a function LPconstr, which is similar to constr in that it
uses the same M-files for computing the objective function, constraints, and their respective gradients. In
addition, a third M-file (lpeqslvr.m) is needed to solve for x1 from the equality constraints, given a value
for x2 . This architecture makes it relatively simple to modify the formulation of the problem and still be
able to use both the constr-based and LP-based solvers.
The algorithm proceeds as follows, where the superscripts denote iteration number:
Step 0:

Set iteration counter k 0 and choose an appropriate initial value, call it x20 , for x2 .

Step 1:

Solve the equality constraint (power flow) equations g1 (x1k , x k2 ) = 0 for x1k .

14

MATPOWER Users Manual

Step 2:

Version 3.1b1

Linearize the problem around xk, solve the resulting LP for x.

f

min
x
x x
x=xk
subject to
g

x g( x k )

x x = x k
x

Step 3: Set k k + 1, update current solution x k = x k 1 + x .

Step 4: If xk meets termination criteria, stop, otherwise go to step 5.
Step 5: Adjust step size limit based on the trust region algorithm in [3], go to step 1.
The termination criteria is outlined below:
L f
g
=
+ T
tolerance1
x x
x
g(x) tolerance2
x tolerance3
Here is the vector of Lagrange multipliers of the LP problem. The first condition pertains to the size of
the gradient, the second to the violation of constraints, and the third to the step size. More detail can be
found in [4].
Quite frequently, the value of xk given by step 1 is infeasible and could result in an infeasible LP problem.
In such cases, a slack variable is added for each violated constraint. These slack variables must be zero at
the optimal solution.
The LPconstr function implements the following three methods:
sparse formulation with full set of inequality constraints
sparse formulation with relaxed constraints (ICS, Iterative Constraint Search)
dense formulation with relaxed constraints (ICS) [11]
These three methods are specified using algorithm codes 160, 140, and 120, respectively, for systems
with polynomial costs, and 260, 240, and 220, respectively, for systems with piecewise-linear costs. As
with the constr-based method, selecting one of the 2xx algorithms for a system with polynomial cost will
cause the cost to be replaced by a piecewise-linear approximation.
In the dense formulation, some of the variables x1 and the equality constraints g1 are eliminated from the
problem before posing the LP sub-problem. This procedure is outlined below. Suppose the LP subproblem is given by:
min c T x
subject to
A x b
x

15

MATPOWER Users Manual

Version 3.1b1

If this is rewritten as:

min c1T x1 + c2T x2
subject to
A11 x1 + A12 x 2 = b1
A21 x1 + A22 x2 b2
x
where A1 1 is a square matrix, x1 can be computed as:
x1 = A111 (b1 A12 x2 )
Substituting back in to the problem, yields a new LP problem:
min (-c1T A111 A12 + c T2 ) x2
subject to
A11 x1 + A12 x 2 = b1
A21 A111 (b1 A12 x2 ) + A22 x 2 b2
1 A111 (b1 A12x2 ) 1
2 x2 2
This new LP problem is smaller than the original, but it is no longer sparse.
As mentioned above, to realize the full potential of the LP-based OPF solvers, it will be necessary to obtain a good LP-solver, such as bpmpd. See Appendix A for more details.
3.4.2 Generalized AC OPF Formulation (fmincon and MINOPF)
The generalized AC OPF formulation, used by the fmincon and MINOPF solvers, offers a number of
extra capabilities relative to the traditional formulation used by the classic MATPOWER OPF solvers:
mixed polynomial and piecewise linear costs
dispatchable loads
generator P-Q capability curves
additional user supplied linear constraints
additional user supplied costs
New in MATPOWER 3.1 are the generalized user supplied cost formulation, the generator capability
curves and a simplification of the general linear constraint specification used in version 3.0.
The problem is formulated in terms of 3 groups of optimization variables, labeled x, y and z. The x variables are the OPF variables, consisting of the voltage angles and magnitudes V at each bus, and real
and reactive generator injections Pg and Q g . The y variables are the helper variables used by the CCV
formulation of the piecewise linear generator cost functions. Additional user defined variables are
grouped in z.

16

MATPOWER Users Manual

Version 3.1b1

The optimization problem can be expressed as follows:

1
min ( f1i (Pgi ) + f2i (Qgi )) + yi + wT Hw + C wT w
x,y,z
2
i
i
subject to
gP (x) =P( ,V ) Pg + Pd = 0
gQ (x) = Q( ,V ) Qg + Qd = 0

gS f (x) = S f ( ,V ) Smax 0
gSt (x) = St ( ,V ) Smax 0

x

l A y u
z
xmin x xmax

(active power balance equations)

(reactive power balance equations)
(apparent power flow limit of lines, from end)
(apparent power flow limit of lines, to end)
(general linear constraints)
(voltage and generation variable limits)

The most significant additions to the traditional, simple OPF formulation appear in the generalized cost
terms containingw and in the general linear constraints involving the matrix A, described in the next two
sections. These
allow tremendous flexibility in customizing the problem formulation,
two frameworks
making MATPOWER
even more useful as a research tool.
Note: In Optimization Toolbox versions 3.0 and earlier, fmincon seems to be providing inaccurate
shadow prices on the constraints. This did not happen with constr and it may be a bug in these versions of the Optimization Toolbox.
General Linear Constraints
In addition to the standard non-linear equality constraints for nodal power balance and non-linear inequality constraints for line flow limits, this formulation includes a framework for additional linear constraints involving the full set of optimization variables.
x

l A y u
(general linear constraints)
z
Some portions of these linear constraints are supplied directly by the user, while others are generated
automatically based on the case data. Automatically generated portions include:
rows for constraints

that define generator P-Q capability curves

rows for constant power factor constraints for dispatchable or price-sensitive loads
rows and columns for the y variables from the CCV implementation of piecewise linear generator
costs and their associated constraints
In addition to these automatically generated constraints, the user can provide a matrix Au and vectors lu
and uu to define further linear constraints. These user supplied constraints could be used, for example, to
restrict voltage angle differences between specific buses. The matrix Au must have at least nx columns

