Linearized optimal power flow

1.0 Some introductory comments

The advantage of the economic dispatch formulation to obtain
minimum cost allocation of demand to the generation units is that
it is computationally very fast and reasonably easy to solve.
However, it suffers from two main drawbacks:
1. It is somewhat inaccurate due to loss approximation and due
to implicit assumption that transmission has infinite capacity;
2. It provides no information about line flows.
The optimal power flow (OPF) compensates for these drawbacks
by replacing the one power balance equation with the nodal power
flow equations. The generator limits are retained as inequality
constraints, but we add to the inequality constraints the constraints
on line flows.

We desire to bring network constraints into the problem in order to

study the influence of transportation (transmission) constraints on
the most economic distribution of generation. There are a number
of ways to solve the problem, including the following.
• Linearized
• Nonlinear, generalized reduced gradient
• Nonlinear, penalty function approach.
In these notes, we will focus on the linearized approach. It is very
attractive to do so because linear optimization problems may be
solved using linear programming (LP). LP has been under
development now for over 50 years, and extremely efficient
algorithms have been developed so that linear programs, even very
large ones, may be solved quickly. In addition, there are available
software packages which make LP solutions relatively easy to
develop. We will not have time to present LP in this course, but
there are courses that do (see IE 312 and IE 534). In addition, the
appendix of Chapter 6 in W&W overviews one method (but not
the simplest method and so not a good place to begin learning LP).

1
The OPF has been a tool for many years in the power engineering
community, and has been effective in solving the following kinds
of problems (see Section 13.1 of W&W):
• Minimizing dispatch cost while meeting transmission
constraints (this is the “classical” OPF)
• Security-constrained OPF (corrective and preventative)
• Voltage-var optimization
• Maximum power transfer
• Loss minimization

But it is the advent of electricity markets that has made OPF a

central, essential part of running power systems. There are two
comments to be made here:
• Most electricity markets utilize the linear programmed security-
constrained OPF (LP-SCOPF). We will focus only on the LP-
OPF in these notes. However, the main difference between LP-
SCOPF and LP-OPF is that security constraints associated with
branch overload are present in the former and not in the latter.
• Most electricity markets allow both offers (to sell) and bids (to
buy). We will allow only the former in these notes, but
introduction of the latter provides nothing conceptually new.

My treatment complements that of Section 13.4 in W&W.

In linearizing the OPF, to maintain simplicity, we will assume that

each generator makes only a single-price offer. Thus, each unit is
represented in the objective function by a constant times the
generation level for that unit. This approach (si), and an alternative
(si1, si2, si3), are illustrated in Fig. 0 below (similar to Fig. 13.6 in
W&W).

2
 Ci si3
$/hr si

si2
si1
Pi3
Pi1 Pi2

Pi,min Pi (MW) Æ

Fig. 0

So here is the formal statement of our problem:

min ∑s P } k gk
k ∈{generator _ buses
1

Subject to:
P = B' θ 2

P B = ( D × A) × θ 3

− P B ,max ≤ P B ≤ P B ,max 4

0 ≤ Pgk ≤ Pgk ,max , ∀k ∈ {generator _ buses} 5

where
Pk = Pgk − Pdk , k = 1,...N 6

1
k is index on the units.
2
These are DC power flow equations to represent the network. However, we must include all nodal
injections P1,…PN and all angles θ1 …θN in this set of equations.
3
These are equation to get line flows. Again, we need to include all angles θ1 …θN in this set of equations.
4
These are the limits on the line flows. Notice that there is only one circuit rating, but it must be enforced
as a limit if the flow is in one direction or in the other.
5
These are the limits on the linear cost curve variables.
6
This relates variables used in the cost curves (Pkj) to variables used in DC power flow equations (Pk).

3
Before we proceed with formulating this problem, we must make
one very important observation. In the DC power flow problem,
we found that if we included an equation for all buses in the
network, using all angles in the network, that there was a
dependency in our set of equations, and the matrix was not
invertable.

Now, however, we will include all N equations for our network

because we want a Lagrange multiplier for each bus, in order to
know the locational marginal price (LMP) of energy. This implies
that we must also include all of the angles as our unknowns. This
will not create the same problem of matrix singularity that we had
before because our overall system of equations will include (2),
(3), and (6). In fact, this system will be under-constrained (there
will be more unknowns than equations). This is acceptable because
• we are not trying to solve them (because they are
underconstrained, there are many solutions);
• rather, we are trying to minimize a function that is subject to
them.(with appropriate convexity requirements, there is only one
solution that will do this).

Therefore we will include in P=B’θ all DC power flow equations,

one for each node, as a function of all angles in the network.

So be aware that in all of what follows, the vector P includes the

reference bus injection, the vector θ includes the reference bus
angle, the matrix B’ is N×N (i.e., it does not have a row and
column eliminated corresponding to the reference bus), and
adjacency (node-arc) matrix D is M×N (i.e, we do not eliminate
the column corresponding to the reference bus).

