Course notes for EE394V

Restructured Electricity Markets:

Locational Marginal Pricing
Ross Baldick
Copyright c 2013 Ross Baldick
www.ece.utexas.edu/baldick/classes/394V/EE394V.html
Title Page 1 of 101 Go Back Full Screen Close Quit
3
Power ow
(i) Review of power concepts,
(ii) Formulation of power ow,
(iii) Problem characteristics and solution,
(iv) Linearized power ow,
(v) Fixed voltage schedule,
(vi) Line ow,
(vii) Direct Current (DC) power ow,
Title Page 2 of 101 Go Back Full Screen Close Quit
(viii) Example,
(ix) DC power ow circuit interpretation,
(x) Homework exercises.
Title Page 3 of 101 Go Back Full Screen Close Quit
3.1 Review of power concepts
Power is the rate of doing work, measured in W, kW, MW, or GW.
Energy is the work accomplished over time, measured in Wh, kWh,
MWh, or GWh.
When power varies over time, the energy is the integral over time of the
power.
Title Page 4 of 101 Go Back Full Screen Close Quit
Review of power concepts, continued
In principle, power could be generated, transmitted, and consumed using
either direct current (DC) or alternating current (AC).
Cost-effective and low loss transmission of bulk power relies on being
able to create high voltages:
power capability is proportional to the product of current and voltage,
so higher voltages allow for higher power levels,
for a given power, higher voltage means lower current, which implies
lower losses for a given resistance of conductor.
Generation and consumption is more convenient at lower voltages:
we will mostly focus on generation and transmission, modeling
consumption through aggregated loads at distribution substations,
probability of generation, transmission, and distribution failures, effects
due to presence of other loads, and other issues affect the quality of
supply to end users,
local effects on distribution system typically affect quality of supply
more noticeably than generation and transmission failures.
Title Page 5 of 101 Go Back Full Screen Close Quit
Review of power concepts, continued
Until the advent of power electronics, only AC power could easily be
transformed from one voltage to another:
basic reason for ubiquity of AC power systems.
AC transmission of power also involves the back-and-forth ow of power
between electric and magnetic elds:
this back-and-forth ow is called reactive power,
to distinguish reactive power from the power that can actually be
consumed by a load, the latter is called real power.
Title Page 6 of 101 Go Back Full Screen Close Quit
Review of power concepts, continued
The relationship between voltage and current in a circuit is determined by
the characteristics of the circuit elements and Kirchhoffs laws.
Kirchhoffs current law:
due to conservation of charge passing a bus or node of circuit,
implies that supply of electric power always equals demand of electric
power plus losses,
mis-match between mechanical and electrical power is smoothed by
inertia of system and results in frequency change,
enforcing supplydemand balance between mechanical power and
electrical power is different to enforcing supplydemand balance in
typical markets, such as a market for apartments to be described in
Section 6.
Kirchhoffs voltage law:
sum of voltages around loop is zero,
electric transmission network behaves differently to most other
transportation networks.
Title Page 7 of 101 Go Back Full Screen Close Quit
Review of power concepts, continued
Kirchhoffs laws implicitly determine the voltages and currents due to the
real and reactive power injections at the generators and the withdrawals at
the loads.
The problem of using Kirchhoffs laws to solve for the voltages and
currents in a circuit consisting of generators, the transmission and/or
distribution system, and loads is called the power ow problem:
assumes a particular operating condition,
quasi-static assumption that ignores dynamics and changes.
The solution provides information about the ow of current and power on
the transmission and distribution lines.
The lines have limited capacities, so calculation of power ow enables us
to decide whether or not a particular pattern of generation would result in
acceptable ows on lines:
constraints on transmission operation implicitly determine limitations
on the patterns of injections and withdrawals,
the locational marginal pricing market reects these limitations into
prices that vary by bus (or node).
Title Page 8 of 101 Go Back Full Screen Close Quit
3.2 Formulation of power ow
3.2.1 Variables
3.2.1.1 Phasors
We can use complex numbers, called phasors, to represent the magnitude
and angle of the AC voltages and currents at a xed frequency.
The magnitude of the complex number represents the root-mean-square
magnitude of the voltage or current.
The angle of the complex number represents the angular displacement
between the sinusoidal voltage or current and a reference sinusoid.
3.2.1.2 Reference angle
The angles of the voltages and currents in the system would all change if
we changed the angle of our reference sinusoid, but this would have no
effect on the physical system.
We can therefore arbitrarily assign the angle at one of the buses to be zero
and measure all the other angles with respect to this angle.
We call this bus the reference bus or the angle reference bus, bus , and
typically number the buses so that = 0 or = 1.
Title Page 9 of 101 Go Back Full Screen Close Quit
3.2.1.3 Representation of complex numbers
To represent a complex number V with real numbers requires two real
numbers, either:
the magnitude |V| and the angle

V, so that V =|V| exp(

1), or
the real {V} and imaginary {V} parts, so that
V ={V}+{V}

1.
Since we need to compare voltage magnitudes to limits to check
satisfaction of voltage limit constraints, we will represent voltages as
magnitudes and angles:
Some recent developments in power ow have used the real and
imaginary parts representation.
Title Page 10 of 101 Go Back Full Screen Close Quit
3.2.1.4 Scaling and per unit
There are voltage transformers throughout a typical power system:
step-up voltage at a generator to transmission voltages to enable
transfer from generator to transmission system,
transform from one transmission voltage to another,
step-down voltage at a distribution substation and in distribution
feeder for convenient use by load.
The nominal voltage magnitude varies considerably across the system by
several orders of magnitude.
We scale the voltage magnitude so that an actual value of 121 kV in the
110 kV part of the system would be represented by a scaled value of:
121 kV
110 kV
= 1.1,
while an actual value of 688.5 kV in the 765 kV part of the system would
be represented by a scaled value of:
688.5 kV
765 kV
= 0.9.
Title Page 11 of 101 Go Back Full Screen Close Quit
3.2.2 Symmetry
3.2.2.1 Three-phase circuits
Generation-transmission systems are usually operated as balanced
three-phase systems, with generators, lines, and (roughly) distribution
system loads arranged as symmetric triplets.
generator
transmission
line load
t
Z
n
n

neutral

b
Z
p
Z
L
Z
L
Z
L
n

a Z
p
@
@

@
@
c
Z
p
@
@
@
@

@
@

@
@

@
@

@
@
@
@

Fig. 3.1. An example

balanced three-phase
system.
Title Page 12 of 101 Go Back Full Screen Close Quit
3.2.2.2 Per-phase equivalent
The behavior of a balanced three-phase circuit can be completely
determined from the behavior of a per-phase equivalent circuit.
Figure 3.2 shows the a-phase equivalent circuit of the three-phase circuit
of Figure 3.1.
generator
transmission
line load
neutral

Z
p
Z
L
Fig. 3.2. Per-phase
equivalent circuit for
the three-phase circuit
in Figure 3.1.
Title Page 13 of 101 Go Back Full Screen Close Quit
3.2.3 Transmission lines
Transmission lines are physically extended objects, so the boxes in
Figures 3.1 and 3.2 are actually distributed parameter circuits,
We can represent the terminal behavior of such distributed parameter
circuits with a -equivalent circuit.
Each component of the -equivalent has an impedance (or, equivalently,
an admittance) determined by the characteristics of the line.
neutral
k
u
u
s
h
u
n
t
series
u
u
s
h
u
n
t
Fig. 3.3. Equivalent
circuit of per-phase
equivalent of transmis-
sion line.
Title Page 14 of 101 Go Back Full Screen Close Quit
3.2.4 Bus admittance matrix and power ow equations
Consider the per-phase equivalent of a three bus, three line transmission
system as illustrated in Figure 3.4.
For each bus = 1, 2, 3, the pair of shunt elements joining node to
neutral can be combined together to form a single shunt element.
neutral
1 2
3
t
t
t
t
t
t
t
t
t
t
t
t
Fig. 3.4. Per-phase
equivalent circuit model
for three bus, three line
system.
Title Page 15 of 101 Go Back Full Screen Close Quit
Bus admittance matrix and power ow equations, continued
This yields a circuit with:
one element corresponding to each of the buses = 1, 2, 3, joining node
to neutral, and
one element corresponding to each line,
as illustrated in Figure 3.5.
neutral
1 2
3
t
Y
1
Y
13
t
t
Y
3
Y
23
t
Y
2
Y
12
Fig. 3.5. Per-phase
equivalent circuit model
for three bus, three line
system with parallel
components combined.
Title Page 16 of 101 Go Back Full Screen Close Quit
Bus admittance matrix and power ow equations, continued
Let us write Y

for the admittance of the element joining node to neutral,

and
Y
k
for the admittance of the series element corresponding to a line joining
buses and k.
The series element is most easily characterized in terms of its impedance.
For a series impedance Z
k
= R
k
+X
k

