Power Flow

The Power Flow Problem

James D. McCalley, Iowa State University
T7.0 Introduction
The power flow problem is a very well known problem in the field of power systems engineering, where
voltage magnitudes and angles for one set of buses are desired, given that voltage magnitudes and power
levels for another set of buses are known and that a model of the network configuration (unit commitment
and circuit topology) is available. A power flow solution procedure is a numerical method that is employed
to solve the power flow problem. A power flow program is a computer code that implements a power flow
solution procedure. The power flow solution contains the voltages and angles at all buses, and from this
information, we may compute the real and reactive generation and load levels at all buses and the real and
reactive flows across all circuits. The above terminology is often used with the word load! substituted for
power,! i.e., load flow problem, load flow solution procedure, load flow program, and load flow solution.
"owever, the former terminology is preferred as one normally does not think of load! as something that
flows.!
The power flow problem was originally motivated within planning environments where engineers
considered different network configurations necessary to serve an e#pected future load. $ater, it became an
operational problem as operators and operating engineers were re%uired to monitor the real&time status of
the network in terms of voltage magnitudes and circuit flows. Today, the power flow problem is widely
recogni'ed as a fundamental problem for power system analysis, and there are many advanced, commercial
power flow programs to address it. (ost of these programs are capable of solving the power flow program
for tens of thousands of interconnected buses. )ngineers that understand the power flow problem, its
formulation, and corresponding solution procedures are in high demand, particularly if they also have
e#perience with commercial grade power flow programs.
The power flow problem is fundamentally a network analysis problem, and as such, the study of it provides
insight into solutions for similar problems that occur in other areas of electrical engineering. For e#ample,
integrated circuit designers also encounter network analysis problems, although of significantly smaller
physical si'e, are %uite similar otherwise to the power flow problem. For e#ample, references *+,,- are
well&known network analysis te#ts in .$/0 design that also provide good insight into the numerical
analysis needed by the power flow program designer. /imilarly, there are numerous classical power system
engineering te#ts, *1&++- are a representative sample, that provide advanced network analysis methods
applicable to .$/0 design and analysis problems.
/ection T2.+ identifies a feature of power generators important to the power flow problem 3 real and
reactive power limits. /ection T2., defines some additional terminology necessary to understand the power
flow problem and its solution procedure. /ection T2.1 introduces the so&called network 4&bus,! otherwise
known more generally as the network admittance matri#. /ection T2.5 develops the power flow e%uations,
building from module T+ where e%uations for real and reactive power flow across a transmission line were
introduced. /ection T2.6 provides an analytical statement of the power flow problem. /ection T2.7 uses a
simple e#ample to introduce the 8ewton&9aphson algorithm for solving systems of non&linear algebraic
e%uations. /ection T2.2 illustrates application of the 8ewton&9aphson algorithm to the power flow
problem. /ection T2.: provides an overview of several interesting and advanced attributes of the problem.
/ection T2.; summari'es basic power flow input and output %uantities and provides an e#ample associated
with a commercial power flow program.

T7.1 Generator eactive !imits
0t is well known that generators have ma#imum and minimum real power capabilities. 0n addition, they also
have ma#imum and minimum reactive power capabilities. The ma#imum reactive power capability
+
corresponds to the ma#imum reactive power that the generator may produce when operating with a lagging
power factor. The minimum reactive power capability corresponds to the ma#imum reactive power the
generator may absorb when operating with a leading power factor. These limitations are a function of the
real power output of the generator, that is, as the real power increases, the reactive power limitations move
closer to 'ero. The solid curve in Figure T2.+ is a typical generator capability curve, which shows the
lagging and leading reactive limitations (the ordinate) as real power is varied (the abscissa). (ost power
flow programs model the generator reactive capabilities by assuming a somewhat conservative value for
Pma# (perhaps ;6< of the actual value), and then fixing the reactive limits =ma# (for the lagging limit) and
=ming (for the leading limit) according to the dotted lines shown in Fig. T,.+.

P
ma#
=
ma#

=
min

P
=
leading
operation
lagging
operation
"i# T7.1$ Generator Ca%a&ility Curve and '%%ro(imate eactive !imits
T7.) Terminolo#y
>ulk high voltage transmission systems are always comprised of three phase circuits. "owever, under
balanced conditions (the currents in all three phases are e%ual in magnitude and phase separated by +,?),
we may analy'e the three phase system using a per&phase e%uivalent circuit consisting of the a&phase and
the neutral conductor. Per&uniti'ation of a per&phase e%uivalent of a three phase, balanced system results in
the per&unit circuit. 0t is the per&uniti'ed, per&phase e%uivalent circuit of the power system that we use to
formulate and solve the power flow problem. For the remainder of this module, we will assume that all
%uantities are in per&unit. The reader unfamiliar with per&phase e%uivalent circuits or the per&unit system
should refer to modules >1 and >5, respectively.
0t is convenient to represent power system networks using the so&called one&line diagram, which can be
thought of as the circuit diagram of the per&phase e%uivalent, but without the neutral conductor (module >1
also provides additional background on the one&line diagram). Figure T2., illustrates the one&line diagram
of a small transmission system.
Fig. T2., illustrates several important elements of the power flow problem. First, one notices we may
categori'e each bus depending on whether generation and@or load is connected to it. /pecifically, a bus may
have generation only (buses >+, >,, and >1), load only (buses >6, >2, and >;), or neither generation or
load (buses >5, >7, and >:). 0n addition, a bus may have both generation and load, although none of the
buses in Fig. + fall into this category. This categori'ation, which focuses on the load and generation, leads
us to define the term bus inAection! or more simply, inAection.! Be will use this term fre%uently, and the
student is advised to carefully note its meaning, given and discussed in the following paragraph.
An injection is the power, either real or reactive, that is being inAected into or withdrawn from a bus by an
element having its other terminal (in the per&phase e%uivalent circuit) connected to ground. /uch an
element would be either a generator or a load. Be define a positive inAection as one where power is flowing
from the element into the bus (i.e., into the network)C a negative inAection is then when power is flowing
from the bus (i.e., from the network) into the element. Denerators normally have positive real power
inAections, although they may also be assigned negative real power inAections, in which case they are
,
Appro#imate
=&limits of
each generator
using dotted
bo#.
For pf analysis,
we use per&
phase circuits
in the per&unit
system.
Ene&line
diagramF per&
phase circuit
diagram without
neutral.
operating as a motor. Denerators may have either positive or negative reactive power inAectionsF positive if
the generator is operating lagging and delivering reactive power to the bus, negative if the generator is
operating leading and absorbing reactive power from the bus, and 'ero if the generator is operating at unity
power factor. $oads normally have negative real and reactive power inAections, although they may also be
assigned positive real power inAections in the case of very special modeling needs. Figure T2.1 (a) and (b)
illustrate the two most common possibilities. Figure T.2.1 (c) illustrates that we must compute a net
inAection as the algebraic sum when a bus has both load and generationC in this case, the net inAection for
both real and reactive power is positive (into the bus). Thus, the net real power inAection is PkGPgk&Pdk, and
the net reactive power inAection is =kG=gk&=dk. Be may also refer to the net comple# power inAection as
/kG/gk&/dk, where /kGPkHA=k.