To solve this problem as a linear program, we need to be able to

write it in a form that a standard LP solver will handle. This means
equality and inequality constraints must be written as a function of
a single vector of unknowns (the “solution vector”).

4
Special note on LP solver: The below formulation was developed
for use in Matlab. However, this formulation may also be
implemented in CPLEX. To do so, you need to realize that CPLEX
is quite flexible in handling input constraints. The below example
illustrates this flexibility:

Equality
constraints

Less-than
inequality
constraint

Greater-than
Less-than, inequality
greater-than constraint
inequality
constraints

You will find a great deal of additional information on how to use

CPLEX at www.iro.umontreal.ca/~gendron/IFT6551/CPLEX/HTML.

2.0 Formulation of the linearized OPF

We define the following solution vector for our problem as:

⎡Pg ⎤
x = ⎢⎢ P B ⎥⎥
⎢⎣ θ ⎥⎦ (1)

5
where
ƒ Pg is the vector of generation increments [Pgk …]T for all bus k
that has generation.
ƒ PB is the vector of line flows [Pb1 Pb2…PbM]T, M is # of
branches.
ƒ θ is the vector of bus angles, in radians [θ1 θ2…θN], N is # of
buses.

We want to capture all of our equality constraints in a single matrix

relation.

There are two kinds of equality constraints: one due to line flows

P B = ( D × A) × θ (2)
and the other due to injections:
P = B' θ (3)
We rewrite these slightly modified to make them more convenient
to embed in a single matrix equation.

− P B + ( D × A) × θ = 0 (4)

− P + B 'θ = 0 (5)
The inequality constraints are straightforward given our definition
of the solution vector. The only issue here is what to use as
constraints on the angles? Clearly, all angles must reside between –
π radians and +π radians. Therefore, inequality constraints will be:

⎡ P g ,min ⎤ ⎡ P g ⎤ ⎡ P g ⎤
⎢− P ⎥ ≤ ⎢P ⎥ ≤ ⎢P ⎥
⎢ B ,max ⎥ ⎢ B ⎥ ⎢ B ,max ⎥
⎢⎣ − π ⎥⎦ ⎢⎣ θ ⎥⎦ ⎢⎣ π ⎥⎦ (6)

6
Example:

We illustrate using an example that has 3 units connected to 3

different buses in a 4 bus network supplying load at 2 different
buses. This system is very similar to that of Fig. 13.1 in your text.

The one-line diagram for the example system is given in Fig. 1.

We will modify the load so that it has a total of 2.1787 per unit (or
217.87 MW), with 1 per unit load at bus 2 (Pd2=1.0) and 1.1787
per unit load at bus 3 (Pd3=1.1787).

Pg2
Pg1

1 2

y12 =-j10
y14 =-j10
y13 =-j10 Pd2=1pu y23 =-j10

y34 =-j10
4 3

Pg4 Pd3=1.1787pu

Fig. 1: One line diagram for example system

The offers and corresponding min and max generation limits are as
follows:
K1 ( Pg1 ) = s1 Pg1 , 50 ≤ Pg1 ≤ 200
K 2 ( Pg 2 ) = s2 Pg 2 , 37.5 ≤ Pg 2 ≤ 150
K 4 ( Pg 4 ) = s4 Pg 4 , 45 ≤ Pg 4 ≤ 180

7
with
s1=13.07 $/MWhr
s2=12.11 $/MWhr
s4=12.54 $/MWhr

A modification is necessary at this point in that we need to change

the generation variables to per-unit. This is necessary in order to
bring in the transmission equations, which are in per-unit. Thus,
the decision variables Pg1, Pg2, and Pg3, are all divided by 100. We
can compensate for this in terms of obtaining the correct evaluation
for the cost by multiplying the offers by 100. Thus, we will use:
s1=1307 $/puMWhr
s2=1211 $/puMWhr
s4=1254 $/puMWhr
Objective function: We explicitly write the solution vector; then we
will be able to write the objective function immediately.
⎡ Pg1 ⎤
⎢P ⎥
⎢ g2 ⎥
⎢ Pg 4 ⎥
⎢ ⎥
⎢ PB1 ⎥
⎢ PB 2 ⎥
⎡Pg ⎤ ⎢ ⎥
⎢P ⎥
x = ⎢⎢ P B ⎥⎥ = ⎢ B 3 ⎥
P
⎢⎣ θ ⎥⎦ ⎢ B 4 ⎥
⎢ PB 5 ⎥
⎢θ ⎥
⎢ 1⎥
⎢ θ2 ⎥
⎢ ⎥
⎢ θ3 ⎥
⎢⎣ θ 4 ⎥⎦