1 between buses and k, the

corresponding admittance Y
k
is given by:
Y
k
=
1
Z
k
,
=
1
R
k
+X
k

1
=
1
R
k
+X
k

R
k
X
k

1
R
k
X
k

1
=
R
k
X
k

1
(R
k
)
2
+(X
k
)
2
. (3.1)
If Z
k
= 0.1+

1, what is Y
k
= 1/Z
k
?
Title Page 17 of 101 Go Back Full Screen Close Quit
Bus admittance matrix and power ow equations, continued
Let V be the vector of phasor voltages at all the buses in the system and
let I be the vector of phasor current injections into the transmission
network at all of the buses in the system.
Using Kirchhoffs laws, we can obtain a relationship of the form AV = I
between current and voltage, where:
, k, A
k
=
_
_
_
Y

J()
Y
k
, if = k,
Y
k
, if k J() or J(k),
0, otherwise,
(3.2)
where J() is the set of buses joined directly by a transmission line to bus
.
The linear simultaneous equations AV = I represent conservation of
current at each of the buses.
A is called the bus admittance matrix:
the -th diagonal entry is the sum of the admittances connected to bus ,
the k-th off-diagonal entry is minus the admittance connecting bus
and k.
Title Page 18 of 101 Go Back Full Screen Close Quit
3.2.5 Generators and loads
When electricity is bought and sold, the (real) power and energy are the
quantities that are usually priced, not the voltage or current.
However, real power does not completely describe the interaction
between generators or loads and the system.
We also have to characterize the injected reactive power.
We can combine the real and reactive powers into the complex power,
which is the sum of:
the real power, and

1 times the reactive power.

Title Page 19 of 101 Go Back Full Screen Close Quit
Generators and loads, continued
The usefulness of this representation is that, for example, the net complex
power S

injected at node into the network is given by:

S

=V

,
where the superscript indicates complex conjugate:
note difference between complex conjugate, denoted superscript , and
optimal or desired value, denoted superscript .
The current I

equals the sum of:

the current owing into the shunt element Y

, and
the sum of the currents owing into each line connecting to a bus
k J() through admittance Y
k
.
We can substitute for the currents to obtain:
S

= V

[A

+

kJ()
A
k
V
k
]

,
= |V

|
2
A

+

kJ()
A

k
V

k
. (3.3)
Title Page 20 of 101 Go Back Full Screen Close Quit
Generators and loads, continued
Let A
k
= G
k
+B
k

1, , k, where we note that by (3.1) and (3.2):

we have that G
k
< 0 and B
k
> 0 for = k, and
we have that G

> 0 and the sign of B

is indeterminate but typically

less than zero;
let S

= P

+Q

1, , with:
for generator buses, P

> 0 and Q

is typically positive,
for load buses, P

< 0 and Q

< 0;
and let V

= u

exp(

1), , with:
the voltage magnitude u

1 in scaled units to satisfy voltage limits,

the voltage angle

typically between /4 and /4 radians.

Sometimes we will explicitly distinguish the real power injected by a
generator from the real power consumed by a load, by writing D

for the
real power load at bus :
the net real power injection at a bus with generation P

and load D

is
then P

, with both P

and D

typically positive.
Similarly, we will write E

for the reactive power load at bus , so that

the net reactive power injection is Q

.
Title Page 21 of 101 Go Back Full Screen Close Quit
Generators and loads, continued
For notational convenience in the following development, we will write
P

and Q

for the net real and reactive injections:

later cases where we explicitly distinguish generation from load will be
clear from context.
We can separate (3.3) into real and imaginary parts:
P

=

kJ(){}
u

u
k
[G
k
cos(

k
) +B
k
sin(

k
)], (3.4)
Q

=

kJ(){}
u

u
k
[G
k
sin(

k
) B
k
cos(

k
)]. (3.5)
The equations (3.4) and (3.5), which are called the power ow equality
constraints, must be satised at each bus .
That is, there are two constraints that must be satised at each bus.
For a 5000 bus system, how many power ow equality constraints must
be satised?
Title Page 22 of 101 Go Back Full Screen Close Quit
3.2.6 The power ow problem
The power ow problem is to nd values of voltage angles and
magnitudes that satisfy the power ow equality constraints.
3.2.6.1 Real and reactive power balance
For convenience, we will say PQ bus for a bus where the real and reactive
power injection is specied.
We specify:
the real and reactive generations at the PQ generator buses according to
the generator control settings, and
the (typically negative) real and reactive net power injections at the PQ
load buses according to supplied data.
At each such bus, we have two specied parameters (the real and reactive
power injection) and two unknowns that are entries in the decision vector,
the voltage magnitude and angle.
However, we cannot arbitrarily specify the real and reactive power at all
the buses since this would typically violate the rst law of
thermodynamics!
Not all the buses can be PQ buses.
Title Page 23 of 101 Go Back Full Screen Close Quit
3.2.6.2 Slack bus
A traditional, but ad hoc approach to nding a solution to the equations is
to single out a slack bus, bus .
At this slack bus, instead of specifying injected real and reactive power,
there is assumed to be a generator that produces whatever is needed to
balance the real and reactive power for the rest of the system, assuming
that such a solution exists.
Typically, the slack bus is the same as the reference bus, but this is not
necessarily the case, and the slack can even be (conceptually)
distributed across multiple buses.
For reasons that will become clear in the context of locational marginal
pricing, we also call the slack bus the price reference bus.
Title Page 24 of 101 Go Back Full Screen Close Quit
Slack bus, continued
We re-interpret P

and Q

to be decision variables in our power ow

formulation and calculate them to satisfy the real and reactive power
balance in the system:
for reasons that will become clear in the next section, we will not have
to represent P

and Q

explicitly in the decision vector x,

in Section 5 in the context of economic dispatch where we are
considering the choice of generation at all the buses, we will also
consider the other real and reactive generations to be decision variables
and so P

and Q

together with all the other real and reactive

generations will be explicitly in the decision vector x.
Title Page 25 of 101 Go Back Full Screen Close Quit
Slack bus, continued
The generator at the slack bus supplies whatever power is necessary for
real and reactive power balance.
To keep the number of unknowns equal to the number of equations, the
voltage magnitude at the slack bus is specied as any particular value:
in Section 9 in the context of optimal power ow we will re-interpret
the voltage magnitude at the slack bus to also be part of the decision
vector.
If the reference bus and the slack bus are the same bus, then we can call it
a V bus, since both the voltage magnitude and angle are specied.
At the V bus, we still have two specied parameters (the voltage
magnitude and angle) and two unknowns (the real and reactive power
injections).
For most of the rest of the development of power ow, we will typically
assume that the reference bus and the slack bus are the same and typically
number the buses so that the reference/slack bus is bus 1:
we will sketch how to consider the case where the reference and slack
buses are different.
Title Page 26 of 101 Go Back Full Screen Close Quit
3.2.7 Non-linear equations
We have n
PQ
PQ buses, including both the PQ generators and the loads.
Let n = 2n
PQ
and dene a decision vector x R
n
consisting of the voltage
magnitudes and angles at the PQ buses:
unknown real P

and reactive Q

generation at the slack bus will be

evaluated in terms of x and so not represented explicitly in the decision
vector.
For every bus (that is, including the slack bus as well as the PQ buses)
dene functions p

: R
n
R and q

: R
n
R by:
x R
n
, p

(x) =

kJ(){}
u

u
k
[G
k
cos(

k
) +B
k
sin(

k
)], (3.6)
x R
n
, q

(x) =

kJ(){}
u

u
k
[G
k
sin(

k
) B
k
cos(

k
)]. (3.7)
The functions p

and q

represent the real and reactive power ow,

respectively, from bus into the lines in the rest of the system.
Kirchhoffs laws require that the net real and reactive ow out of a bus
must be zero, so that p

(x) P

= 0 and q

(x) Q

= 0 at every bus .
Title Page 27 of 101 Go Back Full Screen Close Quit
Non-linear equations, continued
For convenience, let = = 1 and dene a vector function g : R
n
R
n
with entries given by p

and q

for all the PQ buses:

x R
n
, g(x) =
_

_
p
2
(x) P
2
p
3
(x) P
3
.
.
.
q
2
(x) Q
2
q
3
(x) Q
3
.
.
.
_

_
.
If we solve the non-linear simultaneous equations:
g(x) = 0, (3.8)
for x

and then set:

P
1
= p
1
(x

),
Q
1
= q
1
(x

),
then we will have satised the power ow equality constraints at all buses
including the slack bus.
Title Page 28 of 101 Go Back Full Screen Close Quit
Non-linear equations, continued
We have formulated the power ow problem as the solution of non-linear
simultaneous equations:
g(x) = 0.
The vector g includes real and reactive power entries for each bus except
the reference/slack bus:
we will calculate the real and reactive power injection at the slack bus
after we have solved g(x) = 0.
The vector x includes voltage angles and magnitudes for each bus except
the reference/slack bus:
the voltage angle and magnitude for the reference/slack bus are
specied.
Recall that as we develop other problems, we will re-dene x and g as
needed for the formulation.
Title Page 29 of 101 Go Back Full Screen Close Quit
Non-linear equations, continued
For future notational convenience:
Let , u, p, q, P, and Q be vectors consisting, respectively, of the entries

, u

, p

, q

, P

, and Q

for all the buses.