17

MATPOWER Users Manual

Version 3.1b1

where nx is the number of x variables. If Au has more than nx columns, a corresponding z optimization
variable is created for each additional column. These z variables also enter into the generalized cost terms
described below, so Au and N must have the same number of columns.
x
lu Au u u
(user supplied linear constraints)
z
Change from MATPOWER 3.0: The Au matrix supplied by the user no longer includes the (all zero) columns corresponding to the y variables for piecewise linear generator costs. This should simplify significantly the creation
of the desired Au matrix.
Generalized Cost Function
The cost function consists of 3 parts. The first two are the polynomial and piecewise linear costs, respectively, of generation. A polynomial or piecewise linear cost is specified for each generators active output
and, optionally, reactive output in the appropriate row(s) of the gencost matrix. Any piecewise linear
costs are implemented using the CCV formulation described above which introduces corresponding
helper y variables. The general formulation allows generator costs of mixed type (polynomial and piecewise linear) in the same problem.
The third part of the cost function provides a general framework for imposing additional costs on the optimization variables, enabling things such as using penalty functions as soft limits on voltages, additional
costs on variables involved in constraints handled by Langrangian relaxation, etc.
This general cost term is specified through a set of parameters H, Cw, N and fparm, described below. It
consists of a general quadratic function of an nw 1 vector w of transformed optimization variables.
1 T
w Hw + C wT w
2
H is the nw nw symmetric, sparse matrix of quadratic coefficients and Cw is the nw 1 vector of linear
coefficients. The sparse N matrix is nw n xz , where the number of columns must match that of any user
supplied Au matrix. And fparm is nw
4 , where the 4 columns are labeled as
f parm = [d

r h m] .

The vector w is created from the x and z optimization variables by first applying a general linear trans
formation

x
r = N ,
z

followed by a scaled function with a shifted dead zone, defined by the remaining elements of fparm.
Each element of r is transformed into the corresponding element of w as follows:

m
f (r r + h ),
i i i i i
wi =
0,
m f r r h ,
i i( i i i)

ri ri < hi
hi ri ri hi
ri ri > hi

where the function fi is a predetermined function selected by the index in di. The current implementation
includes linear and quadratic options.
t,
f i (t ) = 2
t ,

18

di = 1
di = 2

MATPOWER Users Manual

Version 3.1b1

The linear case, where di = 1, is illustrated below, where wi is found by shifting ri by ri , applying a dead
zone defined by hi and then scaling by mi.
wi

mi

ri

ri
hi

Generator P-Q Capability Curves

The traditional AC OPF formulation models the generator P-Q capability curves as simple box constraints defined by the PMIN, PMAX, QMIN and QMAX columns of the gen matrix. In MATPOWER 3.1, version 2 of the case file format is introduced, which includes 6 new columns in the gen matrix for specifying additional sloped upper and lower portions of the capability curves. The new columns are PC1, PC2,
QC1MIN, QC1MAX, QC2MIN, and QC2MAX. The feasible region for generator operation with this more general
capability curve is illustrated by the shaded region in the figure below.

19

MATPOWER Users Manual

Version 3.1b1

Qg
QC1MAX

QMAX

QC2MAX

Pg

QC2MIN

QMIN

QC1MIN
PC1 PMIN

PMAX

PC2

The particular values of PC1 and PC2 are not important and may be set equal to PMIN and PMAX for convenience. The important point is to set the corresponding QCnMAX (QCnMIN) limits such that the two resulting points define the desired line corresponding to the upper (lower) sloped portion of the capability
curve.
Dispatchable loads
In general, dispatchable or price-sensitive loads can be modeled as negative real power injections with
associated costs. The current test is that if PMIN < PMAX = 0 for a generator, then it is really a dispatchable
load. If a load has a demand curve like the following

20

MATPOWER Users Manual

Version 3.1b1

so that it will consume zero if the price is higher than price2, P1 if the price is less than price2 but higher
than price1, and P2 if the price is equal or lower than price1. Considered as a negative injection, the desired dispatch is zero if the price is greater than price2, -P1 if the price is higher than price1 but lower
than price2, and -P2 if the price is equal to or lower than price1. This suggests the following piecewise
linear cost curve:

Note that this assumes that the demand blocks can be partially dispatched or split; if the price trigger
is reached half-way through the block, the load must accept the partial block. Otherwise, accepting or rejecting whole blocks really poses a mixed-integer problem, which is outside the scope of the current
MATPOWER implementation.
When there are dispatchable loads, the issue of reactive dispatch arises. If the QMIN/QMAX generation limits for the negative generator in question are not set to zero, then the algorithm will dispatch the reactive injection to the most convenient value. Since this is not normal load behavior, in the generalized formulation it is assumed that dispatchable loads maintain a constant power factor. The mechanism for
posing additional general linear constraints is employed to automatically include restrictions for these
injections to keep the ratio of Pg and Qg constant. This ratio is inferred from the values of PMIN and either
QMIN (for inductive loads) or QMAX (for capacitive loads) in the gen table. It is important to set these appropriately, keeping in mind that PG is negative and that for normal inductive loads QG should also be
negative (a positive reactive load is a negative reactive injection). The initial values of the PG and QG columns of the gen matrix must be consistent with the ratio defined by PMIN and the appropriate Q limit.
Problem Data Transformation
Defining a user supplied A matrix to add additional linear constraints requires knowledge of the order of
the optimization variables in the x vector. This requires an understanding of the standard transformations
performed on the input data (bus, gen, branch, areas and gencost tables) before the problem is solved.
All of these transformations are reversed after solving the problem so the output data is correctly placed
in the tables.
The first step filters out inactive generators and branches; original tables are saved for data output.
comgen
onbranch
gen
branch

=
=
=
=

find(gen(:,GEN_STATUS) > 0);

find(branch(:,BR_STATUS) ~= 0);
gen(comgen, :);
branch(onbranch, :);