8
Now we see that the objective function is given by:
⎡ Pg 1 ⎤
⎢P ⎥
⎢ g2 ⎥
⎢ Pg 4 ⎥
⎢ ⎥
⎢ PB1 ⎥
⎢ PB 2 ⎥
⎢ ⎥
⎢P ⎥
Z ( x ) = [1307 1211 1254 0 0 0 0 0 0 0 0 0]⎢ B 3 ⎥
P
⎢ B4 ⎥
⎢ PB 5 ⎥
⎢θ ⎥
⎢ 1 ⎥
⎢ θ2 ⎥
⎢ ⎥
⎢ θ3 ⎥
⎢⎣ θ 4 ⎥⎦
Equality constraints: The equality constraints are given in eqs. (4)
and (5), repeated here for convenience:

− P B + ( D × A) × θ = 0 (4)

− P + B 'θ = 0 (5)
We will build all of these equality constraints into a matrix form of
Aeqx=beq. We begin by noting dimensions.
ƒ Columns: Since the solution vector x is 12×1, Aeq must have 12
columns in order to pre-multiply x.
ƒ Rows: Since there are 5 branches, eq. (4) will contribute 5 rows
to Aeq. Since there are 4 buses, eq. (5) will contribute 4 rows to
Aeq. So Aeq will have total of 9 rows.

Therefore, the dimensions of Aeq will be 9×12.

We begin with the line flow equations, eq. (4). The D matrix is:

9
 ⎡10 0 0 0 0 ⎤
⎢ 0 10 0 0 0 ⎥
⎢ ⎥
D = ⎢ 0 0 10 0 0 ⎥
⎢ ⎥
⎢ 0 0 0 10 0 ⎥
⎢⎣ 0 0 0 0 10⎥⎦
The node-arc incidence matrix, A, is:
⎡1 0 0 - 1⎤
⎢1 - 1 0 0 ⎥
⎢ ⎥
A = ⎢0 1 − 1 0 ⎥
⎢ ⎥
⎢ 0 0 − 1 1 ⎥
⎢⎣1 0 − 1 0 ⎥⎦
The D×A product required by eq. (4) is then given by:
⎡10 0 0 0 0 ⎤ ⎡1 0 0 - 1⎤ ⎡10 0 0 − 10⎤
⎢ 0 10 0 0 0 ⎥ ⎢1 - 1 0 0 ⎥ ⎢10 − 10 0 0 ⎥
⎢ ⎥⎢ ⎥ ⎢ ⎥
D × A = ⎢ 0 0 10 0 0 ⎥ ⎢0 1 − 1 0 ⎥ = ⎢ 0 10 − 10 0 ⎥
⎢ ⎥⎢ ⎥ ⎢ ⎥
⎢ 0 0 0 10 0 ⎥⎢ 0 0 − 1 1 ⎥ ⎢ 0 0 − 10 10 ⎥
⎢⎣ 0 0 0 0 10⎥⎦ ⎢⎣1 0 − 1 0 ⎥⎦ ⎢⎣10 0 − 10 0 ⎥⎦
So based on eq. (4) and the solution vector, we can see that these
elements will occupy the upper right hand corner of Aeq. So that
will take care of the last 4 columns in the first 5 rows.

But what about the first 8 columns? These are the elements in the
line flow equations that multiply the variables Pg1, Pg2, Pg4,
PB1, PB2, PB3, PB4, PB5. Since we do not use the generation variables
within the line flow equations, the first 3 columns of these top 5
rows will be zeros. The next 5 columns in these top 5 rows
(columns 4-8) will also be zeros, except the one element in each of

10
these rows that multiplies the corresponding line flow variable, and
that element will be -1.

Finally, with respect to these top 5 equations, eq. (4) indicates that
the right-hand-side will be 0 for each of them.

Thus, we can now write down all elements in the first 5 rows of
our matrix, as follows:
⎡ Pg1 ⎤
⎢P ⎥
− 10⎤ ⎢ ⎥ ⎡ 0 ⎤
g2
⎡0 0 0 − 1 0 0 0 0 10 0 0
⎢0 ⎢ Pg 4 ⎥
⎢ 0 0 0 − 1 0 0 0 10 − 10 0 0 ⎥⎥ ⎢ ⎥ ⎢⎢ 0 ⎥⎥
PB1
⎢0 0 0 0 0 − 1 0 0 0 10 − 10 0 ⎥⎢ ⎥ ⎢0⎥
⎢ ⎥ ⎢ PB 2 ⎥ ⎢ ⎥
⎢0 0 0 0 0 0 −1 0 0 0 − 10 10 ⎥ ⎢ ⎥ ⎢ 0 ⎥
⎢ PB 3 ⎥
Aeq x = ⎢ 0 0 0 0 0 0 0 − 1 10 0 − 10 0 ⎥⎢ ⎥ = ⎢0⎥
⎢ ⎥ PB 4 ⎢ ⎥
⎢_ _ _ _ _ _ _ _ _ _ _ _ ⎥⎢ ⎥ ⎢_ ⎥
⎢_ ⎢ PB 5 ⎥
_ _ _ _ _ _ _ _ _ _ _ ⎥⎢ ⎥ ⎢_ ⎥
⎢ ⎥ θ1 ⎢ ⎥
⎢_ _ _ _ _ _ _ _ _ _ _ _ ⎥⎢ ⎥ ⎢_ ⎥
⎢_ ⎢ θ2 ⎥
⎣ _ _ _ _ _ _ _ _ _ _ _ ⎥⎦ ⎢ ⎥ ⎢⎣ _ ⎥⎦
⎢ θ3 ⎥
⎢⎣ θ 4 ⎥⎦
Now we need to write the last 4 equations. These are the DC power
flow equations corresponding to eq. (5).