We will often need to refer to a sub-vector with a particular entry omitted:
let subscript k on a vector denote that vector with the entry k omitted,
so
k
, u
k
, p
k
, q
k
, P
k
, and Q
k
are, respectively, the sub-vectors of ,
u, p, q, P, and Q with the entries
k
, u
k
, p
k
, q
k
, P
k
, and Q
k
omitted.
We will maintain these denitions of , u, p, q, P, and Q throughout the
course:
recall that we will change the denition of x and g depending on the
particular problem being formulated.
Also, let subscript k on a matrix denote that matrix with row k omitted:
so A
k
is the admittance matrix with the k-th row omitted.
Title Page 30 of 101 Go Back Full Screen Close Quit
Non-linear equations, continued
With = = 1 the reference/slack bus then, x =
_

1
u
1
_
and
g =
_
(p
1
) P
1
(q
1
) Q
1
_
.
With this notation, g(x) = 0 can also be expressed in the equivalent form:
p
1
(x) = p
1
__

1
u
1
__
= P
1
,
q
1
(x) = q
1
__

1
u
1
__
= Q
1
.
For a 5000 bus system, how many entries are in , u, p and q? How about
in
1
, u
1
, p
1
, q
1
, x, and g?
Title Page 31 of 101 Go Back Full Screen Close Quit
Non-linear equations, continued
In summary, to solve Kirchhoffs equations for the electric power
network, we:
(i) solve (3.8), g(x) = 0, which is a system of non-linear
simultaneous equations, and
(ii) substitute the solution x

into (3.4) and (3.5) for the slack bus

= to nd the real and reactive power generated at the slack bus.
The real power generation at the slack bus is P

= p

(x

), so x

also
satises p(x

) = P and, moreover:
1

P = 1

p(x

).
This expression evaluates the total losses in the system, since it sums the
total net real power injected into the transmission lines.
Line currents and real and reactive power ows can also be calculated
once x is known.
Title Page 32 of 101 Go Back Full Screen Close Quit
3.2.8 Example
For example, for a three bus system with buses = 1, 2, 3 and bus 1 the
reference/slack bus, the entries of x R
4
and g : R
4
R
4
would be:
x =
_

1
u
1
_
=
_

3
u
2
u
3
_

_
,
x R
4
, g(x) =
_
p
1
(x) P
1
q
1
(x) Q
1
_
=
_

_
p
2
(x) P
2
p
3
(x) P
3
q
2
(x) Q
2
q
3
(x) Q
3
_

_
,
where P
2
is the net generation (generation minus demand) at bus 2, and
similarly for other buses and for the reactive power at the buses.
Title Page 33 of 101 Go Back Full Screen Close Quit
Example, continued
If we solve g(x) = 0, we can then use the resulting solution x

to evaluate
the real power and reactive power that must be produced at the
reference/slack bus to satisfy real and reactive power balance at every bus:
P
1
= p
1
(x

),
Q
1
= q
1
(x

).
Losses in the system are given by:
1

p(x

) = P
1
+P
2
+P
3
.
Title Page 34 of 101 Go Back Full Screen Close Quit
3.3 Problem characteristics and solution
3.3.1 Number of variables and equations
There are the same number of variables as equations in (3.8).
For a 5000 bus system, with one V bus and the rest PQ buses, how many
variables and equations are there?
3.3.2 Non-existence of direct algorithms
Because the equations are non-linear, there is no direct algorithm, such as
factorization, to solve for the solution x

for arbitrary systems.

The NewtonRaphson algorithm from Section 2.3 can be applied to this
problem, requiring:
an initial guess x
(0)
,
evaluation of partial derivative terms in the Jacobian,
g
x
, and
solution of the NewtonRaphson update (2.9)(2.10) at each iteration.
Title Page 35 of 101 Go Back Full Screen Close Quit
3.3.3 Number of solutions
There may be no solutions, one solution, or even multiple solutions
to (3.8).
However, power systems are usually designed and operated so that the
voltage magnitudes are near to nominal and the voltage angles are
relatively close to 0

.
If we restrict our attention to solutions such that voltage magnitudes are
all close to 1 (and make some other assumptions) then we can nd
conditions for the there to be at most one solution.
How many solutions are there to 2+sin() = 0?
How many solutions are there to 0.1+sin() = 0?
How many solutions are there to 0.1+sin() = 0 with /4 /4?
Title Page 36 of 101 Go Back Full Screen Close Quit
3.3.4 Admittance matrix
3.3.4.1 Symmetry
The admittance matrix A is symmetric.
3.3.4.2 Sparsity
The matrix A is only sparsely populated with non-zero entries and each
component of g depends on only a few components of x.
Sparsity is the key to practical solution of problems with large numbers of
buses.
For a 5000 bus system having 5000 lines, how many non-zero entries are
there in the admittance matrix A?
Title Page 37 of 101 Go Back Full Screen Close Quit
3.3.4.3 Values
A typical line impedance has positive real and imaginary parts.
The corresponding line admittance Y
k
therefore has positive real part and
negative imaginary part.
If there is a line between buses and k then the entries
A
k
= G
k
+

1B
k
in the admittance matrix satisfy G
k
< 0, B
k
> 0.
The diagonal entries A

= G

1B

in the admittance satisfy

G

> 0 and, typically, B

< 0.
The resistance R
k
of transmission lines is relatively small compared to
the inductive reactance X
k
.
Furthermore, the shunt elements are often also negligible compared to the
inductive reactance.
This means that:
, k J() {}, |G
k
| |B
k
|.
Title Page 38 of 101 Go Back Full Screen Close Quit
3.4 Linearized power ow
3.4.1 Base-case
Suppose that we are given values of real and reactive generation
P
(0)
R
n
PQ
+1
and Q
(0)
R
n
PQ
+1
that specify a base-case.
For example, the base-case real and reactive generations could be the
current operating conditions.
As another example, P
(0)
= 0 is the (unrealistic) condition of zero net
real power injection.
Also suppose that we have a solution x

to the base-case equations.