"i#ure T7.)$ Sin#le !ine Dia#ram *or Sim%le +ower System

PkG+??
=kG1?
(a)
PkG & 5?
=kG &,?
(b)
PkG+??H(&5?)G7?
=kG1?H(&,?)G+?
(c)
"i# T7.,$ Illustration o* -a. %ositive in/ection, -&. ne#ative in/ection, and -c. net in/ection
1
Three bus typesF
+. P.
,. P=
1. /wing (PI)
$osses are a function of flows.
Be do not know flows a priori.
/o one gen&bus has inAection as
a function of the solution.
P+GP$osses(solution)&JkG,,8 Pk
=+G=$osses(solution)&JkG,,8 Pk
9eal, reactive
power inAectionF
into busF gen
out of busF load
into bus is
defined positive.
There are four
variables for
every busF
P
=
.
I
Although it is physically appealing to categori'e buses based on the generation@load mi# connected to it, we
need to be more precise in order to analytically formulate the power flow problem. For proper analytical
formulation, it is appropriate to categori'e the buses according to what information is known about them
before we solve the power flow problem. For each bus, there are four possible variables that characteri'e
the buses electrical condition. $et us consider an arbitrary bus numbered k. The four variables are real and
reactive power inAection, Pk and =k, respectively, and voltage magnitude and angle, K.k K and k,
respectively. From this perspective, there are three basic types of buses. Be refer to the first two types using
terminology that remind us of the known variables.
PV BusesF For type P. buses, we know Pk and K.k K but not =k or k. These buses fall under the
category of voltage-controlled buses because of the ability to specify (and therefore to know) the
voltage magnitude of this bus. (ost generator buses fall into this category, independent of whether it
also has loadC e#ceptions are buses that have reactive power inAection at either the generatorLs upper
limit (=ma#) or its lower limit (=min), and (,) the system swing bus (we describe the swing bus below).
There are also special cases where a non&generator bus (i.e., either a bus with load or a bus with neither
generation or load) may be classified as type P., and some e#amples of these special cases are buses
having switched shunt capacitors or static var systems (/.Ms). Be will not address these special cases
in this module. 0n Fig. T2.,, buses >, and >1 are type P.. The real power inAections of the type P.
buses are chosen according to the system dispatch corresponding to the modeled loading conditions.
The voltage magnitudes of the type P. buses are chosen according to the e#pected terminal voltage
settings, sometimes called the generator set points,! of the units.
PQ BusesF For type P= buses, we know Pk and =k but not K.k K or k. All load buses fall into this
category, including buses that have not either load or generation. 0n Fig. T2.,, buses >5&>; are all type
P=. The real power inAections of the type P= buses are chosen according to the loading conditions
being modeled. The reactive power inAections of the type P= buses are chosen according to the
e#pected power factor of the load.
The third type of bus is referred to as the swing bus. Two other common terms for this bus are slack bus and
reference bus. There is only one swing bus, and it can be designated by the engineer to be any generator bus
in the system. For the swing bus, we know K.K and . The fact that we know is the reason why it is
sometimes called the reference bus. Physically, there is nothing special about the swing busC in fact, it is a
mathematical artifact of the solution procedure. At this point in our treatment of the power flow problem, it
is most appropriate to understand this last statement in the following way. The generation must supply both
the load and the losses on the circuits. >efore solving the power flow problem, we will know all inAections
at P= buses, but we will not know what the losses will be as losses are a function of the flows on the
circuits which are yet to be computed. /o we may set the real power inAections for, at most, all but one of
the generators. The one generator for which we do not set the real power inAection is the one modeled at the
swing bus. Thus, this generator swings! to compensate for the network losses, or, one may say that it
takes up the slack.! Therefore, rather than call this generator a K.K bus (as the above naming convention
would have it), we choose the terminology swing! or slack! as it helps us to better remember its function.
The voltage magnitude of the swing bus is chosen to correspond to the typical voltage setting of this
generator. The voltage angle may be designated to be any angle, but normally it is designated as ?
o
.
A word of caution about the swing bus is in order. >ecause the real power inAection of the swing bus is not
set by the engineer but rather is an output of the power flow solution, it can take on mathematically
tractable but physically impossible values. Therefore, the engineer must always check the swing bus
generation level following a solution to ensure that it is within the physical limitations of the generator.

T7., T0e 'dmittance Matri(
Murrent inAections at a bus are analogous to power inAections. The student may have already been
introduced to them in the form of current sources at a node. Murrent inAections may be either positive (into
the bus) or negative (out of the bus). Nnlike current flowing through a branch (and thus is a branch
%uantity), a current inAection is a nodal %uantity. The admittance matri#, a fundamental network analysis
tool that we shall use heavily, relates current inAections at a bus to the bus voltages. Thus, the admittance
matri# relates nodal %uantities. Be motivate these ideas by introducing a simple e#ample.
5
Mheck the
swing bus after
a solution to
ensure it is
within physical
limits.
Figure T2.5 shows a network represented in a hybrid fashion using one&line diagram representation for the
nodes (buses +&5) and circuit representation for the branches connecting the nodes and the branches to
ground. The branches connecting the nodes represent lines. The branches to ground represent any shunt
elements at the buses, including the charging capacitance at either end of the line. All branches are denoted
with their admittance values yiA for a branch connecting bus i to bus A and yi for a shunt element at bus i. The
current inAections at each bus i are denoted by 0i.