% find online generators

% find online branches

The second step is a renumbering of the bus numbers in the bus table so that the resulting table contains
consecutively-numbered buses starting from 1:
21

MATPOWER Users Manual

Version 3.1b1

[i2e, bus, gen, branch, areas] = ext2int(bus, gen, branch, areas);

where i2e is saved for inverse reordering at the end. Finally, generators are further reordered by bus
number:
ng = size(gen,1);
% number of generators or injections
[tmp, igen] = sort(gen(:, GEN_BUS));
[tmp, inv_gen_ord] = sort(igen); % save for inverse reordering at the end
gen = gen(igen, :);
if ng == size(gencost,1)
% This is because gencost might have
gencost = gencost(igen, :);
% twice as many rows as gen if there
else
% are reactive injection costs.
gencost = gencost( [igen; igen+ng], :);
end

Having done this, the variables inside the x vector now have the same ordering as in the bus, gen tables:
x = [

Theta ;
V
;
Pg ;
Qg ];

%
%
%
%

nb
nb
ng
ng

bus voltage angles

bus voltage magnitudes
active power injections (p.u.) (ascending bus order)
reactive power injections (p.u.)(ascending bus order)

and the nonlinear constraints have the same order as in the bus, branch tables
g = [

gp;
gq;
gsf;
gst ];

%
%
%
%

nb
nb
nl
nl

real power flow mismatches (p.u.)

reactive power flow mismatches (p.u.)
"from" end apparent power injection limits (p.u.)
"to" end apparent power injection limits (p.u.)

With this setup, box bounds on the variables are applied as follows: the reference angle is bounded above
and below with the value specified for it in the original bus table. The V section of x is bounded above
and below with the corresponding values for VMAX and VMIN in the bus table. The Pg and Qg sections of
x are bounded above and below with the corresponding values for PMAX, PMIN, QMAX and QMIN in the gen
table. The nonlinear constraints are similarly setup so that gp and gq are equality constraints (zero RHS)
and the limits for gsf, gst are taken from the RATE_A column in the branch table.
Example of Additional Linear Constraint
The following example illustrates how an additional general linear constraint can be added to the problem
formulation. In the standard solution to case9.m, the voltage angle for bus 7 lags the voltage angle in bus
2 by 6.09 degrees. Suppose we want to limit that lag to 5 degrees at the most. A linear restriction of the
form
Theta(2) Theta(7) <= 5 degrees

would do the trick. We have nb = 9 buses, ng = 3 generators and nl = 9 branches. Therefore the first 9
elements of x are bus voltage angles, elements 10-18 are bus voltage magnitudes, elements 19-21 are active injections corresponding to the generators in buses 1, 2 and 3 (in that order) and elements 22-24 are
the corresponding reactive injections. Note that in this case the generators in the original data already appear in ascending bus order, so no permutation with respect to the original data is necessary. Going back
to the angle restriction, we see that it can be cast as
[ 0 1 0 0 0 0 -1 0 0 zeros(1,nb+ng+ng) ] * x

We can set up the problem as follows:

A = sparse([1;1], [2;7], [1;-1], 1, 24);
l = -Inf;

22

<= 5 degrees

MATPOWER Users Manual

Version 3.1b1

u = 5 * pi/180;
mpopt = mpoption('OPF_ALG', 520); % use fmincon w/generalized formulation
opf('case9', A, l, u, mpopt)

which indeed restricts the angular separation to 5 degrees.

3.4.3 DC OPF Formulation
The DC optimal power flow problem solved by MATPOWER is similar to the traditional AC OPF formulation described above, but using the DC model of the network, which only includes bus voltage angles and real power injections and flows.

min fi (Pgi )
,Pg

subject to
Bbus = Pg Pd Pbus,shift Gsh

B f P max Pf ,shift

B f P max + Pf ,shift

(active power balance equations)

(real power flow limit of lines, from end)
(real power flow limit of lines, to end)

Pgimin Pgi Pgimax

(active power generation limits)

The voltage
angle at the reference bus is also constrained to the specified value. Since all constraints are
problem
linear, the
is a simple LP or QP problem depending on the form of the cost function.
The current implementation of the DC OPF does not allow additional user-supplied linear constraints
and costs as in the generalized AC OPF formulation described above.

3.5 Unit Decommitment Algorithm

The standard OPF formulation described in the previous section has no mechanism for completely shutting down generators which are very expensive to operate. Instead they are simply dispatched at their
minimum generation limits. MATPOWER includes the capability to run an optimal power flow combined
with a unit decommitment for a single time period, which allows it to shut down these expensive units and
find a least cost commitment and dispatch. To run this for a case30, for example, type:
>> runuopf('case30')

MATPOWER uses an algorithm similar to dynamic programming to handle the decommitment. It proceeds through a sequence of stages, where stage N has N generators shut down, starting with N = 0.
The algorithm proceeds as follows:
Step 1: Begin at stage zero (N = 0), assuming all generators are on-line with all limits in place.
Step 2: Solve a normal OPF. Save the solution as the current best.
Step 3: Go to the next stage, N = N + 1. Using the best solution from the previous stage as the base
case for this stage, form a candidate list of generators with minimum generation limits binding.
If there are no candidates, skip to step 5.
Step 4: For each generator on the candidate list, solve an OPF to find the total system cost with this
generator shut down. Replace the current best solution with this one if it has a lower cost.
If any of the candidate solutions produced an improvement, return to step 3.
Step 5: Return the current best solution as the final solution.