That is, g(x

) = 0, or equivalently:
p
1
(x

) = P
(0)
1
,
q
1
(x

) = Q
(0)
1
,
where P
(0)
1
and Q
(0)
1
are the sub-vectors of P
(0)
and Q
(0)
, respectively, that
omit the reference/slack bus.
Title Page 39 of 101 Go Back Full Screen Close Quit
3.4.2 Change-case
Now suppose that the real and reactive power generations change:
from P
(0)
and Q
(0)
,
to P = P
(0)
+P and Q = Q
(0)
+Q, respectively.
Similarly, we suppose that the value of x changes from x

to x

+x to
re-establish satisfaction of the power ow equations g(x) = 0.
That is, the change-case power ow equations are given by:
p
1
(x

+x) = P
(0)
1
+P
1
,
q
1
(x

+x) = Q
(0)
1
+Q
1
,
where P
1
and Q
1
are the sub-vectors of P and Q, respectively, that
omit the reference/slack bus.
The equations are non-linear equations in x.
Title Page 40 of 101 Go Back Full Screen Close Quit
Change-case, continued
Note the change in net generation at the reference/slack bus is required to
be consistent with the change x.
So, we also have that:
p
1
(x

+x) = P

1
+P
1
,
q
1
(x

+x) = Q

1
+Q
1
.
That is, the change in generation at the reference/slack bus can be
calculated (or estimated) once x is known or estimated.
Title Page 41 of 101 Go Back Full Screen Close Quit
3.4.3 First-order Taylor approximation
To nd an approximate solution to the change-case equations, we form
rst-order Taylor approximations to p
1
and q
1
:
p
1
(x

+x) p
1
(x

) +
p
1
x
(x

)x,
q
1
(x

+x) q
1
(x

) +
q
1
x
(x

)x.
For future reference, note that the matrices
p
1
x
(x

) and
q
1
x
(x

) form
the Jacobian of the system of equations p
1
(x) = P
1
, q
1
(x) = Q
1
:
_

_
p
1
x
q
1
x
_

_
.
Title Page 42 of 101 Go Back Full Screen Close Quit
3.4.4 Linearization of change-case equations
Substituting the rst-order Taylor approximations into the change-case
equations, we obtain:
p
1
(x

) +
p
1
x
(x

)x P
(0)
1
+P
1
,
q
1
(x

) +
q
1
x
(x

)x Q
(0)
1
+Q
1
.
From the base-case solution, we have p
1
(x

) = P
(0)
1
and q
1
(x

) = Q
(0)
1
.
Ignoring the error in the rst-order Taylor approximation, we have:
p
1
x
(x

)x = P
1
,
q
1
x
(x

)x = Q
1
.
Title Page 43 of 101 Go Back Full Screen Close Quit
Linearization of change-case equations, continued
Typically, the Jacobian
_

_
p
1
x
(x

)
q
1
x
(x

)
_

_
is non-singular.
That is, we can solve:
_

_
p
1
x
(x

)
q
1
x
(x

)
_

_
x =
_
P
1
Q
1
_
,
for x.
These are sparse linear equations, which can be solved efciently for x.
This approximation to the solution of the change-case power ow
equations is equivalent to performing one iteration of the
NewtonRaphson method, starting at the base-case specied by x

.
For a 5000 bus system, what is the size of the coefcient matrix of the
linear equations?
Title Page 44 of 101 Go Back Full Screen Close Quit
Linearization of change-case equations, continued
Moreover, the change in real and reactive power at the reference/slack bus
will approximately satisfy:
P
1
=
p
1
x
(x

)x,
Q
1
=
q
1
x
(x

)x.
Title Page 45 of 101 Go Back Full Screen Close Quit
3.4.5 Jacobian
3.4.5.1 Terms
Recall that the entries in p : R
n
R
n
PQ
+1
are dened by:
x R
n
, p

(x) =

kJ(){}
u

u
k
[G
k
cos(

k
) +B
k
sin(

k
)].
The entries in q : R
n
R
n
PQ
+1
are dened by: q

: R
n
R:
x R
n
, q

(x) =

kJ(){}
u

u
k
[G
k
sin(

k
) B
k
cos(

k
)].
The entries in the vector x are either of the form
k
or of the form u
k
.
To examine the terms in the Jacobian, partition x so that all the voltage
angles appear rst in a sub-vector
1
followed by all the voltage
magnitudes in a sub-vector u
1
.
Title Page 46 of 101 Go Back Full Screen Close Quit
Terms, continued
There are four qualitative types of partial derivative terms corresponding
to each combination:
x R
n
,
p

k
(x)
=
_

jJ()
u

u
j
[G
j
sin(

j
) +B
j
cos(

j
)], if k = ,
u

u
k
[G
k
sin(

k
) B
k
cos(

k
)], if k J(),
0, otherwise,
x R
n
,
p

u
k
(x)
=
_

_
2u

+

jJ()
u
j
[G
j
cos(

j
) +B
j
sin(

j
)], if k = ,
u

[G
k
cos(

k
) +B
k
sin(

k
)], if k J(),
0, otherwise,
Title Page 47 of 101 Go Back Full Screen Close Quit
Terms, continued
x R
n
,
q

k
(x)
=
_

jJ()
u

u
j
[G
j
cos(

j
) +B
j
sin(

j
)], if k = ,
u

u
k
[G
k
cos(

k
) B
k
sin(

k
)], if k J(),
0, otherwise,
x R
n
,
q

u
k
(x)
=
_

_
2u

+

jJ()
u
j
[G
j
sin(

j
) B
j
cos(

j
)], if k = ,
u

[G
k
sin(

k
) B
k
cos(

k
)], if k J(),
0, otherwise.
Title Page 48 of 101 Go Back Full Screen Close Quit
3.4.5.2 Partitioning by types of terms
Based on the partitioning of x, we can partition the Jacobian into four
blocks:
_

_
p
1

1
(x)
p
1
u
1
(x)
q
1

1
(x)
q
1
u
1
(x)
_

_
.
For a 5000 bus system, with 5000 lines, how many non-zero entries are
there in each of these blocks?
Title Page 49 of 101 Go Back Full Screen Close Quit
3.4.6 Decoupled equations
Recall that for typical lines , k J() {}, |G
k
| |B
k
|.
Also note that for typical lines k, |

k
| /2.
This implies that the terms in the matrices
p
1
u
1
and
q
1

1
are small
compared to the terms in the matrices
p
1

1
and
q
1
u
1
.
If we neglect all the terms in
p
1
u
1
and
q
1

1
, then we can then
approximate the Jacobian by
_

_
p
1

1
(x) 0
0
q
1
u
1
(x)
_

_
.
Title Page 50 of 101 Go Back Full Screen Close Quit
Decoupled equations, continued
Letting x =
_

1
u
1
_
, this allows decoupling of the linearized equations
into:
p
1

1
(x

)
1
= P
1
,
q
1
u
1
(x

)u
1
= Q
1
,
The rst set of equations relate real power and angles, while the second
set of equations relate reactive power and voltage magnitudes.
These decoupled equations require less computation than solving the full
system.
For a 5000 bus system, what is the size of the coefcient matrix of each of
the the decoupled linear equations?
Title Page 51 of 101 Go Back Full Screen Close Quit
3.5 Fixed voltage schedule
If real power generations and ows are our main concern and there is
adequate voltage support in the form of controllable reactive sources
then we may be justied in assuming that the voltage magnitudes can be
held xed by controlling reactive power:
instead of each bus except the reference/slack bus being a PQ bus, we
re-interpret them as having a specied real power and voltage
magnitude.
These are called PV buses.
A typical assumption is that all voltage magnitudes are 1 per unit, u = 1.
More generally, any xed voltage schedule u
(0)
can be used.
Title Page 52 of 101 Go Back Full Screen Close Quit
Fixed voltage schedule, continued
Assuming all buses, except the reference/slack bus, are PV buses:
The unknowns are: the voltage angles at all the buses except the
reference/slack bus; the real power generation at the reference/slack
bus; and the reactive power generations at all buses.
We rst solve p
1
__

1
u
(0)
1
__
= P
1
for

1
, given the xed voltage
schedule u
(0)
.
To complete the solution:
real power at the reference/slack bus, P

1
is chosen to satisfy
P

1
= p
1
__

1
u
(0)
1
__
, and
reactive generations at all buses, including the reference/slack bus, Q

are chosen to satisfy Q

= q
__

1
u
(0)
1
__
, in order to achieve the voltage
schedule u
(0)
.
Title Page 53 of 101 Go Back Full Screen Close Quit
3.6 DC power ow
We combine the ideas of xed voltage prole and linearization.
3.6.1 Fixed voltage schedule
We again assume that there are controllable voltage sources available to
provide a xed voltage schedule u
(0)
.
Based on the analysis in the previous section, we could rst solve
p
1
__

1
u
(0)
1
__
= P
1
for

1
and then evaluate P

1
= p
1
__

1
u
(0)
1
__
.
This again enables us to focus on real power generation and angles.
However, instead of solving p
1
__

1
u
(0)
1
__
= P
1
exactly for

1
, we
solve a linearized version that is linearized about a base-case in order to
estimate a change-case solution.
Title Page 54 of 101 Go Back Full Screen Close Quit
3.6.2 Linearization
We linearize about a xed base-case solution, x
(0)
=
_

(0)
1
u
(0)
1
_
.
The change-case power ow equations are given by:
p
1
__

(0)
1
+
1
u
(0)
1
__
= P
(0)
1
+P
1
.
To nd an approximate solution to the change-case equations, we form a
rst-order Taylor approximation to p
1
:
p
1
__