y
5
y
1
y
+
y
,
0
5
0
1
0
,
0
+
5 1
,
+
y
15
y
,1
y
+1
y
+,
"i#. T7.1$ 2etwor3 *or Motivatin# 'dmittance Matri(
OirchoffLs Murrent $aw (OM$) re%uires that each of the current inAections be e%ual to the sum of the
currents flowing out of the bus and into the lines connecting the bus to other buses, or to the ground.
Therefore, recalling EhmLs $aw, 0G.@PG.4, the current inAected into bus + may be written asF
0+G(.+&.,)y+, H (.+&.1)y+1 H .+y+ (T2.+)
To be complete, we may also consider that bus + is connected! to bus 5 through an infinite impedance,
which implies that the corresponding admittance y+5 is 'ero. The advantage to doing this is that it allows us
to consider that bus + could be connected to any bus in the network. Then, we haveF
0+G(.+&.,)y+, H (.+&.1)y+1 H (.+&.5)y+5 H .+y+ (T2.,)
8ote that the current contribution of the term containing y+5 is 'ero since y+5 is 'ero. 9earranging e%. T2.,,
we haveF
0+G .+( y+ H y+, H y+1 H y+5) H .,(&y+,)H .1(&y+1) H .5(&y+5) (T2.1)
/imilarly, we may develop the current inAections at buses ,, 1, and 5 asF
0,G .+(&y,+) H .,( y, H y,+ H y,1 H y,5) H .1(&y,1) H .5(&y,5) (T2.5)
01G .+(&y1+)H .,(&y1,) H .1( y1 H y1+ H y1, H y15) H .5(&y15)
05G .+(&y5+)H .,(&y5,) H .1(&y15)H .5( y5 H y5+ H y5, H y51)
where we recogni'e that the admittance of the circuit from bus k to bus i is the same as the admittance from
bus i to bus k, i.e., ykiGyik From e%s. (T2.1) and (T2.5), we see that the current inAections are linear functions
of the nodal voltages. Therefore, we may write these e%uations in a more compact form using matrices
according toF
6
)#press current
inAections 0k using
OM$. To be
general, assume a
fully connected
network.
1
1
1
1
]
1
1
1
1
1
]
1
+ + +
+ + +
+ + +
+ + +
1
1
1
1
]
1
5
1
,
+
51 5, 5+ 5 51 5, 5+
15 15 1, 1+ 1 1, 1+
,5 ,1 ,5 ,1 ,+ , ,+
+5 +1 +, +5 +1 +, +
5
1
,
+
V
V
V
V
y y y y y y y
y y y y y y y
y y y y y y y
y y y y y y y
I
I
I
I
(T2.6)
The matri# containing the network admittances in e%. (T2.6) is the admittance matri#, also known as the 4&
bus, and denoted asF
1
1
1
1
]
1
+ + +
+ + +
+ + +
+ + +
51 5, 5+ 5 51 5, 5+
15 15 1, 1+ 1 1, 1+
,5 ,1 ,5 ,1 ,+ , ,+
+5 +1 +, +5 +1 +, +
y y y y y y y
y y y y y y y
y y y y y y y
y y y y y y y
(T2.7)
Qenoting the element in row i, column A, as 4iA, we rewrite e%. (T2.7) asF
1
1
1
1
]
1
55 51 5, 5+
15 11 1, 1+
,5 ,1 ,, ,+
+5 +1 +, ++

(T2.2)
where the terms 4iA are not admittances but rather elements of the admittance matri#. Therefore, e%. (T2.7)
becomesF
1
1
1
1
]
1
1
1
1
1
]
1
1
1
1
1
]
1
5
1
,
+
55 51 5, 5+
15 11 1, 1+
,5 ,1 ,, ,+
+5 +1 +, ++
5
1
,
+
V
V
V
V

I
I
I
I
(T2.:)
>y using e%. (T2.2) and (T2.:), and defining the vectors . and 0, we may write e%. (T2.:) in compact form
according toF
1
1
1
1
]
1
1
1
1
1
]
1
5
1
,
+
5
1
,
+
,
I
I
I
I
I
V
V
V
V
V V I
(T2.;)
Be make several observations about the admittance matri# given in e%s. (T2.7) and (T2.2). These
observations hold true for any linear network of any si'e.
+. The matri# is symmetric, i.e., 4iAG4Ai.
7
Ebserve the
symmetry of
the 4&bus.
4&bus
formulation
rules.
,. A diagonal element 4ii is obtained as the sum of admittances for all branches connected to bus i,
including the shunt branch, i.e.,

+
!
i k k
ik i ii
y y
, +
, where we emphasi'e once again that yik is
non&'ero only when there e#ists a physical connection between buses i and k.
1. The off&diagonal elements are the negative of the admittances connecting buses i and A, i.e., 4iAG&yAi.
These observations enable us to formulate the admittance matri# very %uickly from the network based on
visual inspection. The following e#ample will clarify.
4(am%le T7.1
Monsider the network given in Fig. T2.6, where the numbers indicate admittances.

A?.5

A?.1

A?.+

A?.,

0
5
0
1 0
,
0
+
5 1
,
+
,&A1

,&A6

+&A5

,&A5

"i#. T7.5$ Circuit *or 4(am%le T7.1
The admittance matri# is given by inspection asF
1
1
1
1
]
1
+
+ + +
+ +
+ +
1
1
1
1
]
1
7 . , , 1 , ? ?
1 , 2 . ++ 6 6 , 5 +
? 6 , : . : 5 5 ,
? 5 + 5 , ; . 2 ,
55 51 5, 5+
15 11 1, 1+
,5 ,1 ,, ,+
+5 +1 +, ++
j j
j j j j
j j j
j j j

T7.1 T0e %ower *low e6uations

Be have defined the net comple# power inAection into a bus, in /ection T2.,, as /kG/gk&/dk. 0n this section,
we desire to derive an e#pression for this %uantity in terms of network voltages and admittances. Be begin
by reminding the reader that all %uantities are assumed to be in per unit, so we may utili'e single&phase
power relations. Qrawing on the familiar relation for comple# power, we may e#press /k asF
/kG.k0k
R
(T.2.+?)
From e%. (T2.:), we see that the current inAection into any bus k may be e#pressed as
j
!
j
kj k
V I

+
(T2.++)

where, again, we emphasi'e that the 4kA terms are admittance matri# elements and not admittances.
/ubstitution of e%. (T2.++) into e%. (T2.+?) yieldsF
2
The elements of 4&bus are
designated as 4kAG DkAHA>kA.
Ebserve >kAS? for diagonal
elementsC >kAT? otherwise.
The shunt elements all have
positive susceptance, and
must therefore be capacitive.
The most important step in
deriving pf e%tsF substitute 4&
bus relation into /G.0R.
R
+
R
R
+
j
!
j
kj k j
!
j
kj k k
V V V V "

,
_
(T2.+,)
9ecall that .k is a phasor, having magnitude and angle, so that .kGK.kKk. Also, 4kA, being a function of
admittances, is therefore generally comple#, and we define DkA and >kA as the real and imaginary parts of the
admittance matri# element 4kA, respectively, so that 4kAGDkAHA>kA. Then we may rewrite e%. (T2.+,) as
( ) ( )
( ) ( )

+
!
j
kj kj j k j k
!
j
kj kj j j k k
j j
!
j
kj kj k k j j
!
j
kj kj k k j
!
j
kj k k
jB # V V jB # V V
V jB # V V jB # V V V "
+ +
+
R
+
R R
+
R
) ( ) ( ) (
) ( ) (

(T2.+1)
9ecall, from the )uler relation, that a phasor may be e#pressed as comple# function of sinusoids, i.e., .GK.K
GK.KUcosHAsinV, we may rewrite e%. (T2.+1) as
( )
( )
+

!
j
kj kj j k j k j k
!
j
kj kj j k j k k
jB # j V V
jB # V V "
+
+
) ( ) sin( ) cos(
) ( ) (

(T2.+5)
0f we now perform the algebraic multiplication of the two terms inside the parentheses of e%. (T2.+5), and
then collect real and imaginary parts, and recall that /kGPkHA=k, we can e#press e%. (T2.+5) as two
e%uations, one for the real part, Pk, and one for the imaginary part, =k, according toF
( )
( )

+
!
j
j k kj j k kj j k k
!
j
j k kj j k kj j k k
B # V V Q
B # V V P
+
+
) cos( ) sin(
) sin( ) cos(

(T2.+6)
The two e%uations of (T2.+6) are called the power flow e%uations, and they form the fundamental building
block from which we attack the power flow problem.
0t is interesting to consider the case of e%s. (T2.+6) if bus k, relabeled as bus p, is only connected to one
other bus, letLs say bus %. Then the bus p inAection is the same as the flow into the line p%. The situation is
illustrated in Fig. T2.7.