Again, we must remember that the solution vector contains all 4

angles, and therefore the DC power flow matrix needs to be a 4×4.

This augmented DC power flow matrix is given below:

11
 ⎡ 30 − 10 − 10 − 10⎤
⎢− 10 20 − 10 0 ⎥
B' = ⎢ ⎥
⎢− 10 − 10 30 − 10⎥
⎢ ⎥
⎣− 10 0 − 10 20 ⎦
So based on eq. (5) and the solution vector, we can see that this
matrix will occupy the lower right hand side of the Aeq matrix. So
that will take care of the last 4 columns in the bottom 4 rows. The
resulting matrix appears as:
⎡ Pg1 ⎤
⎢P ⎥
− 10⎤ ⎢ ⎥ ⎡ 0 ⎤
g2
⎡0 0 0 −1 0 0 0 0 10 0 0
⎢0 ⎢ Pg 4 ⎥
⎢ 0 0 0 −1 0 0 0 10 − 10 0 0 ⎥⎥ ⎢ ⎥ ⎢⎢ 0 ⎥⎥
PB1
⎢0 0 0 0 0 −1 0 0 0 10 − 10 0 ⎥⎢ ⎥ ⎢0⎥
⎢ ⎥ ⎢ PB 2 ⎥ ⎢ ⎥
⎢0 0 0 0 0 0 −1 0 0 0 − 10 10 ⎥ ⎢ ⎥ ⎢ 0 ⎥
⎢ PB 3 ⎥
Aeq x = ⎢ 0 0 0 0 0 0 0 −1 10 0 − 10 0 ⎥⎢ ⎥ = ⎢0⎥
⎢ ⎥ PB 4 ⎢ ⎥
⎢_ _ _ _ _ _ _ _ 30 − 10 − 10 − 10⎥ ⎢ ⎥ ⎢ _ ⎥
⎢_ ⎢ PB 5 ⎥
_ _ _ _ _ _ _ − 10 20 − 10 0 ⎥⎢ ⎥ ⎢_ ⎥
⎢ ⎥ θ1 ⎢ ⎥
⎢_ _ _ _ _ _ _ _ − 10 − 10 30 − 10⎥ ⎢ ⎥ ⎢ _ ⎥
⎢_ ⎢ θ2 ⎥
⎣ _ _ _ _ _ _ _ − 10 0 − 10 20 ⎥⎦ ⎢ ⎥ ⎢⎣ _ ⎥⎦
⎢ θ3 ⎥
⎢⎣ θ 4 ⎥⎦

Once again, we need to consider the first eight columns. Columns

4-8 correspond to the line flow variables, which do not appear in
the DC power flow equations, so these will be zero.

12
 ⎡ Pg1 ⎤
⎢P ⎥
− 10⎤ ⎢ ⎥ ⎡ 0 ⎤
g2
⎡0 0 0 −1 0 0 0 0 10 0 0
⎢0 ⎢ Pg 4 ⎥
⎢ 0 0 0 −1 0 0 0 10 − 10 0 0 ⎥⎥ ⎢ ⎥ ⎢⎢ 0 ⎥⎥
PB1
⎢0 0 0 0 0 −1 0 0 0 10 − 10 0 ⎥⎢ ⎥ ⎢0⎥
⎢ ⎥ ⎢ PB 2 ⎥ ⎢ ⎥
⎢0 0 0 0 0 0 −1 0 0 0 − 10 10 ⎥ ⎢ ⎥ ⎢ 0 ⎥
⎢ PB 3 ⎥
Aeq x = ⎢ 0 0 0 0 0 0 0 −1 10 0 − 10 0 ⎥⎢ ⎥ = ⎢0⎥
⎢ ⎥ PB 4 ⎢ ⎥
⎢_ _ _ 0 0 0 0 0 30 − 10 − 10 − 10⎥ ⎢ ⎥ ⎢ _ ⎥
⎢_ ⎢ PB 5 ⎥
_ _ 0 0 0 0 0 − 10 20 − 10 0 ⎥⎢ ⎥ ⎢_⎥
⎢ ⎥ θ1 ⎢ ⎥
⎢_ _ _ 0 0 0 0 0 − 10 − 10 30 − 10⎥ ⎢ ⎥ ⎢ _ ⎥
⎢_ ⎢ θ2 ⎥
⎣ _ _ 0 0 0 0 0 − 10 0 − 10 20 ⎥⎦ ⎢ ⎥ ⎢⎣ _ ⎥⎦
⎢ θ3 ⎥
⎢⎣ θ 4 ⎥⎦
The first three columns multiply the generation variables Pg1, Pg2,
and Pg4. However, the DC power flow equations, eq. (5), require
the negative of the injections for all buses, and the injections are
the generation minus the load, i.e., Pgk-Pdk.