(0)
1
+
1
u
(0)
1
__
p
1
__

(0)
1
u
(0)
1
__
+
p
1

1
__

(0)
1
u
(0)
1
__

1
.
Substituting the rst-order Taylor approximations into the change-case
equations, we obtain:
p
1
__

(0)
1
u
(0)
1
__
+
p
1

1
__

(0)
1
u
(0)
1
__

1
P
(0)
1
+P
1
.
Title Page 55 of 101 Go Back Full Screen Close Quit
Linearization, continued
From the base-case solution, we have p
1
__

(0)
1
u
(0)
1
__
= P
(0)
1
.
Ignoring the error in the rst-order Taylor approximation, we have:
p
1

1
__

(0)
1
u
(0)
1
__

1
=P
1
,
which can be solved for
1
.
The change in the real power generation at the slack bus is then
approximately:
P
1
=
p
1

1
__

(0)
1
u
(0)
1
__

1
.
Title Page 56 of 101 Go Back Full Screen Close Quit
Linearization, continued
The base-case power generations P
(0)
that determine the base-case
solution are chosen to be convenient for calculations.
A typical base-case involves:
zero net real power generation at all buses, so that P
(0)
= 0, and
all voltage magnitudes 1 per unit, so that u
(0)
= 1.
If the transmission lines have zero real values for their shunt elements
then
(0)
= 0 solves the base-case.
x
(0)
=
_

(0)
1
u
(0)
1
_
=
_
0
1
_
is called a at start.
At the at start, the linearization yields the following equations:
p
1

1
(x
(0)
)
1
=P
1
.
Title Page 57 of 101 Go Back Full Screen Close Quit
Linearization, continued
To summarize, we have linearized about the at start condition to
approximate the change-case solution =
(0)
+ = 0+ =
corresponding to injections P = P
(0)
+P = 0+P =P.
We now interpret:
P
(0)
+P =P = P to be the power generation for the change-case we
are trying to solve, and

(0)
+ = = to be the solution for the angles for the change-case
we are trying to solve.
That is, we solve the linearized power ow equations for
1
:
p
1

1
__
0
1
__

1
= P
1
,
where
p
1

1
__
0
1
__
is a constant matrix,

1
is the vector of unknown angles at the change-case solution, and
P
1
is the sub-vector of P that omits the slack bus.
Title Page 58 of 101 Go Back Full Screen Close Quit
Linearization, continued
These equations are in the form J
1

1
= P
1
, where the coefcient
matrix is:
J
1
=
p
1

1
__
0
1
__
.
The coefcient matrix J
1
relates real power and angles.
The subscript 1 is referring to bus 1 as the slack bus:
for the general case of bus as the slack bus, we will consider
J

=
p

1
__
0
1
__
,
where the reference bus is still bus 1.
The equations J
1

1
= P
1
are sparse linear equations, which can be
solved efciently for
1
.
Paralleling the earlier observation, this approximation to the solution of
the power ow equations for power generation P
1
is equivalent to
performing one iteration of the NewtonRaphson method, starting at a at
start.
Title Page 59 of 101 Go Back Full Screen Close Quit
Linearization, continued
Moreover, the real power at the slack bus for the change-case can be
estimated by:
P
1
=
p
1

1
__
0
1
__

1
, (3.9)
or P

=
p

1
__
0
1
__

1
if bus is the slack bus.
We will see that the estimation of the real power generation at the slack
bus can be simplied under certain assumptions on the base-case system.
Letting J =
p

1
__
0
1
__
, note that the approximation satises J
1
= P,
which are called the DC power ow equations.
Values of
1
and P that satisfy the DC power ow equations then
approximately satisfy the power ow equality constraints (3.4) for all
buses .
Recall that we have assumed that Q is chosen to satisfy the power ow
equality constraints (3.5) for all buses .
Title Page 60 of 101 Go Back Full Screen Close Quit
3.6.3 Interpretation
We have interpreted the DC power ow approximation as equivalent to
performing one iteration of the NewtonRaphson method, starting at a
base-case specied by a at start, or equivalently the power ow
equations linearized about a at start.
This differs from the traditional interpretation that emphasizes:
the small angle approximations for cos and sin, and
the solution of DC power ow being the same as the solution of an
analogous DC circuit with current sources specied by the power
injections and voltages specied by the angles.
Our interpretation provides a clearer and more general perspective on the
conditions when the DC power ow provides a good approximation:
see homework.
It also provides a connection to decomposition algorithms:
iteration between solution and linearization of the power ow equations,
and calculation of a desired generation operating point.
The traditional interpretation is useful for solving small systems by hand:
See in Section 3.9.
Title Page 61 of 101 Go Back Full Screen Close Quit
3.6.4 Terms in Jacobian
The entries for the sub-matrix J
1
=
p
1

1
__
0
1
__
of the Jacobian are:
p

k
__
0
1
__
=
_

jJ()
B
j
, if k = ,
B
k
, if k J(),
0, otherwise,
(3.10)
=
_

jJ()
_
minus the susceptance
joining buses and j
_
, if k = ,
B
k
, if k J(),
0, otherwise,
=
_
B
k
, if k J() {},
0, otherwise,
_
if the shunt susceptances
are all equal to zero.
Note that these entries correspond to the imaginary part of the admittance
matrix, B, where A = G+B

1, not to the inverse of the line inductive

reactances, as is often stated in derivations of the DC power ow:
these are different if the resistance is non-zero.
Title Page 62 of 101 Go Back Full Screen Close Quit
Terms in Jacobian, continued
In the next slides, we will consider entries of J =
p

1
__
0
1
__
, which
includes a row consisting of the derivatives of p
1
corresponding to the
slack bus.
If the shunt admittances are all equal to zero then:
J is minus the imaginary part of the admittance matrix, that is, B, with
the column corresponding to the reference bus deleted,
J
1
is minus the imaginary part of the admittance matrix, that is, B,
with the column corresponding to the reference bus deleted and the
row corresponding to the slack bus deleted, and
if the slack bus is bus , then J

is minus the imaginary part of the

admittance matrix, that is, B, with the column corresponding to
the reference bus deleted and the row corresponding to the slack bus
deleted.
If the shunt admittances are non-zero then the entries of J corresponding
to diagonal entries B

of B will differ from the entries of B by the shunt

admittance connected to bus .
Title Page 63 of 101 Go Back Full Screen Close Quit
Terms in Jacobian, continued
Summing the entries in the -th column of
p

1
__
0
1
__
, we obtain:
k,

k
__
0
1
__
=
p
k

k
__
0
1
__
+

=k
p

k
__
0
1
__
,
where the summation over includes
the slack bus,
=

jJ(k)
B
k j

J(k)
B
k
, general case from (3.10),
considering possibly non-zero shunt admittances,
=

jJ(k)
B
k j

J(k)
B
k
, since B
k
= B
k
,
= 0.
That is, each column of J =
p

1
__
0
1
__
sums to zero.
Title Page 64 of 101 Go Back Full Screen Close Quit
3.6.5 Slack injection
Equivalently,
1

1
__
0
1
__
= 0. (3.11)
From (3.9), we can estimate the net injection at the slack bus as:
P
1
=
p
1

1
__
0
1
__

1
,
= 1

p
1

1
__
0
1
__

1
,
since 1

1
__
0
1
__
= 0 from (3.11),
= 1

P
1
,
given the DC power ow approximation.
Title Page 65 of 101 Go Back Full Screen Close Quit
3.6.6 Losses
Moreover, the losses in the system are estimated as:
1

P = 1

_
P
1
P
1
_
,
= 1

_
p
1

1
__
0
1
__

1
p
1

1
__
0
1
__

1
_

_
,
= 1

1
__
0
1
__

1
,
= 0
1
, by (3.11),
= 0,
so that the linearized representation is lossless, given the DC power ow
assumption of the at start as base-case:
note that this applies whether or not the shunt admittances are non-zero.
Title Page 66 of 101 Go Back Full Screen Close Quit
3.6.7 Solving for the angles
In the usual case that the matrix J
1
=
p
1

1
__
0
1
__
is non-singular, we
can write
1
= [J
1
]
1
P
1
, allowing us to solve for the angles
1
.
(If J
1
is singular, then the system is at a steady-state stability limit.)
Given the lossless assumption and a specication of P
1
, we have already
evaluated the net generation at the reference/slack bus:
P
1
=1

P
1
,
so that the reference/slack bus exactly compensates for the net demand or
withdrawal summed across all other buses, since the approximation is
lossless.
Summarizing, the DC power ow equations J
1
= P are equivalent to:
P
1
= 1