>us % >us p
/eries
admittance
D&A>
>us
inAection
P
p
and =
p

$ine flow
P
p%
and =
p%

"i#. T7.7$ 8us % Connected to 9nly 8us 6
For the situation illustrated in Fig. T2.7, e%s. (T2.+6) becomeF
:
The pf e%uationsW Be will be
very wise to ensure that the
left&hand&side is always
known.
) cos( ) sin(
) sin( ) cos(
,
,
$ p p$ $ p $ p p$ $ p pp p p
$ p p$ $ p $ p p$ $ p pp p p
B V V # V V B V Q
B V V # V V # V P

+
+ +
(T2.+7)
0f the line p% admittance is yGD&A>
+
, as shown in Fig. T2.7, then Dp%G&D and >p%G> (see e%. T2.7). 0f there
is no bus p shunt reactance or line charging, then DppGD and >ppG>. Nnder these conditions, e%s. (T2.+7)
becomeF
) cos( ) sin(
) sin( ) cos(
,
,
$ p $ p $ p $ p p p
$ p $ p $ p $ p p p
B V V # V V B V Q
B V V # V V # V P

+
(T2.+2)
0f we simply rearrange the order of the terms in the reactive e%uation, then we haveF
) sin( ) cos(
) sin( ) cos(
,
,
$ p $ p $ p $ p p p
$ p $ p $ p $ p p p
# V V B V V B V Q
B V V # V V # V P

+
(T2.+:)
T7.5 'nalytic statement o* t0e %ower *low %ro&lem
Monsider a power system network having 8 buses, 8D of which are voltage&regulating generators. Ene of
these must be the swing bus. Thus there are 8D&+ type P. buses, and 8&8D type P= buses. Be assume that
the swing bus is numbered bus +, the type P. buses are numbered ,,X, 8D, and the type P= buses are
numbered 8DH+,X,8 (this assumption on numbering is not necessary, but it makes the following
development notationally convenient). 0t is typical that we know, in advance, the following information
about the network (implying that it is input data to the problem)F
+. The admittances of all series and shunt elements (implying that we can obtain the 4&bus),
,. The voltage magnitudes .k, kG+,X,8D, at all 8D generator buses,
1. The real power inAection of all buses e#cept the swing bus, Pk, kG,,X,8
5. The reactive power inAection of all type P= buses, =k, kG8DH+, X, 8
/tatements 1 and 5 indicate power flow e%uations for which we know the inAections, i.e., the values of the
left&hand side of e%s. (T2.+6). These e%uations are very valuable because they have one less unknown than
e%uations for which we do not know the left&hand&side. The number of these e%uations for which we know
the left&hand&side can be determined by adding the number of buses for which we know the real power
inAection (statement 1 above) to the number of buses for which we know the reactive power inAection
(statement 5 above). This is (8&+)H(8&8D)G,8&8D&+. Be repeat the power flow e%uations here, but this
time, we denote the appropriate number to the right.
( )
( ) ! ! k B # V V Q
! k B # V V P
#
!
j
j k kj j k kj j k k
!
j
j k kj j k kj j k k
,..., + , ) cos( ) sin(
,..., , , ) sin( ) cos(
+
+
+
+

(T2.+;)
Be are trying to find the following information about the networkF
a. The angles for the voltage phasors at all buses e#cept the swing bus (it is ? at the swing bus), i.e.,
k, kG,,X,8
b. The magnitudes for the voltage phasors at all type P= buses, i.e., K.kK, kG8DH+, X, 8
+
Be have defined the line p% admittance to be yGD&A>, instead of yGDHA>. The reason for defining in this way is because, since the
line admittance y always represents inductive susceptance, the imaginary part of y must be negativeC therefore the definition used here
re%uires that > to be a positive number.
;
Y1 is $"/ of
8&+ real pf
e%uations.
Y5 is $"/ of
8&8D reactive
pf e%uations.
/ince these
$"/ %uantities
are known,
inclusion of
the
corresponding
pf e%uation
does not add
an unknown.
/tatements a and b imply that we have 8&+ angle unknowns and 8&8D voltage magnitude unknowns, for a
total number of unknowns of (8&+)H(8&8D)G,8&8D&+. 9eferring to the power flow e%uations, e%. (T2.+;),
we see that there are no other unknowns on the right&hand side besides voltage magnitudes and angles (the
real and imaginary parts of the admittance values, DkA and >kA, are known, based on statement + above).
Thus we see that the number of e%uations having known left&hand side (inAections) is the same as the
number of unknown voltage magnitudes and angles. Therefore it is possible to solve the system of ,8&8D&+
e%uations for the ,8&8D&+ unknowns. "owever, we note from e%. (T2.+;) that these e%uations are not
linear, i.e., they are nonlinear e%uations. This nonlinearity comes from the fact that we have terms
containing products of some of the unknowns and also terms containing trigonometric functions of some of
the unknowns. >ecause of these nonlinearities, we are not able to put them directly into the familiar matri#
form of A#Gb! (where A is a matri#, # is the vector of unknowns, and b is a vector of constants) to obtain
their solution. Be must therefore resort to some other methods that are applicable for solving nonlinear
e%uations. Be describe such a method in /ection T2.7. >efore doing that, however, it may be helpful to
more crisply formulate the e#act problem that we want to solve.
$etLs first define the vector of unknown variables. This we do in two steps. First, define the vector of
unknown angles (an underline beneath the variable means it is a vector or a matri#) and the vector of
unknown voltage magnitudes K.K.
1
1
1
1
]
1
1
1
1
1
]
1
+
+
% %V
% %V
% %V
% V %
!
!
!
!
#
#

,
+
1
,
,
(T2.,?)
/econd, define the vector # as the composite vector of unknown angles and voltage magnitudes.
1
1
1
1
1
1
1
1
1
1
]
1
1
1
1
1
1
1
1
1
1
1
]
1
1
]
1
+
+
+ ,
+
+
,
+
,
+
1
,
#
#
#
#
! !
!
!
!
!
!
!
!
x
x
x
x
x
x
% %V
% %V
% %V
&
&
&
% V %
&
(T2.,+)
Bith this notation, we see that the right&hand sides of e%s. (T2.+;) depend on the elements of the unknown
vector #. )#pressing this dependence more e#plicitly, we rewrite e%s. (T2.+;) as
'...'! ! k x Q Q
'...'! k x P P
# k k
k k
+ , ) (
, , ) (
+

(T2.,,)
0n e%s. (T2.,,), Pk and =k are the specified inAections (known constants) while the right&hand sides are
functions of the elements in the unknown vector #. >ringing the left&hand side over to the right&hand side,
we have that
'...'! ! k Q x Q
'...'! k P x P
# k k
k k
+ , ? ) (
, , ? ) (
+