We do not have load variables Pdk included in the solution vector.

In addition, we do not have generation for bus 3 (it is just a load
bus) included in the solution vector. So what do we do?

The answer to this lies in recognizing that the “variables” we do

not have in the solution vector, Pd1, Pd2, Pd3, Pd4, and Pg3, are not
(when the electricity market does not allow demand bids)
“variables” at all! In fact, they are known quantities, constants,
given by:
Pd1=0, Pd2=1.0, Pd3=1.1787, Pd4=0, Pg3=0
Since these are constants, they can go to the right-hand side.
But with what sign? Eq. (5)
− P + B 'θ = 0 (5)

13
indicates that the injection, if modeled on the left-hand-side,
should be negative, i.e., on the left-hand side, Pgk-Pdk should be
negative. So we should see on the left-hand-side –Pgk+Pdk. But now
we will take the load term onto the right-hand-side by subtracting
it from both sides.

Thus, we see that the load term should show up on the right-hand-
side as a negative number. This same logic also shows us that the
elements multiplying the generation terms in the Aeq matrix should
be -1.

So we are now prepared to complete the matrix relation for the

equality constraints.
⎡ Pg1 ⎤
⎢P ⎥
− 10⎤ ⎢ ⎥ ⎡ 0 ⎤
g2
⎡ 0 0 0 − 1 0 0 0 0 10 0 0
⎢ 0 0 0 0 − 1 0 0 0 10 − 10 0 ⎢ Pg 4 ⎥
⎢ 0 ⎥⎥ ⎢ ⎥ ⎢⎢ 0 ⎥⎥
PB1
⎢ 0 0 0 0 0 −1 0 0 0 10 − 10 0 ⎥⎢ ⎥ ⎢ 0 ⎥
⎢ ⎥ ⎢ PB 2 ⎥ ⎢ ⎥
⎢ 0 0 0 0 0 0 −1 0 0 0 − 10 10 ⎥ ⎢ ⎥ ⎢ 0 ⎥
⎢ PB 3 ⎥
Aeq x = ⎢ 0 0 0 0 0 0 0 − 1 10 0 − 10 0 ⎥⎢ ⎥ = ⎢ 0 ⎥
⎢ ⎥ PB 4 ⎢ ⎥
⎢− 1 0 0 0 0 0 0 0 30 − 10 − 10 − 10⎥ ⎢ ⎥ ⎢ 0 ⎥
⎢ 0 − 1 0 0 0 0 0 0 − 10 20 − 10 ⎢ PB 5 ⎥
0 ⎥⎢ ⎥ ⎢ −1 ⎥
⎢ ⎥ θ1 ⎢ ⎥
⎢ 0 0 0 0 0 0 0 0 − 10 − 10 30 − 10⎥ ⎢ ⎥ ⎢− 1.1787 ⎥
⎢ 0 0 − 1 0 0 0 0 0 − 10 0 ⎢ θ2 ⎥
⎣ − 10 20 ⎥⎦ ⎢ ⎥ ⎢⎣ 0 ⎥⎦
⎢ θ3 ⎥
⎢⎣ θ 4 ⎥⎦

Inequality constraints:

The inequality constraints are simple, as given in what follows:

14
 ⎡ 0.5 ⎤ ⎡ Pg1 ⎤ ⎡ 2 ⎤
⎢0.375⎥ ⎢ P ⎥ ⎢ 1.5 ⎥
⎢ ⎥ ⎢ g2 ⎥ ⎢ ⎥
⎢ 0.45 ⎥ ⎢ Pg 4 ⎥ ⎢ 1.8 ⎥
⎢ ⎥ ⎢ ⎥ ⎢ ⎥
⎢ − 500⎥ ⎢ PB1 ⎥ ⎢500⎥
⎢ − 500⎥ ⎢ PB 2 ⎥ ⎢500⎥
⎢ ⎥ ⎢ ⎥ ⎢ ⎥
⎢ − 500⎥ ≤ ⎢ PB 3 ⎥ ≤ ⎢500⎥
⎢ − 500⎥ ⎢ P ⎥ ⎢500⎥
⎢ ⎥ ⎢ B4 ⎥ ⎢ ⎥
⎢ − 500⎥ ⎢ PB 5 ⎥ ⎢500⎥
⎢ −π ⎥ ⎢ θ ⎥ ⎢ π ⎥
⎢ ⎥ ⎢ 1⎥ ⎢ ⎥
⎢ − π ⎥ ⎢ θ2 ⎥ ⎢ π ⎥
⎢ ⎥ ⎢ ⎥ ⎢ ⎥
⎢ − π ⎥ ⎢ θ3 ⎥ ⎢ π ⎥
⎢⎣ − π ⎥⎦ ⎢⎣ θ 4 ⎥⎦ ⎢⎣ π ⎥⎦
Solution by Matlab: The code for solving this linear program using
Matlab is given below:

15
%Load is system load plus losses
Load=2.1787;

%Build objective function vector.

c=[1307 1211 1254 0 0 0 0 0 0 0 0 0]';

%Build Aeq matrix for equality constraints.

Aeq=[0 0 0 -1 0 0 0 0 10 0 0 -10;
0 0 0 0 -1 0 0 0 10 -10 0 0;
0 0 0 0 0 -1 0 0 0 10 -10 0;
0 0 0 0 0 0 -1 0 0 0 -10 10;
0 0 0 0 0 0 0 -1 10 0 -10 0;
-1 0 0 0 0 0 0 0 30 -10 -10 -10
0 -1 0 0 0 0 0 0 -10 20 -10 0;
0 0 0 0 0 0 0 0 -10 -10 30 -10;
0 0 -1 0 0 0 0 0 -10 0 -10 20;];

%Build right-hand side of equality constraint.

beq=zeros(9,1);
beq(7)=-1;
beq(8)=-1.1787;

%Build upper and lower bounds on decision variables.

LB=[.50 .375 .45 -500 -500 -500 -500 -500 -pi -pi -pi -pi]';
UB=[2.00 1.50 1.80 500 500 500 500 500 pi pi pi pi]';
[X,FVAL,EXITFLAG,OUTPUT,LAMBDA]=LINPROG(c,A,b,Aeq,beq,LB,UB);

16
The solution vector x is given by:

⎡ Pg1 ⎤ ⎡ 0.5 ⎤
⎢ P ⎥ ⎢ 1.2287 ⎥
⎢ g2 ⎥ ⎢ ⎥
⎢ Pg 4 ⎥ ⎢ 0.45 ⎥
⎢ ⎥ ⎢ ⎥
P
⎢ B1 ⎥ ⎢ − 0. 0152 ⎥
⎢ PB 2 ⎥ ⎢ 0.0955 ⎥
⎢ ⎥ ⎢ ⎥
⎢ ⎥ 0.3242 ⎥
x = ⎢ B 3 ⎥ = ⎢⎢
P
PB 4 0.4348 ⎥
⎢ ⎥ ⎢ ⎥
⎢ PB 5 ⎥ ⎢ 0.4197 ⎥
⎢ θ ⎥ ⎢ 0.0125 ⎥ , Z=2705.8
⎢ 1⎥ ⎢ ⎥
⎢ θ 2 ⎥ ⎢ 0.003 ⎥
⎢ ⎥ ⎢ ⎥
θ
⎢ ⎥ ⎢3 − 0 . 0295 ⎥
⎢⎣ θ 4 ⎥⎦ ⎢⎣ 0.0140 ⎥⎦

The solution is provided on the one-line diagram of Fig. 4.

Pg2=1.2287pu
Pg1=0.5pu

1 2

PB2=0.0955
PB3
PB1 = =0.3242
-0.0152 PB5=0.4197 Pd2=1pu

PB4 =
4 0.4348 3

Pg4=0.45pu Pd3=1.1787pu

Fig. 4: Result in terms of generation levels and flows

17
One can easily check to see that the power is conserved at the
buses.

The objective function that Matlab provides, is given above as

Z=2705.8 $/hr.

It is of interest to compare this solution with a solution obtained

from an economic dispatch (implying no representation of
transmission). This problem will be
min 1307 Pg1 + 1211Pg 2 + 1254Pg 4
Subject to:
Pg1 + Pg 2 + Pg 4 = 2.1787
0.5 ≤ Pg1 ≤ 2.0
0.375 ≤ Pg 2 ≤ 1.5
0.45 ≤ Pg 4 ≤ 1.8
This is a LP that we can use Matlab or CPLEX to solve; however,
it is actually a rather trivial solution. You can see that Pg1 and Pg4
are the more expensive units, and so we will take as little of those
units as possible. Therefore we will assume
Pg1=0.5
Pg4=0.45
This means that Pg2= 2.1787-0.95=1.2287. The fact that this value
of Pg2 is feasible (between its limits) means that this must be the
answer to the problem, since this is the maximum amount of the
cheapest unit that we can have.

Compare this to the solution we obtained in our LPOPF, and you

see it is the same!!!

18
Should we expect the solutions to be the same, given the network
representation in the LOPF approach? The answer is NO, if the
flows are at the branch capacities.