23

MATPOWER Users Manual

Version 3.1b1

3.6 MATPOWER Options

MATPOWER uses an options vector to control the many options available. It is similar to the options
vector produced by the foptions function in early versions of MATLABs Optimization Toolbox. The
primary difference is that modifications can be made by option name, as opposed to having to remember
the index of each option. The default MATPOWER options vector is obtained by calling mpoption with
no arguments. So, typing:
>> runopf('case30', mpoption)

is another way to run the OPF solver with the all of the default options.
The MATPOWER options vector controls the following:
power flow algorithm
power flow termination criterion
OPF algorithm
OPF default algorithms for different cost models
OPF cost conversion parameters
OPF termination criterion
verbose level
printing of results
The details are given below:
>> help mpoption
MPOPTION Used to set and retrieve a MATPOWER options vector.
opt = mpoption
returns the default options vector
opt = mpoption(name1, value1, name2, value2, ...)
returns the default options vector with new values for up to 7
options, name# is the name of an option, and value# is the new
value. Example: options = mpoption('PF_ALG', 2, 'PF_TOL', 1e-4)
opt = mpoption(opt, name1, value1, name2, value2, ...)
same as above except it uses the options vector opt as a base
instead of the default options vector.
The currently defined options are as follows:
idx - NAME, default
description [options]
--------------------------------------------------power flow options
1 - PF_ALG, 1
power flow algorithm
[
1 - Newton's method
]
[
2 - Fast-Decoupled (XB version)
]
[
3 - Fast-Decoupled (BX version)
]
[
4 - Gauss Seidel
]
2 - PF_TOL, 1e-8
termination tolerance on per unit
P & Q mismatch
3 - PF_MAX_IT, 10
maximum number of iterations for
Newton's method

24

MATPOWER Users Manual

4

Version 3.1b1

- PF_MAX_IT_FD, 30

maximum number of iterations for

fast decoupled method
5 - PF_MAX_IT_GS, 1000
maximum number of iterations for
Gauss-Seidel method
6 - ENFORCE_Q_LIMS, 0
enforce gen reactive power limits,
at expense of |V|
[
0 or 1 ]
10 - PF_DC, 0
use DC power flow formulation, for
power flow and OPF
[
0 - use AC formulation & corresponding algorithm opts ]
[
1 - use DC formulation, ignore AC algorithm options
]
OPF options
11 - OPF_ALG, 0
algorithm to use for OPF
[
0 - choose best default solver available in the
]
[
following order, 500, 520 then 100/200
]
[ Otherwise the first digit specifies the problem
]
[ formulation and the second specifies the solver,
]
[ as follows, (see the User's Manual for more details)
]
[ 100 - standard formulation (old), constr
]
[ 120 - standard formulation (old), dense LP
]
[ 140 - standard formulation (old), sparse LP (relaxed)
]
[ 160 - standard formulation (old), sparse LP (full)
]
[ 200 - CCV formulation (old), constr
]
[ 220 - CCV formulation (old), dense LP
]
[ 240 - CCV formulation (old), sparse LP (relaxed)
]
[ 260 - CCV formulation (old), sparse LP (full)
]
[ 500 - generalized formulation, MINOS
]
[ 520 - generalized formulation, fmincon
]
[ See the User's Manual for details on the formulations.
]
12 - OPF_ALG_POLY, 100
default OPF algorithm for use with
polynomial cost functions
(used only if no solver available
for generalized formulation)
13 - OPF_ALG_PWL, 200
default OPF algorithm for use with
piece-wise linear cost functions
(used only if no solver available
for generalized formulation)
14 - OPF_POLY2PWL_PTS, 10
number of evaluation points to use
when converting from polynomial to
piece-wise linear costs
16 - OPF_VIOLATION, 5e-6
constraint violation tolerance
17 - CONSTR_TOL_X, 1e-4
termination tol on x for copf & fmincopf
18 - CONSTR_TOL_F, 1e-4
termination tol on F for copf & fmincopf
19 - CONSTR_MAX_IT, 0
max number of iterations for copf & fmincopf
[
0 => 2*nb + 150
]
20 - LPC_TOL_GRAD, 3e-3
termination tolerance on gradient for lpopf
21 - LPC_TOL_X, 1e-4
termination tolerance on x (min step size)
for lpopf
22 - LPC_MAX_IT, 400
maximum number of iterations for lpopf
23 - LPC_MAX_RESTART, 5
maximum number of restarts for lpopf
24 - OPF_P_LINE_LIM, 0
use active power instead of apparent power
for line flow limits
[
0 or 1 ]
25 - OPF_IGNORE_ANG_LIM, 0 ignore angle difference limits for branches
even if specified
[
0 or 1 ]

25

MATPOWER Users Manual

Version 3.1b1

output options
31 - VERBOSE, 1
amount of progress info printed
[
0 - print no progress info
]
[
1 - print a little progress info
]
[
2 - print a lot of progress info
]
[
3 - print all progress info
]
32 - OUT_ALL, -1
controls printing of results
[ -1 - individual flags control what prints
]
[
0 - don't print anything
]
[
(overrides individual flags, except OUT_RAW)
]
[
1 - print everything
]
[
(overrides individual flags, except OUT_RAW)
]
33 - OUT_SYS_SUM, 1
print system summary
[
0 or 1 ]
34 - OUT_AREA_SUM, 0
print area summaries
[
0 or 1 ]
35 - OUT_BUS, 1
print bus detail
[
0 or 1 ]
36 - OUT_BRANCH, 1
print branch detail
[
0 or 1 ]
37 - OUT_GEN, 0
print generator detail [
0 or 1 ]
(OUT_BUS also includes gen info)
38 - OUT_ALL_LIM, -1
control constraint info output
[ -1 - individual flags control what constraint info prints]
[
0 - no constraint info (overrides individual flags)
]
[
1 - binding constraint info (overrides individual flags)]
[
2 - all constraint info (overrides individual flags)
]
39 - OUT_V_LIM, 1
control output of voltage limit info
[
0 - don't print
]
[
1 - print binding constraints only
]
[
2 - print all constraints
]
[
(same options for OUT_LINE_LIM, OUT_PG_LIM, OUT_QG_LIM) ]
40 - OUT_LINE_LIM, 1
control output of line limit info
41 - OUT_PG_LIM, 1
control output of gen P limit info
42 - OUT_QG_LIM, 1
control output of gen Q limit info
43 - OUT_RAW, 0
print raw data for Perl database
interface code
[
0 or 1 ]
other options
51 - SPARSE_QP, 1
pass sparse matrices to QP and LP
solvers if possible
[
0 or 1 ]