(T2.,1)
+?
8umber of
e%uationsG
number of
unknownsW
/olution
vector.
Mompact
notation for PF
e%ts. $"/ is a
numberC 9"/ is
a function.
Put in form
f(#)G?.
Be now define a vector&valued function f(#) asF
?
?
?
?
?
) (
) (
) (
) (
) (
) (
) (
) (
) (
+
,
+ +
, ,
+ ,
+
+
1
1
1
1
1
1
1
1
1
]
1
1
1
1
1
1
1
1
1
1
]
1
1
1
1
1
1
1
1
1
1
]
1
1
1
1
1
1
1
1
1
1
]
1

+ + +

!
!
!
! !
! !
! !
! !
!
!
Q
Q
P
P
Q x Q
Q x Q
P x P
P x P
x f
x f
x f
x f
x f
# # #
#
(T2.,5)
)%uation (T2.,5) is in the form of f(#)G?, where f(#) is a vector&valued function and ? is a vector of 'erosC
both f(#) and ? are of dimension (,8&8D&+)+, which is also the dimension of the vector of unknowns, #.
Be have also introduced nomenclature representing the mismatc( vector in e%. (T2.,5), as the vector of
PkLs and =kLs. This vector is used during the solution algorithm, which is iterative, to identify how good
the solution is corresponding to any particular iteration. 0n the ne#t section, we introduce this solution
algorithm, which can be used to solve this kind of system of e%uations. The method is called the 8ewton&
9aphson method.
T7.7 T0e 2ewton:a%0son Solution +rocedure
There are two basic methods for solving the power flow problemF Dauss&/iedel (D/) and 8ewton&9aphson
(89). >oth of these methods are iterative root finding sc(emes.
The D/ and 89 methods are often classified as root finding schemes because they are geared towards
solving e%uations like f(#)G? (or f(#)G?). The solution to such an e%uation, call it #R (or #R), is clearly a
root of the function f(#) (or f(#)).
The methods are called iterative because they re%uire a series of successive appro#imations to the solutions.
The procedure is generally as follows. First, guess a solution. Nnless we are very fortunate, the guess will
be, of course, wrong. /o we determine an update to the old! solution that moves to a new! solution with
the intention that the new! solution is closer to the correct solution than was the old! solution. A key
aspect to this type of procedure is the way we obtain the update. 0f we can guarantee that the update is
always improving the solution, such that the new! solution is in fact always closer to the correct solution
than the old! solution, then such a procedure can be guaranteed to work if only we are willing to compute
enough updates, i.e., if only we are willing to iterate enough times.
Mommercial grade power flow programs may make several different solutions procedures available, but
almost all such programs will have available, minimally, the 89 method. 0t is fair to say that the 89
method has become the de&facto industry standard. The main reason for this is that the convergence
properties of the 89 scheme are very desirable when the initial, guessed solution is %uite good, i.e., when it
is chosen close to the correct solution. 0n the power flow problem, it is usually possible to make a good
initial guess regarding the solution. Ene reason for this is that often, we may actually know the solution of a
particular set of conditions because we have already gone through the solution procedure, and we want to
resolve for a set of conditions that are almost the same as the previous ones, e.g., maybe remove one circuit
or change the load level a little. 0n this case, we may utili'e the previous solution as the initial guessed
solution for the new conditions. This is sometimes referred to as a hot! start. >ut even if we do not have a
previous solution, we still may do very well with our guess. The reason for this is that the power flow
problem is always solved with all %uantities in per&unit. >ecause of the way we choose per&unit voltage
bases, the per&unit voltages for all buses, under any reasonably normal condition, will be close to +.? per&
++
Brite out all
e%uations.
The mismatch
vector.
0terative
methods
re%uired to
solve this
nonlinear
algebraic
system of
e%uations.
(ethod is
guess,
check,
update,
check,
update,
check, X
0nitial guess
is very
important.
0f you guess
poorly, you
may get
wrong
solution or
no solution.
Qivergence
indicates
bad initial
guess or no
solution.
8o solution
may be due
to stressed
condition or
bad data.
unit. Ef course, this tells us nothing about the angles, but it is something, and often it is enough to simply
guess that all voltages are +.? per&unit and all angles are ? degrees. This is sometimes called a flat! start.
>ut what are convergence properties! of a root finding methodZ There are basically two of them. Ene is
w(et(er the method will converge. The second one is (ow fast the method will converge. For 89, whether
the method will converge depends on two thingsF how close the guessed solution is to the correct solution
and the nature of the function close to the correct solution. 0f the guessed solution is close, and if the
function is reasonably smooth! close to the correct solution, then the 89 will converge. 8ot only that, but
it will converge $uadratically. =uadratic convergence means that each iteration increases the accuracy of
the solution by two decimal places. For e#ample, if the correct solution for a particular problem is
?.+,15672:;, and we guess ?.+????????, then the first iteration will yield ?.+,1######, the second
iteration will yield ?.+,156####, the third iteration will yield ?.+,15672##, and the correct solution will be
obtained e#actly on the fourth iteration.
0n this module, we will not discuss the D/ method, but the interested reader may find information about it
in many te#ts on power systems analysis or in books on numerical methods. Be will introduce the 89
method with a simple illustration, obtained from *1-.
4(am%le T7.)
Monsider the scalar function f(#)G#
,
&6#H5. This function may be easily factored to find the roots as #RG5,+.
$et us now illustrate how the 89 method finds one of these roots. Be first need the derivativeF f!(#)G,#&6.
Assume we are bad guessers, and try an initial guess of #
(?)
G7. The following provides the first two
iterationsF
+. f(#
(?)
)Gf(7)G7
,
&6(7)H5G+?
,. f!(#
(?)
)GfL(7)G,(7)&6G2
1. #
(?)
G &f(#
(?)
)@fL(#
(?)
)G &+?@2G&+.5,;
5. #
(+)
G#
(?)
H#
(?)
G7H(&+.5,;)G5.62+
+. f(#
(+)
)Gf(5.62+)G,.?1;?5
,. f!(#
(+)
)GfL(5.62+)G5.+5,
1. #
(+)
G &f(#
(+)
)@fL(#
(+)
)G &,.?1;?5@5.+5,G&?.5;,,:5
5. #
(,)
G#
(+)
H#
(+)
G5.62+H(&?.5;,,:5)G5.?2:2
Ene more iteration yields #
(1)
G5.??,. 8ote that by the third iteration, as it is getting very close to the correct
solution, the algorithm has almost obtained %uadratic convergence. Fig. T2.2 illustrates how the first
solution #
(+)
is found from the initial guessed solution #
(?)
during the first iteration of this algorithm.
The 89 algorithm is not smart enough to know which root you want, rather, it generally finds the closest
root. This is another reason for making a good initial guess in regards to the solution. Fortunately, in the
case of the power flow problem, alternative solutions are typically far away! from initial guesses that have
near&unity bus voltage magnitudes. En the other hand, it is possible for the solution to diverge, i.e., not to
converge at all. This may occur if there is simply no solution, which is a case that engineers encounter
fre%uently when studying highly stressed loading conditions served by weak transmission systems. 0t also
might occur if the initial guessed solution is too far away from the correct solution. For this reason, flat!
starts encounter solution divergence more fre%uently than hot! starts.
+,
8umerical
illustration
of 8&9
procedure
for a single
variable
root finding
problem.
0f your initial guess is ,.6,
the derivative is ?, and the
algorithm fails (corresponds
to matri# singularity in the
multi&dimensional case). 0f
your initial guess is T,.6,
you get the solution 5. 0f
your initial guess is S,.6,
you get the solution +.