But if the flows are all within the branch capacities (and we are not
representing losses), then the transmission system has no impact on
the dispatch, and in this case, the solution will be identical to the
solution that we obtained in the economic dispatch scenario.

The LPOPF solution is for the case where all flows are within their
branch capacities, since we have modeled all branch capacities to
be 500 pu. In per-unit, this is effectively infinite branch capacity.

BASE CASE:
Now we will investigate the Lagrange multipliers for this loading
level, assuming infinite capacity lines. These Lagrange multipliers,
which are the same as the dual variables, are given in Table 2.

Table 2: Lagrange multipliers for Pd2=1.0, Pd3=1.1787

and infinite transmission capacity ($/per unit-hr)
Equality constraints Lower bounds Upper bounds
3
Equation Value*10 Variable value variable value
PB1 -0.0000 Pg1 96.0000 Pg1 0.0000
PB2 -0.0000 Pg2 0 Pg2 0.0000
PB3 0.0000 Pg4 43.0000 Pg4 0.0000
PB4 -0.0000 PB1 0 PB1 0.0000
PB5 -0.0000 PB2 0 PB2 0.0000
P1 1.2110 PB3 0 PB3 0.0000
P2 1.2110 PB4 0 PB4 0.0000
P3 1.2110 PB5 0 PB5 0.0000
P4 1.2110 θ1 0 θ1 0.0000
θ2 0 θ2 0.0000
θ3 0 θ3 0.0000
θ4 0 θ4 0.0000

19
Lagrange multipliers on the last four equality constraints are very
interesting to us, since they give the improvement in the objective
function if we increase the right-hand-side of the corresponding
equation by 1 unit. These are the so-called nodal prices, i.e., the
LMPs, given in $/per unit-hr. We see that the numbers are all
$1211/per unit-hr, and if we divide this by the power base of 100
MVA, we get $12.11/MW-hr.

We also see that all Lagrange multipliers on the lower bounds are
all zero, with the exception of the ones on Pg1 and Pg4, which are
96 and 43, respectively, with units of $/per unit-hr. Converting to
$/MW-hr, these values are 0.96 and 0.43, respectively, indicating
the amount of improvement we can expect if we increase the
corresponding right-hand-side of these inequalities by 1 unit. Since
these equations look like, for example, -Pg1<-0.5, an increase in the
right-hand-side corresponds to a decrease in the lower limit. Thus,
for Pg1, if we move the lower limit from 50 MW to 49 MW, we can
expect to improve the objective function by 96 cents per hour.

Lagrange multipliers for all other lower bounds, and for all upper
bounds, are zero, since none of the other inequality constraints are
binding.

CASE 1: Pd2=1.01, Pd3=1.1787, and infinite transmission capacity

We now make a slight modification, by changing the loading of

bus 2 from 1.0 to 1.01, an increase of .01 per unit or 1 MW. The
changes necessary to the Matlab code are:
%Build right-hand side of equality constraint. It will be vector of zeros
%except for element in first row, which is load-sum of minimum
generation
beq=zeros(9,1);
beq(7)=-1.01;
beq(8)=-1.1787;
The solution is almost the same as in the base case, except for the
objective function now evaluates to Z=2717.9. The previous

20
objective function was Z=2705.8, an increase of 12.1 $/hr,
confirming our understanding of the meaning of the nodal prices.

We find that a 1 MW increase in the loading anywhere in this

network has the same effect on the objective function, to increase it
by 12.1$/hr. Thus, nodal prices have no locational variation in this
network. Any network with infinite transmission capacity will
have this attribute (assuming no losses), since the transmission
system effectively makes the entire system look like a single bus.

Case 2: Pd2=1.0, Pd3=1.1787, and 0.3 capacity constraint on PB3

In this case, we maintain loading at the same levels as the base
case, where we saw the flows as in Fig. 5:
Pg2=1.2287pu
Pg1=0.5pu

1 2

PB2=0.0955
PB3
PB1 = =0.3242
-0.0152 PB5=0.4197 Pd2=1pu

PB4 =
4 0.4348 3

Pg4=0.45pu Pd3=1.1787pu

Fig. 5: Base case flows

We observe the flow on branch 3, PB3=0.342. So let’s consider that
this branch has capacity of 0.3. This means that upper and lower
bounds on PB3 should be changed from -500,500 to -0.3, 0.3. The
resulting solution is given in Fig. 6:

21
 Pg2=1.1803pu
Pg1=0.5pu

1 2

PB2=0.1197
PB3 =0.3
PB1 =
-0.0393 PB5=0.4197 Pd2=1pu

PB4 =
4 0.4590 3

Pg4=0.4984pu Pd3=1.1787pu

Fig. 6: Cases 2 flows

In comparing the previous two diagrams, we observe that
ƒ The flow on branch 3 is constrained to 0.3 as desired.
ƒ The flows all over the network have changed.
ƒ The generation levels at buses 2 and 4 have changed.
Thus, the activation of a transmission constraint has changed the
dispatch. This will affect the energy prices! This conclusion can be
verified by looking at the Lagrange multipliers, given in the table
below.