26

MATPOWER Users Manual

Version 3.1b1

MINOPF options
61 - MNS_FEASTOL, 0 (1E-3)
62
63
64
65
66
67
68
69
70

71
72
73

primal feasibility tolerance,

set to value of OPF_VIOLATION by default
- MNS_ROWTOL, 0 (1E-3) row tolerance
set to value of OPF_VIOLATION by default
- MNS_XTOL, 0
(1E-3) x tolerance
set to value of CONSTR_TOL_X by default
- MNS_MAJDAMP, 0 (0.5)
major damping parameter
- MNS_MINDAMP, 0 (2.0)
minor damping parameter
- MNS_PENALTY_PARM, 0 (1.0) penalty parameter
- MNS_MAJOR_IT, 0 (200) major iterations
- MNS_MINOR_IT, 0 (2500) minor iterations
- MNS_MAX_IT, 0 (2500)
iterations limit
- MNS_VERBOSITY, -1
[ -1 - controlled by VERBOSE flag (0 or 1 below)
]
[
0 - print nothing
]
[
1 - print only termination status message
]
[
2 - print termination status and screen progress
]
[
3 - print screen progress, report file (usually fort.9) ]
- MNS_CORE, 1200 * nb + 5000
- MNS_SUPBASIC_LIM, 0 (2*ng) superbasics limit
- MNS_MULT_PRICE, 0 (30) multiple price

A typical usage of the options vector might be as follows:

Get the default options vector:
>> opt = mpoption;

Use the fast-decoupled method to solve power flow:

>> opt = mpoption(opt, 'PF_ALG', 2);

Display only system summary and generator info:

>> opt = mpoption(opt, 'OUT_BUS', 0, 'OUT_BRANCH', 0, 'OUT_GEN', 1);

Show all progress info:

>> opt = mpoption(opt, 'VERBOSE', 3);

Now, run a bunch of power flows using these settings:

>> runpf('case57', opt)
>> runpf('case118', opt)
>> runpf('case300', opt)

27

MATPOWER Users Manual

Version 3.1b1

3.7 Summary of the Files

Documentation files:
README
basic intro to MATPOWER
README.txt
basic intro to MATPOWER, with DOS line endings (for Windows)
docs/CHANGES
modification history of MATPOWER
docs/CHANGES.txt
modification history of MATPOWER, with DOS line endings
docs/manual.pdf
PDF version of the MATPOWER Users Manual
(see also caseformat.m & genform.m below)

Top-level programs:
cdf2matp.m
converts data from IEEE CDF to MATPOWER format
runcomp.m
runs 2 OPFs and compares results
rundcopf.m
runs a DC optimal power flow
rundcpf.m
runs a DC power flow
runduopf.m
runs a DC OPF with unit decommitment
runopf.m
runs an optimal power flow
runpf.m
runs a power flow
runuopf.m
runs an OPF with unit decommitment
(see also opf.m, copf.m, fmincopf.m, lpopf.m below which can also be used as top-level pro-

grams)
Input data files:
caseformat.m
case_ieee30.m
case118.m
case14.m
case30.m
case300.m
case30pwl.m
case30Q.m
case39.m
case4gs.m
case57.m
case6ww.m
case9.m
case9Q.m

documentation for input data file format

IEEE 30 bus system
IEEE 118 bus system
IEEE 14 bus system
modified IEEE 30 bus system
IEEE 300 bus system
case30.m with piecewise linear costs
case30.m with reactive power costs
39 bus system
4 bus system from Grainger & Stevenson
IEEE 57 bus system
6 bus system from Wood & Wollenberg
3 generator, 9 bus system (default case file)
case9.m with reactive power costs

28

MATPOWER Users Manual

Version 3.1b1

Common source files and utility functions used by multiple programs:

bustypes.m
creates vectors of bus indices for ref bus, PV buses, PQ buses
compare.m
prints summary of differences between 2 solved cases
dAbr_dV.m
computes partial derivatives of branch apparent power flows wrt. voltage,
used by OPF
dSbr_dV.m
computes partial derivatives of branch complex power flows wrt. voltage,
used by OPF & state estimator
dSbus_dV.m
computes partial derivatives of bus complex power injections wrt. voltage,
used by OPF, Newton PF, state estimator
ext2int.m
converts data matrices from external to internal bus numbering
hasPQcap.m
checks for generator P-Q capability curve constraints
have_fcn.m
checks for availability of optional functionality
idx_area.m
named column index definitions for areas matrix
idx_brch.m
named column index definitions for branch matrix
idx_bus.m
named column index definitions for bus matrix
idx_cost.m
named column index definitions for gencost matrix
idx_gen.m
named column index definitions for gen matrix
int2ext.m
converts data matrices from internal to external bus numbering
isload.m
checks if generators are actually dispatchable loads
loadcase.m
loads data from a case file or struct into data matrices
makeB.m
forms B matrix used by fast decoupled power flow
makeBdc.m
forms B matrix used by DC PF and DC OPF
makeSbus.m
forms bus complex power injections from specified generation and load
injections
makeYbus.m
forms complex bus admittance matrix
mp_lp.m
solves an LP problem with best solver available
mp_qp.m
solves a QP problem with best solver available
mpver.m
prints MATPOWER version information
printpf.m
pretty prints solved PF or OPF case
savecase.m
saves data matrices to a case file
mpoption.m
set MATPOWER options
Power Flow (PF):
dcpf.m
fdpf.m
gausspf.m
newtonpf.m
pfsoln.m

implements DC power flow solver

implements fast decouple power flow solver
implements Gauss-Seidel power flow solver
implements Newton power flow solver
updates data matrices with PF solution