#
(+)
#
(?)
"i#. T7.7$ Illustration o* t0e *irst iteration o* t0e 2ewton:a%0son al#orit0m
8e#t, we develop the 89 update formula. Be begin with the scalar case, where the update formula may be
easily inferred from )#ample T2.,.
8ewton 9aphson for the /calar MaseF
Assume that we have guessed a solution #
(?)
to the problem f(#)G?. Then f(#
(?)
)? because #
(?)
is Aust a
guess. >ut there must be some #
(?)
which will make f(#
(?)
H #
(?)
)G?. Ene way to study this problem is to
e#pand the function f(#) in a Taylor series, as followsF
? ... ) )( ( [ [
,
+
) ( [ ) ( ) (
, ) ? ( ) ? ( ) ? ( ) ? ( ) ? ( ) ? ( ) ? (
+ + + + x x f x x f x f x x f (T2.,6)
0f the guess is a good one, then #
(?)
will be small, and if this is true, then (#
(?)
)
,
will be very small, and
any higher order terms (h.o.t.) in e%. (T2.,6), which will contain #
(?)
raised to even higher powers, will be
infinitesimal. As a result, it is reasonable to appro#imate e%. (T2.,6) as
? ) ( [ ) ( ) (
) ? ( ) ? ( ) ? ( ) ? ( ) ? (
+ + x x f x f x x f (T2.,7)
Taking f(#
(?)
) to the right hand side, we have
) ( ) ( [
) ? ( ) ? ( ) ? (
x f x x f (T2.,2)
Be may easily solve e%. (T2.,2) for #
(?)
according toF
{ } ) ( ) ( [
) ? (
+
) ? ( ) ? (
x f x f x

(T2.,:)
>ecause f L(#
(?)
) in e%. (T2.,:) is scalar, itLs inverse is very easily evaluated using simple division so thatF
+1
Draphical
illustration
of 89
iteration +
for a single
variable
root finding
problem.
T/) for
scalar case.
Appro#imation
depends on a
good guess.
Morrection formula.
) ( [
) (
) ? (
) ? (
) ? (
x f
x f
x

(T.,;)
)%uation (T2.,:) provides the basis for the update formula to be used in the first iteration of the scalar 89
method. This update formula isF
) ( [
) (
) ? (
) ? (
) ? ( ) ? ( ) ? ( ) + (
x f
x f
x x x x

+ + (T2.1?)
and from e%. (T2.,:), we may infer the update formula for any particular iteration asF
) ( [
) (
) (
) (
) ( ) ( ) ( ) + (
j
j
j j j j
x f
x f
x x x x

+ +
+
(T2.1+)
8e#t we develop the update formula for the case where we have n e%uations and n unknowns. Be call this
the multidimensional case.
8ewton 9aphson for the (ultidimensional MaseF
Assume we have n nonlinear algebraic e%uations and n unknowns characteri'ed by f(#)G?, and that we
have guessed a solution #
(?)
. Then f(#
(?)
)? because #
(?)
is Aust a guess. >ut there must be some #
(?)
which
will make f(#
(?)
H #
(?)
)G?. Again, we e#pand the function f(#) in a Taylor series, as followsF
? ... ) )( ( [ [
,
+
) ( [ ) ( ) (
, ) ? ( ) ? (
+
) ? ( ) ? (
+
) ? (
+
) ? ( ) ? (
+
+ + + + x x f x x f x f x x f
? ... ) )( ( [ [
,
+
) ( [ ) ( ) (
, ) ? ( ) ? (
,
) ? ( ) ? (
,
) ? (
,
) ? ( ) ? (
,
+ + + + x x f x x f x f x x f (T2.1,)
? ... ) )( ( [ [
,
+
) ( [ ) ( ) (
, ) ? ( ) ? ( ) ? ( ) ? ( ) ? ( ) ? ( ) ? (
+ + + + x x f x x f x f x x f
n n n n
)%uations (T2.1,) may be written more compactly as
? ... ) )( ( [ [
,
+
) ( [ ) ( ) (
, ) ? ( ) ? ( ) ? ( ) ? ( ) ? ( ) ? ( ) ? (
+ + + + x x f x x f x f x x f
(T2.11)
Assuming the guess is a good one such that #
(?)
is small, then the higher order terms are also small and we
can write
? ) ( [ ) ( ) (
) ? ( ) ? ( ) ? ( ) ? ( ) ? (
+ + x x f x f x x f (T2.15)
Ene reasonable %uestion to ask at this point isF \ust what is fL(#
(?)
) Z! That is, what is the derivative of a
vector&valued function of a vectorZ /ince we have n functions and n variables, we could compute a
derivative for each individual function with respect to each individual unknown, like fk(#)@#A, which gives
the derivative of the k
th
function with respect to the A
th
unknown. Thus, there will be a number of such
derivatives e%ual to the product of the number of functions by the number of unknowns, in this case, nn.
Thus, it is convenient to store all of these derivatives in a matri#. This matri# has become %uite well&known
+5
Npdate formula.
Bhat is fL(#)Z
as the )acobian matri#, and it is often denoted using the letter \. >ut how should the nn derivatives be
stored in this matri# \Z
The rows of \ should be ordered in the same order as the functions, that is, the k
th
row should contain the
derivatives of the k
th
functions. 0n e%. (T2.15), since the product fL(#
(?)
) #
(?)
must provide a correction to
the function f(#
(?)
H#
(?)
), i.e., since f(#
(?)
) G fL(#
(?)
) #
(?)
, it must be the case that any row of the matri# \
must be ordered so that the term in the A
th
column contains a derivative with respect to the A
th
unknown of
the vector #.
The reasoning in the last paragraph suggests that we write the \acobian matri# asF
1
1
1
1
1
1
1
1
]
1
n
n n n
n
n
x
x f
x
x f
x
x f
x
x f
x
x f
x
x f
x
x f
x
x f
x
x f
)
) ( ) ( ) (
) ( ) ( ) (
) ( ) ( ) (
) ? (
,
) ? (
+
) ? (
) ? (
,
,
) ? (
,
+
) ? (
,
) ? (
+
,
) ? (
+
+
) ? (
+
(T2.16)
0n e%. (T2.15), taking f(#
(?)
) to the right hand side, we have
) ( ) ( [
) ? ( ) ? ( ) ? (
x f x x f (T2.17)
or, in terms of the \acobian matri# \, we haveF
) (
) ? ( ) ? (
x f x ) (T2.12)
/olving e%. (T2.12) for #
(?)
, we haveF
{ } ) ( ) ( ) ( [
) ? ( + ) ? (
+
) ? ( ) ? (
x f ) x f x f x