22
 Table 3: Lagrange multipliers for Pd2=1.0, Pd3=1.1787
and infinite transmission capacity except for and
0.3 capacity constraint on PB3 ($/per unit-hr)
Equality constraints Lower bounds Upper bounds
3
Equation Value*10 Variable value variable value
PB1 -0.0000 Pg1 63.7500 Pg1 0.0000
PB2 -0.0000 Pg2 0.0000 Pg2 0.0000
PB3 0.0860 Pg4 0.0000 Pg4 0.0000
PB4 0.0000 PB1 0.0000 PB1 0.0000
PB5 -0.0000 PB2 0.0000 PB2 0.0000
P1 1.2432 PB3 0.0000 PB3 86.0000
P2 1.2110 PB4 0.0000 PB4 0.0000
P3 1.2647 PB5 0.0000 PB5 0.0000
P4 1.2540 θ1 0.0000 θ1 0.0000
θ2 0.0000 θ2 0.0000
θ3 0.0000 θ3 0.0000
θ4 0.0000 θ4 0.0000

Some comments about the Lagrange multipliers in Table 3:

1. Generation limits: We still see a non-zero Lagrange multiplier
on the Pg1 lower limit, as before, but the Lagrange multiplier on
the Pg4 lower limit has become zero, reflecting that Pg4 had to
increase and come off of its lower limit to compensate for the
decrease in Pg2 necessary to redispatch around the PB3
constraint.
2. Branch limits: The Lagrange multiplier on the PB3 upper bound
is 86, and after dividing by 100 to change from per-unit to MW,
it is 0.86 $/MW-hr, reflecting the improvement in objective
function that can be obtained from increasing the PB3 branch
limit by 1 MW (from 0.30 per-unit to 0.31 per-unit).
3. Branch flows: The Lagrange multiplier on the PB3 flow (the PB3
equality constraint) is 0.086*103=86, and after dividing by 100
to change from per-unit to MW, it is 0.86 $/MW-hr, reflecting
the improvement in objective function that can be obtained from

23
 increasing the flow in this branch by 1 MW (from 0.30 per-unit
to 0.31 per-unit). It makes sense that the Lagrange multiplier on
the flow would be the same as the Lagrange multiplier on the
branch limit.
4. Nodal prices: The Lagrange multipliers on the equality
constraints corresponding to the 4 nodes are the nodal prices.
Without transmission constraints, these prices were all the same,
at 12.11 $/MW-hr, a price set entirely by the generator at bus 2
since it was the bus 2 generator that responded to any load
change. But now they are all different, with only bus 2 price at
12.11 $/MW-hr. This difference reflects that, because of the
transmission constraint, a load increase at one bus will incur a
different cost than a load increase at another bus.

Comment #4 is worth investigating further. Let’s increase the load

at the highest price bus, bus #3, from 1.1787 to 1.1887 per unit, an
increase of 1 MW. The resulting dispatch and flows are shown in
Fig. 7. In order to gain intuition into what has happened, we have
repeated Fig. 6 just below it so as to provide a convenient
comparison.
Pg2=1.1778pu
Pg1=0.5pu

1 2

PB2=0.1222
PB3 =0.3
PB1 =
-0.0443 PB5=0.4222 Pd2=1pu

PB4 =
4 0.4665 3

Pg4=0.5109pu Pd3=1.1887pu

Fig. 7: Flows for case 2 with 1 MW increase in Pd3

24
 Pg2=1.1803pu
Pg1=0.5pu

1 2

PB2=0.1197
PB3 =0.3
PB1 =
-0.0393 PB5=0.4197 Pd2=1pu

PB4 =
4 0.4590 3

Pg4=0.4984pu Pd3=1.1787pu

Fig. 6: Case 1 flows

The comparison shows that in order to supply an additional MW at

bus 3, the generation levels of 2 different units had to be modified.
Specifically, Unit 2 was decreased from 1.1803 to 1.1778, a
decrease of 0.0025 per unit (0.25 MW) and Unit 4 was increased
from 0.4984 per unit to 0.5109 per unit, an increase of 0.0125
(1.25 MW). Thus, Unit 4 was increased enough to supply the
increased load at bus 3 and the decreased generation at bus 2.

Question: Why did we not just increase Unit 2 or increase Unit 4

by 1 MW?

Answer: Because the resulting flow on branch 3 would exceed its

capacity!!!

In fact, it is not possible to supply additional load at bus 3 with

only a single unit increase. We will always have to compensate for
the load AND redispatch to compensate for the additional flow on
the branch 3. As a result, the nodal price at bus 3 is a function of
the generation costs at those buses that are used in the particular
redispatch that achieves the minimum cost.

25