P
1
,

1
= [J
1
]
1
P
1
.
See Exercise 3.4 for a formal demonstration of the equivalence of the DC
power ow equations to this representation.
Title Page 67 of 101 Go Back Full Screen Close Quit
3.6.8 Demand
So far, the vector P has represented the vector of net injections at the
buses.
In some formulations, we want to consider demand and generation
separately.
For example, if the net injection is PD, where:
P is now the vector of generations, and
D is the vector of demands,
then the DC power ow equations are equivalent to:
1

P = 1

D,

1
= [J
1
]
1
(P
1
D
1
),
where P
1
and D
1
are the sub-vectors of P and D, respectively, that omit
the reference/slack bus.
Title Page 68 of 101 Go Back Full Screen Close Quit
3.6.9 Slack bus and reference bus choices
If the slack bus is bus then the DC power ow equations are equivalent
to:
1

P = 1

D,

1
= [J

]
1
(P

),
where P

and D

are the sub-vectors of P and D, respectively, that omit

the slack bus, and
where the reference bus is still assumed to be bus 1.
If the slack bus is bus and the reference bus is bus then the DC power
ow equations are equivalent to:
1

P = 1

D,

=
_
J

1
(P

),
where J

=
p

__
0
1
__
is the matrix of partial derivatives with the
reference bus assumed to be bus and where the matrix J

differs from J
in one column.
Title Page 69 of 101 Go Back Full Screen Close Quit
3.7 Line ow
We typically use the results of power ow to evaluate whether the ow
along a line is within limits:
this is most straightforward for ow limits expressed in terms of real
power ow,
we can also consider ow limits expressed in terms of current
magnitude or the magnitude of complex power.
There is typically a ow limit in each direction on the line.
For a line joining bus to bus k we can consider:
real and reactive ows p
k
and q
k
along the line from bus in the
direction of bus k, and
real and reactive ows p
k
and q
k
along the line from bus k in the
direction of bus .
Without loss of generality, we explicitly consider only p
k
and q
k
.
Title Page 70 of 101 Go Back Full Screen Close Quit
Line ow, continued
Ignoring shunt elements in the models, we have that the real and reactive
ows are given by:
x R
n
, p
k
(x) = u

u
k
[G
k
cos(

k
) +B
k
sin(

k
)] (u

)
2
G
k
,
x R
n
, q
k
(x) = u

u
k
[G
k
sin(

k
) B
k
cos(

k
)] +(u

)
2
B
k
.
(The linearization analysis including shunt elements has the same result
that we will present, but is notationally inconvenient since we need to
dene parameters for the shunt elements in each line.)
We will approximate these expressions by again linearizing about a
base-case:
for convenience, we will again assume = = 1 in the derivation and
then sketch the extensions to the general case.
We linearize the expressions for p
k
and q
k
about
(0)
1
.
We continue to assume that the voltage magnitudes are xed at u
(0)
.
Title Page 71 of 101 Go Back Full Screen Close Quit
3.7.1 Linearized line ow
p
k
__

(0)
1
+
1
u
(0)
1
__
p
k
__

(0)
1
u
(0)
1
__
+
p
k

1
__

(0)
1
u
(0)
1
__

1
,
= p
k
__

(0)
1
u
(0)
1
__
+u
(0)

u
(0)
k
_
G
k
sin(
(0)

(0)
k
)
+B
k
cos(
(0)

(0)
k
)
_
(

k
),
q
k
__

(0)
1
+
1
u
(0)
1
__
q
k
__

(0)
1
u
(0)
1
__
+
q
k

1
__

(0)
1
u
(0)
1
__

1
,
= q
k
__

(0)
1
u
(0)
1
__
+u
(0)

u
(0)
k
_
G
k
cos(
(0)

(0)
k
)
+B
k
sin(
(0)

(0)
k
)
_
(

k
).
Title Page 72 of 101 Go Back Full Screen Close Quit
Linearized Line ow, continued
We focus on the real power p
k
owing along the line from bus in the
direction of bus k.
Dene the row vector K
(k)
of partial derivatives by:
j =1, K
(k) j
=
_

_
u
(0)

u
(0)
k
[G
k
sin(
(0)

(0)
k
) +B
k
cos(
(0)

(0)
k
)],
if j = ,
u
(0)

u
(0)
k
[G
k
sin(
(0)

(0)
k
) +B
k
cos(
(0)

(0)
k
)],
if j = k,
0, otherwise,
That is, K
(k) j
is the j-th entry in the row vector K
(k)
, which has entries
for every bus except the reference bus.
Then the linear approximation to p
k
is given by:
p
k
__

(0)
1
+
1
u
(0)
1
__
p
k
__

(0)
1
u
(0)
1
__
+K
(k)

1
.
Title Page 73 of 101 Go Back Full Screen Close Quit
3.7.2 Line ow constraints
Suppose that we have line ow constraints of the form p
k
(x) p
k
.
Using the linear approximation, we obtain:
p
k
__

(0)
1
u
(0)
1
__
+K
(k)

1
p
k
.
We now consider the case that there are line ow constraints on each line
(k).
By dening a matrix K with rows K
(k)
and a vector d with entries d
(k)
of
the form:
d
(k)
= p
k
p
k
__

(0)
1
u
(0)
1
__
,
we can approximate the collection of line ow constraints in the form
K
1
d.
Title Page 74 of 101 Go Back Full Screen Close Quit
3.7.3 DC power ow approximation to line ow constraints
Using a at start x
(0)
=
_

(0)
1
u
(0)
1
_
=
_
0
1
_
as the base-case for the
linearization, we nd:
p
k
__
0
u
(0)
1
__
= u
(0)

(u
(0)
k
u
(0)

)G
k
,
p
k
__
0
1
__
= 0,
K
(k) j
=
p
k

j
__
0
u
(0)
1
__
=
_
u

u
k
B
k
, if j = ,
u

u
k
B
k
, if j = k,
0, otherwise,
K
(k) j
=
p
k

j
__
0
1
__
=
_
B
k
, if j = ,
B
k
, if j = k,
0, otherwise,
d
(k)
= p
k
p
k
__
0
1
__
,
= p
k
.
Title Page 75 of 101 Go Back Full Screen Close Quit
DC power ow approximation to line ow constraints, continued
Summarizing, we can approximate the ows at the angle

1
=
(0)
1
+
1
=
1
using the linearized equations K
1
d, where:
(k), j = 1, K
(k) j
=
_
B
k
, if j = ,
B
k
, if j = k,
0, otherwise,
(k), d
(k)
= p
k
.
If the reference bus changes to bus then the linearized line ow
constraints would be K

d, where:
(k), j =, K

(k) j
=
_
B
k
, if j = ,
B
k
, if j = k,
0, otherwise,
(k), d
(k)
= p
k
.
The matrix K

differs from the matrix K in one column.

Title Page 76 of 101 Go Back Full Screen Close Quit
3.7.4 Eliminating the angles
We previously found that the power ow equations could be expressed as:
1

P = 1

D,

1
= [J

]
1
(P

),
where is the slack bus and = 1 is the reference bus.
We use the second equation to substitute into K
1
d to obtain the
equality and inequality constraints with the angles eliminated:
1

P = 1

D, (3.12)
K[J

]
1
P

K[J

]
1
D

+d. (3.13)
This approximation to the ows is not always good:
it is used to represent transmission constraints in most day-ahead
electricity markets,
will explore accuracy in homework exercise.
Title Page 77 of 101 Go Back Full Screen Close Quit
3.7.5 Shift factor matrix
The matrix K[J

]
1
is the matrix of DC shift factors.
That is, entries in the matrix represent the fraction of ow along each line
for:
injection at the buses represented in the vector P

, and
withdrawal at the slack bus .
We occasionally want to express line ows in terms of the vector P of all
net injections.
For = 1, dene the augmented shift factor matrix

C =
_
0 K[J
1
]
1

.
That is,

C consists of the columns of K[J
1
]
1
augmented by an additional
zero column corresponding to P
1
.
Each entry of

C
k
represents the fraction of the generation from generator
at bus k that ows on the corresponding line.
The ows are given by

C(PD).
Similarly, if the slack bus is some other bus , we can again dene a
corresponding augmented shift factor matrix

C such that the ows are
given by

C(PD).
Title Page 78 of 101 Go Back Full Screen Close Quit
3.8 Example
Consider the following system with MW capacities and per unit
impedances (on a 1 MVA base) as shown.
Bus = 0 is the slack bus and there are no shunt admittances.
Bus = 1 is reference bus, so the unknown angles are
1
=
_

3
_
.