29

MATPOWER Users Manual

Version 3.1b1

Optimal Power Flow (OPF):

common files shared by multiple OPF solvers
opf_form.m
returns code for formulation given OPF algorithm code
opf_slvr.m
returns code for solver given OPF algorithm code
opf.m
top-level OPF solver routine
poly2pwl.m
creates piecewise linear approximation to polynomial cost function
pqcost.m
splits gencost into real and reactive power costs
totcost.m
computes total cost for given dispatch
files used only by DC OPF
dcopf.m
implements DC optimal power flow
files used only for traditional OPF formulation (constr- and LP-based)
fg_names.m
returns names of function and gradient evaluators for given algorithm
fun_ccv.m
computes objective function and constraints for CCV formulation
fun_std.m
computes objective function and constraints for standard formulation
grad_ccv.m
computes gradients of obj fcn & constraints for CCV formulation
grad_std.m
computes gradients of obj fcn & constraints for standard formulation
opfsoln.m
updates data matrices with OPF solution
files used only by constr-based OPF
copf.m
implements constr-based OPF solver
files used only by LP-based OPF
lpopf.m
implements LP-based OPF solver
LPconstr.m
solves a non-linear optimization via sequential linear programming
LPeqslvr.m
runs Newton power flow
LPrelax.m
solves LP problem with constraint relaxation
LPsetup.m
solves LP problem using specified method
files used only for generalized OPF formulation (fmincon- and MINOS-based)
genform.m
documentation for generalized OPF formulation
makeAy.m
forms A matrix and b vector for generalized OPF formulation
files used only by fmincon-based OPF
consfmin.m
computes value and gradient of constraints
costfmin.m
computes value and gradient of objective function
fmincopf.m
implements fmincon-based OPF solver
files used only for OPF with unit decommitment
fairmax.m
same as MATLABs built-in max(), except breaks ties randomly
uopf.m
implements unit decommitment for OPF

30

MATPOWER Users Manual

Version 3.1b1

Extras: (in extras subdirectory)

auction market software (in smartmarket subdirectory)
auction.m
clears set of bids and offers based on pricing rules and OPF result
case2off.m
creates set of price/quantity bids/offers given gen and gencost matrices
idx_disp.m
named column index definitions for dispatch matrix
off2case.m
updates gen and gencost matrices based on quantity/price bids/offers
printmkt.m
prints the market output
runmkt.m
top-level program runs an OPF-based auction
SM_CHANGES
modification history of the smartmarket software
smartmkt.m
implements the smartmarket solver
unfinished state estimator (in state_estimator subdirectory)
runse.m
runs a state estimator
state_est.m
implements a state estimator
Tests: (in t subdirectory)
soln9_dcopf.mat
data used for tests
soln9_dcpf.mat
data used for tests
soln9_opf.mat
data used for tests
soln9_opf_extras1.matdata used for tests
soln9_opf_Plim.mat
data used for tests
soln9_opf_PQcap.mat data used for tests
soln9_pf.mat
data used for tests
t_auction.m
tests auction.m in extras/smartmarket
t_auction_case.m
test case for t_auction.m
t_auction_fmincopf.m fmincon-based tests of auction.m in extras/smartmarket
t_begin.m
starts a set of tests
t_case9_opf.m
case file (version 1 format) for OPF tests
t_case9_opfv2.m
case file (version 2 format) for OPF tests
t_case9_pf.m
case file (version 1 format) for power flow tests
t_case9_pfv2.m
case file (version 2 format) for power flow tests
t_end.m
finishes a set of tests and prints statistics
t_hasPQcap.m
tests hasPQcap.m
t_is.m
tests if two matrices are identical to some tolerance
t_jacobian.m
does numerical test of partial derivatives
t_loadcase.m
tests load_case.m
t_off2case.m
tests off2case.m
t_ok.m
tests if an expression is true
t_opf.m
tests OPF solvers
t_pf.m
tests PF solvers
t_run_tests.m
framework for running a series of tests
t_runmarket.m
tests runmarket.m
t_skip.m
skips a specified number of tests
test_matpower.m
runs all available MATPOWER tests

31

MATPOWER Users Manual

Version 3.1b1

4 Acknowledgments
The authors would like to acknowledge contributions from several people. Thanks to Chris DeMarco,
one of our PSERC associates at the University of Wisconsin, for the technique for building the Jacobian
matrix. Our appreciation to Bruce Wollenberg for all of his suggestions for improvements to version 1.
The enhanced output functionality in version 2.0 are primarily due to his input. Thanks also to Andrew
Ward for code which helped us verify and test the ability of the OPF to optimize reactive power costs.
Thanks to Alberto Borghetti for contributing code for the Gauss-Seidel power flow solver. Thanks also
to many others who have contributed code, bug reports and suggestions over the years. Last but not least,
we would like to acknowledge the input of Bob Thomas throughout the development of MATPOWER
here at PSERC Cornell.