(T2.1:)
)%uation (T2.1:) provides the basis for the update formula to be used in the first iteration of the multi&
dimensional case. This update formula isF
) (
) ? ( + ) ? ( ) ? ( ) ? ( ) + (
x f ) x x x x

+ (T2.1;)
and from e%. (T2.1;), we may infer the update formula for any particular iteration asF
) (
) ( + ) ( ) ( ) ( ) + ( i i i i i
x f ) x x x x
+
+ (T2.5?)
For problems of relatively small dimension, where the inverse of the \acobian is easily obtainable, e%.
(T2.5?) is an appropriate update formula. 0n general, however, it is a good rule, in programming, to always
avoid matri# inversion if at all possible, because for high&dimension problems, as is usually the case for
large scale power networks, matri# inversion is very time consuming. Be always want to avoid matri#
inversion if possible, and it usually is.
To see how to avoid matri# inversion, we will state the update formula a little differently. To do this, we
write e%. (T2.5?) as
+6
The \acobian matri#X
The correction formula
for first iteration.
The update formula.
>ut to avoid
matri#
inversion, we
write the update
formula like
this
where the
correction is
computed from
this
) ( ) ( ) + ( i i i
x x x +
+
(T2.5+)
where #
(i)
is found from
) (
) ( ) ( i i
x f x ) (T2.5,a)
)%uation (T2.5,a) is a very simple relation. Ebserving that \ is Aust a constant n]n matri#, #
(i)
is an n]+
vector of unknowns, and f(#
(i)
) is an n]+ vector of knowns, we see that e%. (T2.5,) is Aust the linear matri#
e%uation
A 'Gb (T2.5,b)
There are a very many methods of solving (T2.5,b). Be will cover this topic later in these notes. First,
however, letLs illustrate the 8ewton&9aphson procedure for a multi&dimensional case. Be will use the
simplest multi&dimensional case we can, a two&variable problem.
4(am%le T7.,
/olve the following two e%uations algebraically and using 89F
,#+
,
H#+#,&#+&,G?,
#+
,
&#,G?
The steps for the algebraic solution are to first solve both e%uations for #,, resulting in #,G(&,#+
,
H#+H,)@#+
and #,G#+
,
. )%uating these two e#pressions for #,, and manipulating, results in a cubic #+
1
H,#+
,
&#+&,G?. This
e#pression may be factored asF (#+&+) (#+H+)(#+H,)G?, and we see that the solutions to the cubic in #+ are +,
&+ and 3,. Plugging these values for #+ back into either e#pression for #, yields, respectively, +, +, and 5,
and therefore there are three solutions to the original problemC they areF (#+, #,)G(+,+), (&+,+), (&,,5).
8ow letLs solve this same problem using 89Z
Qefine the functions f+(#+,#,)G ,#+
,
H#+#,&#+&, and f,(#+,#,)G #+
,
&#,. Then the \acobian matri# isF
1
]
1
1
1
1
1
]
1
+ ,
+ ,
) , ( ) , (
) , ( ) , (
+
+ , +
,
, + ,
+
, + ,
,
, + +
+
, + +
x
x x x
x
x x f
x
x x f
x
x x f
x
x x f
)
$etLs act like we do not know the solution and guess at (#+
(?)
, #,
(?)
)G(?.;,+.+). Then the \acobian \, evaluated
at this guessed solution, is
1
]
1
1
]
1
+ : . +
; . ? ; . +
+ ,
+ ,
) , (
) ? ( +
+ , + ) ? (
,
) ? (
+
x
x
x x x
x x )
0nverting the \acobian results inF
1
]
1
1
]
1
1
]
1
61;22 . ? 6++17 . ?
,667: . ? ,:5?; . ?
; . + : . +
; . ? +
6, . 1
+
+ : . +
; . ? ; . +
+
+
)
Be also need to evaluateF
1
]
1
1
]
1
1
]
1
,; . ?
++ . ? , ,
) (
) (
) (
) + . + , ; . ? (
,
,
+
+ , + +
) ? (
,
) ? (
+
) ? (
) ? (
x
x x
x x x x
x f
x f
x f
Be can now update the solution using e%. (T2.5?), as
+7
A system of
linear
e%uationsW
1
]
1
1
]
1
1
]
1
1
]
1
1
]
1
1
]
1
1
]
1
+

;;;2+7 ?
??61;2 +
+??,:5 ?
+?61;2 ?
+ +
; ?
,; ?
++ ?
61;22 ? 6++17 ?
,667: ? ,:5?; ?
+ +
; ?
) (
) + (
,
) + (
+
) ? ( + ) ? ( ) ? ( ) ? ( ) + (
.
.
.
.
.
.
.
.
. .
. .
.
.

x
x
x f ) x x x x
Be see that the first update results in a solution that is very close to the actual solution of (+,+). This good
performance is due to the fact that we made a good initial guess. The student should repeat the above
procedure, but try starting from other points, e.g., (&?.;,+.+), (&+.;,5.+), and (?,+.+), using two iterations
each time. Briting a simple program will greatly reduce the effort.
0n general, of course, we usually need to iterate several times in order to obtain a satisfactory solution. "ow
many times is enoughZ The 89 algorithm must employ a stopping criterion in order to determine when the
solution is satisfactory. There are two ways to do this.
Type + stopping criterionF Test the ma#imum change in the solution elements from one iteration to the
ne#t, and if this ma#imum change is smaller than a certain predefined tolerance, then stop. This means
to compare the ma#imum absolute value of elements in # against a small number, call it +. 0n
e#ample (T2.1), # G *&?.+?61;2, ?.+??,:5-
T
, so the ma#imum absolute value of elements in # is
?.+?61;2. 0f we had +G?.+6, we could stop. >ut if we had +G?.?6, we would need to continue to the
ne#t iteration.
Type , stopping criterionF Test the ma#imum value in the function elements of the most current
iteration f(#), and if this ma#imum value of elements in f(#) is smaller than a certain predefined
tolerance, then stop. This means to compare the ma#imum absolute value of elements in f(#) against a
small number, call it +. 0n e#ample (T2.1), f(#)G*&?.++, &?.,;-
T
, so the ma#imum absolute value of
elements in f(#) is ?.,;. 0f we had +G?.1, we could stop. >ut if we had +G?.,, we would need to
continue to the ne#t iteration. This is the most common stopping criterion for power flow solutions,
and the value of each element in the function is referred to as the power mismatch! for the bus
corresponding to the function. For type P= buses, we test both real and reactive power mismatches.
For type P. buses, we test only real power mismatches.
T7.7 '%%lication o* 2 to +ower "low Solution
$etLs revisit the power flow problem outlined in /ection T2.6, in light of the 89 solution procedure
described in /ection T2.7. Be desire to solve e%. (T2.,5), with the vector of unknowns are given by e%.
(T2.,+) and the functions are in the form of e%. (T2.+;). These e%uations are repeated here for convenienceF
?
?
?
?
?
) (
) (
) (
) (
) (
) (
) (
) (
) (
+
,
+ +
, ,
+ ,
+
+
1
]
1
1
1
1
1
1
1
1
1
1
]
1
1
1
1
1
1
1
1
1
1
]
1
1
1
1
1
1
1
1
1
1
]
1
1
1
1
1
1
1
1
1
1
]
1