@
@
@
@
@
@
@
@

@
@
@
@
@
@
@
@

@
@
@
@
@
@
@
@

@
@
@
@
@
@
@
@

@
@
@
@
@
@
@
@

P
1
P
2
D
0
P
3
1 2
0 3
0+0.001

1
3000 MW
0+0.001

1
3000 MW
0+0.002

1
300 MW
0+0.001

1
3000 MW
@
@
@R

@
@
@I

? ?
Fig. 3.6. Four-line
four-bus network with
generators at buses 1, 2,
and 3, and demand at
bus 0.
Title Page 79 of 101 Go Back Full Screen Close Quit
3.8.1 Admittance matrix
The line admittances are:
Y
01
=Y
10
=Y
12
=Y
21
=Y
03
=Y
30
=
1
0+0.001

1
=1000

1,
Y
23
=Y
32
=
1
0+0.002

1
=500

1.
Title Page 80 of 101 Go Back Full Screen Close Quit
Admittance matrix, continued
The bus admittance matrix is:
_

_
Y
01
+Y
03
Y
01
0 Y
03
Y
10
Y
10
+Y
12
Y
12
0
0 Y
21
Y
21
+Y
23
Y
23
Y
30
0 Y
32
Y
30
+Y
32
_

_
=
_

_
2000

1 1000

1 0 1000

1
1000

1 2000

1 1000

1 0
0 1000

1 1500

1 500

1
1000

1 0 500

1 1500

1
_

_
,
=
_

_
B
00

1 B
01

1 0 B
03

1
B
10

1 B
11

1 B
12

1 0
0 B
21

1 B
22

1 B
23

1
B
30

1 0 B
32

1 B
33

1
_

_
.
Title Page 81 of 101 Go Back Full Screen Close Quit
3.8.2 Jacobian
Since there are no shunts, J =
p

1
__
0
1
__
is minus the imaginary part
of the admittance matrix, that is, B, with the column corresponding to
the reference bus deleted:
J =
p

1
__
0
1
__
=
_

_
B
00
0 B
03
B
10
B
12
0
0 B
22
B
23
B
30
B
32
B
33
_

_
,
=
_

_
2000 0 1000
1000 1000 0
0 1500 500
1000 500 1500
_

_
.
Note that the rows of J are indexed by 0, 1, 2, 3, while the columns are
indexed by 0, 2, 3.
Title Page 82 of 101 Go Back Full Screen Close Quit
3.8.3 DC power ow
We can solve for
1
to obtain the following form for the DC power ow
equations:
1

P = 1

D,

1
= [J
0
]
1
(P
0
D
0
).
where J
0
is J with the row corresponding to = 0 deleted, so that:
J
0
=
_
1000 1000 0
0 1500 500
1000 500 1500
_
,
[J
0
]
1
=
_
0.0008 0.0006 0.0002
0.0002 0.0006 0.0002
0.0006 0.0002 0.0006
_
.
Note that the subscript 1 on
1
is referring to = 1, the reference bus,
with the entry
1
omitted, (and columns of J also omit terms for
1
),
whereas the subscript 0 on J
0
, P
0
, and D
0
is referring to = 0, the
slack bus, with row
p
0

1
, and terms P
0
and D
0
, respectively, omitted.
Title Page 83 of 101 Go Back Full Screen Close Quit
DC power ow, continued
So far, the development considered generation and demand at all buses.
The example only has demand at bus 0 and has generation at buses 1, 2,
and 3.
Since there is only demand at bus 0 then the DC power ow equations are:
P
1
P
2
P
3
= D
0
,

1
=
_
0.0008 0.0006 0.0002
0.0002 0.0006 0.0002
0.0006 0.0002 0.0006
__
P
1
P
2
P
3
_
.
Title Page 84 of 101 Go Back Full Screen Close Quit
3.8.4 DC power ow approximation to line ow constraints
In principle, there are limits on ow in both directions on each line.
We will assume that the only binding limits are in the directions from
buses 1 to 0, 2 to 1, 2 to 3, and 3 to 0, respectively, as suggested by the
arrows in Figure 3.6.
These four line ow inequality constraints are then specied by
K
1
d, where:
d =
_

_
p
10
p
21
p
23
p
30
_

_
=
_

_
3000
3000
300
3000
_

_
,
K =
_

_
B
10
0 0
0 B
21
0
0 B
23
B
23
B
30
0 B
30
_

_
=
_

_
1000 0 0
0 1000 0
0 500 500
1000 0 1000
_

_
.
Note that the rows of K are indexed by (10), (21), (23), (30), while the
columns are indexed by 0, 2, 3.
Title Page 85 of 101 Go Back Full Screen Close Quit
3.8.5 DC shift factors
The matrix of DC shift factors is:
K[J
0
]
1
=
_

_
1000 0 0
0 1000 0
0 500 500
1000 0 1000
_

_
_
0.0008 0.0006 0.0002
0.0002 0.0006 0.0002
0.0006 0.0002 0.0006
_
,
=
_

_
0.8 0.6 0.2
0.2 0.6 0.2
0.2 0.4 0.2
0.2 0.4 0.8
_

_
.
The augmented shift factor matrix is:

C =
_
0 K[J
0
]
1

,
=
_

_
0.0 0.8 0.6 0.2
0.0 0.2 0.6 0.2
0.0 0.2 0.4 0.2
0.0 0.2 0.4 0.8
_

_
.
Title Page 86 of 101 Go Back Full Screen Close Quit
DC shift factors, continued
For example, for power injected at bus 1 and withdrawn at bus = 0, the
shift factors to the lines 1 to 0, 2 to 1, 2 to 3, and 3 to 0 are, respectively
0.8, 0.2, 0.2, 0.2.
Moreover, the ow on any particular line is the sum of the ows due to
individual injections at particular buses.
For power injected at bus 0 and withdrawn at bus = 0, what are the shift
factors to the lines from buses 1 to 0, 2 to 1, 2 to 3, and 3 to 0?
If 1 MW is injected at bus 1, 10 MW is injected at bus 2, and 100 MW is
injected at bus 3, with 111 MW withdrawn at bus = 0, what is the ow
on the line from bus 1 to bus 0?
Title Page 87 of 101 Go Back Full Screen Close Quit
3.8.6 Line ow constraints in terms of shift factors
The ows on the lines are given by

C(PD) or, equivalently,
K[J
1
]
1
(P
0
D
0
).
The equality and inequality constraints with angles eliminated are:
1

P = 1

D,
K[J
0
]
1
P
0
K[J
0
]
1
D
0
+d.
Again note that D
0
= 0 for this particular example.
Also, d =
_

_
3000
3000
300
3000
_

_
, so these constraints become:
P
1
P
2
P
3
= D
0
,
_

_
0.8 0.6 0.2
0.2 0.6 0.2
0.2 0.4 0.2
0.2 0.4 0.8
_

_
_
P
1
P
2
P
3
_

_

_
3000
3000
300
3000
_

_
.
Title Page 88 of 101 Go Back Full Screen Close Quit
Line ow constraints in terms of shift factors, continued

@
@
@
@
@
@
@
@

@
@
@
@
@
@
@
@

@
@
@
@
@
@
@
@

@
@
@
@
@
@
@
@

@
@
@
@
@
@
@
@

P
1
P
2
D
0
P
3
1 2
0 3
0.2P
1
+0.6P
2
+0.2P
3
3000 MW
0.8P
1
+0.6P
2
+0.2P
3
3000 MW
0.2P
1
+0.4P
2
0.2P
3
300 MW
0.2P
1
+0.4P
2
+0.8P
3
3000 MW
@
@
@R

@
@
@I

? ?
Fig. 3.7. DCpower owapproximation to line owconstraints for four-line four-
bus network.
Title Page 89 of 101 Go Back Full Screen Close Quit
3.8.7 Line ow constraints at other operating points
The derivation so far used the at start condition as the base-case for
evaluating the shift factors and the line ow constraints.
Other base-cases could be used, such as:
another assumed operating point, or
a measured or estimated operating point from a state estimator.
The lossless assumption will typically not hold at other base-cases nor for
the estimated change-case.
Title Page 90 of 101 Go Back Full Screen Close Quit
3.9 DC power ow circuit interpretation
As mentioned in Section 3.6.3, we can interpret the DC power ow
approximation in terms of an analogous DC circuit:
the DC circuit interpretation is useful to solve small systems by hand.
Recall that the power ow equations are in the form J