5 References
1. R. van Amerongen, A General-Purpose Version of the Fast Decoupled Loadflow, IEEE Transactions on Power Systems, Vol. 4, No. 2, May 1989, pp. 760-770.
2. O. Alsac, J. Bright, M. Prais, B. Stott, Further Developments in LP-based Optimal Power Flow,
IEEE Transactions on Power Systems, Vol. 5, No. 3, Aug. 1990, pp. 697-711.
3. R. Fletcher, Practical Methods of Optimization, 2n d Edition, John Wiley & Sons, p. 96.
4. P. E. Gill, W. Murry, M. H. Wright, Practical Optimization, Academic Press, London, 1981.
5. A. F. Glimm and G. W. Stagg, Automatic calculation of load flows, AIEE Transactions (Power
Apparatus and Systems), vol. 76, pp. 817-828, Oct. 1957.
6. A. Grace, Optimization Toolbox, The MathWorks, Inc., Natick, MA, 1995.
7. C. Li, R. B. Johnson, A. J. Svoboda, A New Unit Commitment Method, IEEE Transactions on
Power Systems, Vol. 12, No. 1, Feb. 1997, pp. 113-119.
8. C. Mszros, The efficient implementation of interior point methods for linear programming and
their applications, Ph.D. Thesis, Etvs Lornd University of Sciences, 1996.
9. B. Stott, Review of Load-Flow Calculation Methods, Proceedings of the IEEE, Vol.62, No.7,
July 1974, pp. 916-929.
10. B. Stott and O. Alsac, Fast decoupled load flow, IEEE Transactions on Power Apparatus and
Systems, Vol. PAS-93, June 1974, pp. 859-869.
11. B. Stott, J. L. Marino, O. Alsac, Review of Linear Programming Applied to Power System Rescheduling, 1979 PICA, pp. 142-154.
12. W. F. Tinney and C. E. Hart, Power Flow Solution by Newtons Method, IEEE Transactions on
Power Apparatus and Systems, Vol. PAS-86, No. 11, Nov. 1967, pp. 1449-1460.
13. A. J. Wood and B. F. Wollenberg, Power Generation, Operation, and Control, 2n d Edition, John
Wiley & Sons, p. 108-111.
14. B.A Murtagh andM.A. Saunders, MINOS 5.5 Users Guide, Stanford University Systems Optimization Laboratory Technical Report SOL83-20R.

32

MATPOWER Users Manual

Version 3.1b1

Appendix A: Notes on LP-Solvers for MATLAB

When MATPOWER was initially developed the LP and QP solvers available in MATLABs Optimization
Toolbox, lp.m and qp.m, did not exploit sparsity and were therefore very slow for the large sparse problems typically encountered in power system simulation. Fortunately, there were some third party LP and
QP-solvers for MATLAB with much better performance.
Several LP and QP-solvers were tested for use in the context of an LP-based OPF. Some of them we
were unable to get to compile on our architecture of choice and others proved to be less than robust in an
OPF context.
Here is a list of the solvers we tested at the time:
bpmpd
- QP-solver from http://www.sztaki.hu/~meszaros/bpmpd/
Please see http://www.pserc.cornell.edu/bpmpd/ for a MATLAB MEX version.
lp.m
- LP-solver included with Optimization Toolbox 1.x and 2.x (from MathWorks)
lp_solve - LP-solver from ftp://ftp.ics.ele.tue.nl/pub/lp_solve/
loqo
- LP-solver from http://www.princeton.edu/~rvdb/
sol_qps.m - LP-solver developed at U. of Wisconsin (not publicly available)
Of all of the packages tested, the bpmpd solver, has been the only one which worked reliably for us. It
has proven to be very robust and has exceptional performance.
More information about free optimizers is available in Decision Tree for Optimization Software
maintained by Mittenlmonn Hans and P. Spellucci at http://plato.la.asu.edu/guide.html.
Since the initial development of MATPOWER, more recent versions of the MATLAB Optimization Toolbox have moved to new LP and QP solvers, linprog.m and quadprog.m. The LP solver, base on
LIPSOL, does support sparsity, but is still typically slower than bpmpd. The QP solver does not support
sparsity in general, only for certain restricted special cases.

Appendix B: Additional Notes

Some versions of MATLAB 5 were slow at selecting rows of a large sparse matrix, but much faster at
transposing and selecting columns.
fmincon.m seems to compute inaccurate shadow prices for Optimization Toolbox 3.0 and earlier.

33

MATPOWER Users Manual

Version 3.1b1

Appendix C: Auction Code

MATPOWER 3 includes in the extras/smartmarket directory code which implements a smart market auction clearing mechanism. The purpose of this code is to take a set of offers to sell and bids to
buy and use MATPOWERs optimal power flow to compute the corresponding allocations and prices. It
has been used extensively by the authors with the optional MINOPF package2 in the context of
POWERWEB3 but has not been widely tested in other contexts.
MATPOWER 3.1 includes a new version that supports offers and bids for reactive power and uses a new
interface.
The smart market algorithm consists of the following basic steps:
1. Convert block offers and bids into corresponding generator capacities and costs.
2. Run an optimal power flow with decommitment option (uopf) to find generator allocations and
nodal prices ( P ).
3. Convert generator allocations and nodal prices into set of cleared offers and bids.
4. Print results.
For step 1,
the offers and bids are supplied as two structs, offers and bids, each with fields P for real
power and Q for reactive power (optional). Each of these is also a struct with matrix fields qty and prc,
where the element in the i-th row and j-th column of qty and prc are the quantity and price, respectively
of the j-th block of capacity being offered/bid by the i-th generator. These block offers/bids are converted
to the equivalent piecewise linear generator costs and generator capacity limits by the off2case function.
See help off2case for more information.
Offer blocks must be in non-decreasing order of price and the offer must correspond to a generator with
0PMIN<PMAX. A set of price limits can be specified via the lim struct, e.g. and offer price cap on real
energy would be stored in lim.P.max_offer. Capacity offered above this price is considered to be
withheld from the auction and is not included in the cost function produced. Bids must be in nonincreasing order of price and correspond to a generator with PMIN<PMAX0 (see Dispatchable loads
on page 20). A lower limit can be set for bids in lim.P.min_bid. See help pricelimits for more information.
The data specified by a MATPOWER case file, with the gen and gencost matrices modified according
the step 1, are then used to run an OPF. A decommitment mechanism is used to shut down generators if
doing so results in a smaller overall system cost (see Section 3.5 Unit Decommitment Algorithm).
In step 3 the OPF solution is used to determine for each offer/bid block, how much was cleared and at
what price. These values are returned in co and cb, which have the same structure as offers and bids.
The mkt parameter is a struct used to specify a number of things about the market, including the type of
auction to use, type of OPF (AC or DC) to use and the price limits.
There are two basic types of pricing options available through mkt.auction_type, discriminative pricing
and uniform pricing. The various uniform pricing options are best explained in the context of an unconstrained lossless network. In this context, the allocation is identical to what one would get by creating bid
and offer stacks and finding the intersection point. The nodal prices ( P ) computed by the OPF and returned in bus(:,LAM_P) are all equal to the price of the marginal block. This is either the last accepted
offer (LAO) or the last accepted bid (LAB), depending which is the marginal block (i.e. the one that is

See http://www.pserc.cornell.edu/minopf/
3
See http://www.pserc.cornell.edu/powerweb/

34

MATPOWER Users Manual

Version 3.1b1

split by intersection of the offer and bid stacks). There is often a gap between the last accepted bid and
the last accepted offer. Since any price within this range is acceptable to all buyers and sellers, we end up
with a number of options for how to set the price, as listed in the table below.
Auction
Type

Name

discriminative

LAO

Uniform price equal to the last accepted offer.

FRO

Uniform price equal to the first rejected offer.

LAB

Uniform price equal to the last accepted bid.

FRB

Uniform price equal to the first rejected bid.

first price

second price

7
8

Description
The price of each cleared offer (bid) is equal to the offered (bid) price.

Uniform price equal to the offer/bid price of marginal unit.

Uniform price equal to min(FRO, LAB) if the marginal unit is an offer, or max(FRB, LAO) if it is a bid.

split-the-difference Uniform price equal to the average of the LAO and LAB.
dual LAOB

Uniform price for sellers equal to LAO, for buyers equal to LAB.

Generalizing to a network with possible losses and congestion results in nodal prices P which vary according to location. These P values can be used to normalize all bids and offers to a reference location
by adding a locational adjustment. For bids and offers at bus i, the adjustment is P,ref P,i , where P,ref
is the nodal price at the reference bus. The desired uniform pricing rule can then be applied to the ad This uniform price is then
justed offers and bids to get the appropriate uniform price at the reference bus.
by subtracting the locational adjustment. The appropriate locationally adjusted uniadjusted for location
form price is then used for all cleared bids and offers.

There are certain circumstances under which the price of a cleared offer determined by the above procedures can be less than the original offer price, such as when a generator is dispatched at its minimum
generation limit, or greater than the price cap lim.P.max_cleared_offer. For this reason all cleared
offer prices are clipped to be greater than or equal to the offer price but less than or equal to
lim.P.max_cleared_offer. Likewise, cleared bid prices are less than or equal to the bid price but
greater than or equal to lim.P.min_cleared_bid.
Handling Supply Shortfall
In single sided markets, in order to handle situations where the offered capacity is insufficient to meet the
demand under all of the other constraints, resulting in an infeasible OPF, we introduce the concept of
emergency imports. We model an import as a fixed injection together with an equal sized dispatchable
load which is bid in at a high price. Under normal circumstances, the two cancel each other and have no
effect on the solution. Under supply shortage situations, the dispatchable load is not fully dispatched, resulting in a net injection at the bus, mimicking an import. When used in conjunction with the LAO pricing rule, the marginal load bid will not set the price if all offered capacity can be used.
Example
Six generators with three blocks of capacity each, offering as follows:

35

MATPOWER Users Manual

Version 3.1b1

Generator

Block 1

Block 2

Block 3

MW @ $/MWh MW @ $/MWh

MW @ $/MWh

12 @ $20

24 @ $50

24 @ $60

12 @ $20

24 @ $40

24 @ $70

12 @ $20

24 @ $42

24 @ $80

12 @ $20

24 @ $44

24 @ $90

12 @ $20

24 @ $46

24 @ $75

12 @ $20

24 @ $48

24 @ $60

One fixed load of 151.64 MW.

Three dispatchable loads, bidding three blocks each as follows:
Block 1

Block 2

Block 3

MW @ $/MWh

MW @ $/MWh

MW @ $/MWh

10 @ $100

10 @ $70

10 @ $60

10 @ $100

10 @ $50

10 @ $20

10 @ $100

10 @ $60

10 @ $50

Load

The case file t/t_auction_case.m, used for this example, is a modified version of the 30-bus system
that has 9 generators, where the last three have negative PMIN to model the dispatchable loads.
To solve this case using an AC optimal power flow and a last accepted offer (LAO) pricing rule, we use
mkt.OPF = 'AC';
mkt.auction_type = 1;

and set up the problem as follows:

offers.P.qty = [ ...
12 24 24;
12 24 24;
12 24 24;
12 24 24;
12 24 24;
12 24 24 ];
offers.P.prc = [ ...
20 50 60;
20 40 70;
20 42 80;
20 44 90;
20 46 75;
20 48 60 ];

36

MATPOWER Users Manual

Version 3.1b1

bids.P.qty = [ ...
10 10 10;
10 10 10;
10 10 10 ];
bids.P.prc =
100 70
100 50
100 60

[ ...
60;
20;
50 ];

[mpc_out, co, cb, f, dispatch, success, et] = runmarket(mpc, offers, bids, mkt);

The resulting cleared offers and bids are:

>> co.P.qty
ans =
12.0000
12.0000
12.0000
12.0000
12.0000
12.0000

23.3156
24.0000
24.0000
24.0000
24.0000
24.0000

0
0
0
0
0
0

50.0000
50.2406
50.3368
51.0242
52.1697
52.9832

50.0000
50.2406
50.3368
51.0242
52.1697
52.9832

10.0000
0
10.0000

10.0000
0
0

51.8207
54.0312
55.6208

51.8207
54.0312
55.6208

>> co.P.prc
ans =
50.0000
50.2406
50.3368
51.0242
52.1697
52.9832
>> cb.P.qty
ans =
10.0000
10.0000
10.0000
>> cb.P.prc
ans =
51.8207
54.0312
55.6208

37

MATPOWER Users Manual

Version 3.1b1

In other words, the generators sold:

Generator Quantity Sold Selling Price
MW

$/MWh

35.3

$50.00

36

$50.24

36

$50.34

36

$51.02

36

$52.17

36

$52.98

And the dispatchable loads bought:

Load Quantity Bought Purchase Price
MW

$/MWh

30.0

$51.82

10.0

$54.03

20.0

$55.62

38