= P, where is
the reference bus.
If the shunt admittances are zero then J is given by B with the column
corresponding to the reference bus deleted.
Consider the following analogy with a DC circuit:
Bus is the datum node in the circuit with DC voltage dened to be 0,
Real power injections P are analogous to DC current injections i at all
buses,
Angles

are analogous to DC voltages v

at all nodes except the

datum node,
Entries in J are analogous to the admittance matrix of a circuit having
resistors joining nodes and k with conductance g
k
=|B
k
|.
The analogous DC circuit satises Jv

= i:
applies whether or not there are non-zero shunt admittances.
Title Page 91 of 101 Go Back Full Screen Close Quit
DC power ow circuit interpretation, continued
Current injections and ows in the DC circuit correspond to power
injections and ows in the power system.
Recall that currents in a DC circuit can be superposed:
the current owing in a branch due to multiple current injections is
equal to the sum of the currents owing in that branch due to each
current injection considered separately.
Therefore, we can superpose power ow in the DC power ow
approximation in the same way as we superpose current ow in a DC
circuit.
Suppose that the DC circuit has two nodes joined by two conductances,
which we view as two paths between these nodes:
From circuit theory, recall that if current is injected at one node and
withdrawn at another node then current is shared on these paths in
proportion to their conductances.
Moreover, if there is current injected at multiple nodes then the resulting
total current in any branch is equal to the superposition of the currents in
that branch due to the individual current injections.
Title Page 92 of 101 Go Back Full Screen Close Quit
DC power ow circuit interpretation, continued
Recall the example from Figure 3.6, repeated in Figure 3.8.
We will illustrate the DC circuit using this example.

@
@
@
@
@
@
@
@

@
@
@
@
@
@
@
@

@
@
@
@
@
@
@
@

@
@
@
@
@
@
@
@

@
@
@
@
@
@
@
@

P
1
P
2
D
0
P
3
1 2
0 3
0+0.001

1
3000 MW
0+0.001

1
3000 MW
0+0.002

1
300 MW
0+0.001

1
3000 MW
@
@
@R

@
@
@I

? ?
Fig. 3.8. Four-line
four-bus network re-
peated from Figure 3.6.
Title Page 93 of 101 Go Back Full Screen Close Quit
DC power ow circuit interpretation, continued
Suppose current is injected at node 1 in the analogous DC circuit and
withdrawn at the node = 0.
Note that the actual impedance directly joining buses 1 and 0 in the power
system is 0+0.001

1:
analogous conductance of this path is g
10
=|B
10
| = 1000,
current on this analogous conductance is proportional to 1000.
We can also think of the lines from buses 1 to 2, 2 to 3, and 3 to 0 as
another impedance joining buses 1 and 0 in the power system:
total impedance in this path is:
0+0.001

1+0+0.002

10+0.001

1 = 0.004

1,
analogous conductance of path is g
1230
= 250,
current on this analogous conductance due to the lines 1 to 2, 2 to 3, and
3 to 0 is proportional to 250.
Current injected at node 1 and withdrawn at node 0 is shared between the
analogous conductances in the paths in the proportion
1000 : 250 = 0.8 : 0.2.
Title Page 94 of 101 Go Back Full Screen Close Quit
DC power ow circuit interpretation, continued
In the power system, the DC power ow approximation means that power
injected at bus 1 and withdrawn at the slack bus = 0 will be shared in
the ratio 0.8:0.2 between:
the path consisting of the line directly joining buses 1 and 0, and
the lines forming the path from buses 1 to 2, 2 to 3, 3 to 0.
That is, the shift factors to the lines from buses 1 to 0, 2 to 1, 2 to 3, and 3
to 0 are, respectively 0.8, 0.2, 0.2, 0.2, exactly as calculated in the
previous section.
By superposition, we can calculate the total power ow on a line as the
sum of the power ows due to individual power injections.
Title Page 95 of 101 Go Back Full Screen Close Quit
3.10 Summary
In this chapter we formulated the power ow problem.
We considered a linearization of power ow.
We considered xed voltage proles.
We considered the DC power ow.
This chapter is based on:
Sections 8.2 and 9.2 of Applied Optimization: Formulation and
Algorithms for Engineering Systems, Cambridge University Press 2006.
Ross Baldick, Variation of Distribution Factors with Loading, IEEE
Transactions on Power Systems, 18(4):13161323, November 2003.
Brian Stott, Jorge Jardim, and Ongun Alsac, DC Power Flow Revisited,
IEEE Transactions on Power Systems, 24(3):12901300, August 2009.
Title Page 96 of 101 Go Back Full Screen Close Quit
Homework exercises
3.1 Consider a power system consisting of two buses and one transmission line:
bus 1 (the reference/slack bus), where there is a generator, and
bus 2, where there is load.
Suppose that the reference/slack bus voltage is specied to be V
1
= 1

and that
real power ow from bus 2 into the line is given by:
u
2
R
+
,
2
R, p
2
(
2
, u
2
) = u
2
sin
2
.
(That is, we assume that G
22
= G
12
= 0 and B
12
= 1.) Suppose u
2
= 1.0.
(i) What is the largest value of demand D
2
at bus 2 for which there is a
solution to the equation p
2
(
2
, 1.0) +D
2
= 0? What is the corresponding
value of
2
? We will write
2
for this value of
2
.
(ii) What happens if
2
is smaller than
2
?
(iii) Show that there are two solutions to the equation p
2
(
2
, 1.0) +D
2
= 0
with 0
2
>2 if D
2
= 0.5. What are the corresponding values of
2
?
(iv) Use the DC power ow to approximate the relationship between
2
and
D
2
.
(v) When do you expect the DC power ow to be a poor approximation to
the exact solution?
Title Page 97 of 101 Go Back Full Screen Close Quit
3.2 Consider the example in Section 3.8, but suppose that bus 0 is both the
reference and the slack bus, so that = = 0.
(i) What is vector of unknown angles
0
?
(ii) Evaluate J

=
p

0
__
0
1
__
.
(iii) Evaluate [J

0
]
1
.
(iv) Write down the DC power ow equations in terms of generation at buses
1, 2, and 3, and demand at bus 0.
(v) Evaluate the matrix K

in the linearized representation of line ow

inequality constraints K

0
d.
(vi) Evaluate the shift factor matrix K

[J

0
]
1
.
(vii) Write down the line ow inequality constraints in terms of the shift
factors.
(viii) What do you notice about the line ow inequality constraints? Did the
choice of reference bus change the form of the line ow constraints?
(ix) Repeat the previous parts, but with bus 1 both the reference and the slack
bus, so that = = 1.
Title Page 98 of 101 Go Back Full Screen Close Quit
3.3 Consider the three bus, two line system shown in Figure 3.9 and suppose
that bus 1 is both the reference and the slack bus, so that = = 1. The line
capacities are shown. Assume that the susceptance joining bus 2 to bus 1 and the
susceptance joining bus 3 to bus 2 are both non-zero.
Bus 3 Bus 2 Bus 1
Line Capacity
100 MW
Line Capacity
1000 MW
Fig. 3.9. Three bus,
two line radial network.
(i) What is vector of unknown angles
1
?
(ii) Evaluate J =
p

1
__
0
1
__
. (Hint: See (3.10).)
(iii) Write down the DC power ow equations in terms of generation and
demand at buses 2 and 3.
Title Page 99 of 101 Go Back Full Screen Close Quit
(iv) Evaluate [J
1
]
1
.
(v) Assume that the only ow constraints are from bus 2 to bus 1, and from
bus 3 to bus 2. Evaluate the matrix K in the linearized representation of
real power line ow limit inequality constraints K
1
d.
(vi) Evaluate the shift factor matrix K[J
1
]
1
.
(vii) Write down the real power line ow limit inequality constraints in terms
of the shift factors as in (3.13).
(viii) Interpret these inequality constraints in terms of the gure.
Title Page 100 of 101 Go Back Full Screen Close Quit
3.4 Assuming that J
1
is invertible, show that the DC power ow equations
J
1
= P are equivalent to:
P
1
= 1

P
1
,

1
= [J
1
]
1
P
1
.
That is, show that
1
satises J
1
= P if and only if
1
satises

1
= [J
1
]
1
P
1
and P
1
=1

P
1
. (Hint: See discussion in Section 3.6.4.)
Title Page 101 of 101 Go Back Full Screen Close Quit