+ + +

Q
P
Q
Q
P
P
Q x Q
Q x Q
P x P
P x P
x f
x f
x f
x f
x f
!
!
!
! !
! !
! !
! !
!
!
# # #
#
(T2.,5)
+2
Ene of these
two stopping
criterion may be
applied.
Type +F #
Type ,F f(#)
Type , is power
mismatch! and
gives indication
of where
convergence
problems may lie.
(ismatch
vector.
1
1
1
1
1
1
1
1
1
1
]
1
1
1
1
1
1
1
1
1
1
1
]
1
1
]
1
+
+
+ ,
+
+
,
+
,
+
1
,
#
#
#
#
! !
!
!
!
!
!
!
!
x
x
x
x
x
x
% %V
% %V
% %V
&
&
&
% V %
&
(T2.,+)
( )
( ) ! ! k B # V V Q
! k B # V V P
#
!
j
j k kj j k kj j k k
!
j
j k kj j k kj j k k
,..., + , ) cos( ) sin(
,..., , , ) sin( ) cos(
+
+
+
+

(T2.+;)
The solution update formula is given by e%. (T2.5?), repeated here for convenienceF
) (
) ( + ) ( ) ( ) ( ) + ( i i i i i
x f ) x x x x
+
+ (T2.5?)
Mlearly, an essential step in applying 89 to the power flow problem is to enable calculation of the \acobian
elements, given for the general case by e%. (T2.16) as
1
1
1
1
1
1
1
1
]
1
n
n n n
n
n
x
x f
x
x f
x
x f
x
x f
x
x f
x
x f
x
x f
x
x f
x
x f
)
) ( ) ( ) (
) ( ) ( ) (
) ( ) ( ) (
) ? (
,
) ? (
+
) ? (
) ? (
,
,
) ? (
,
+
) ? (
,
) ? (
+
,
) ? (
+
+
) ? (
+
(T2.16)
)valuation of these elements is facilitated by the recognition, from e%. (T2.,5), that there are only two
kinds of e%uations (real power e%uations and reactive power e%uations), and from e%. (T2.,+), that there are
only two kinds of unknowns (voltage angle unknowns and voltage magnitude unknowns). Therefore, there
are only four basic types of derivatives in the \acobian. Be denote four sub&matrices
corresponding to these four basic types of derivatives as \
P
, \
=
, \
P.
, \
=.
, where the first
superscript indicates the type of e%uation we differentiate, and the second superscript indicates
the unknown with respect to which we differentiate. Therefore,
+:
/olution vector.
PF e%uations to
solve.
\acobian.
n]n differentiationsZ
EbservationF
Enly , kinds of
e%uations (P, =)
Enly , kinds of
unknowns (K.K, I)
Mompact e#pression of
\acobian using above
observation.
/till n]n differentiations,
but are differentiations of
same formZ
/tudy dimensionality
of each sub&matri#.

1
1
1
1
]
1

) ( ) ( ) + ( ) (
) ( ) + ( ) + ( ) + (
) + , ( ) + , (
# # #
#
# #
! ! ! !
QV
! ! !
Q
! ! !
PV
! !
P
! ! ! !
) )
) )
)
(T2.51)
The numbers above each sub&matri# in e%. (T2.51) indicate its dimensions, which can be inferred by
identifying the number of e%uations of that type (the number of rows of the sub&matri#) and the number of
unknowns of that type (the number of columns of the sub&matri#). Be may then identify an individual
element of each sub&matri# asF
k
j QV
jk
k
j PV
jk
k
j Q
jk
k
j P
jk
V
Q
)
V
P
)
Q
)
P
)

(T2.55)
8otationally, observe that the element \Ak
P
is not the element in row A, column k of the submatri# \
P
, rather
it is the derivative of the real power inAection e%uation for bus A with respect to the angle of bus k. /ince the
swing bus is numbered +, the \acobian matri# will have \,,
P
as the element in row +, column +. The
situation is similar for the other submatrices.
The update e%uation (T2.5,a) is repeated here for convenienceF
) (
) ( ) ( i i
x f x ) (T2.5,a)
(ultiplying both sides by &+, we obtain
) (
) ( ) ( i i
x f x ) (T2.5,c)
Nsing (T2.,+), (T2.,5), and (T2.51) we can write (T2.5,c) as
1
]
1

1
]
1
1
1
]
1
Q
P
% V %
&
) )
) )
QV Q
PV P
(T2.5,d)
From (T2.5,d) we observe that
Q V % ) & )
P V % ) & )
QV Q
PV P
+
+
K
K
(T2.5,e)
To get the needed derivatives, it is helpful to more e#plicitly write out the functions of e%. (T2.,5). They
areF
+;
/ubscripts
refer to bus
numbers, not
matri#
location.
/oftware needs
to track
relation
between bus
numbers ^
matri# location
/ubscripts A, k
are not row ^
column
number due to
(a) removing
pf e% ^ I for
bus +C (b)
stacking = e%s
^ .&variables
on top of P e%s
^ I variables.
( ) ( ) ( )
( ) ( ) ( )
( ) ( ) ( )
( ) ( ) ( )
1
1
1
1
1
1
1
1
1
1
1
1
1
]
1
+
+

+
+
1
1
1
1
1
1
1
1
1
]
1
+ + + + +
+ +
! j ! j ! j ! j ! j
!
j
!
! j ! j ! j ! j ! j
!
j
!
! j ! !j j ! !j j
!
j
!
j j j j j
!
j
! !
! !
! !
P B # V V
P B # V V
P B # V V
P B # V V
Q x Q
Q x Q
P x P
P x P
x f
# # # # # #
# #

sin sin
sin sin
sin cos
sin cos
) (
) (
) (
) (
) (
, ,
+
+ + , + + , +
+
+
+
, , , , ,
+
,
+ +
, ,
(T2.56)
/o each of the four sub&matrices of e%. (T2.51) has elements given by the e#pressions of e%. (T2.55),
respectively. These e#pressions are evaluated by taking the appropriate derivatives of the functions in e%.
(T2.56). Ene might think that this represents a formidable problem, since, based on e%. (T2.51), we have
(,8&+&8D)(,8&+&8D) elements in the \acobian and therefore the same number of derivatives to evaluate. A
typical power flow model for a N/ control area might have 6??? nodes (8G6???) and +??? generators
(8DG+???), resulting in a ;:;:;:;: \acobian matri# containing ;2,;2?,5?5 elements, with each element
re%uiring a differentiation of a function like those represented in e%. (T2.56). For a power flow model
having 6???? nodes and 6??? generators, the dimension is ;5;;:;5;;:, giving ;,?,5,7??,??? elements.
Fortunately, all of the derivatives can be e#pressed by one of Aust a few differentiations. At first glance, one
might think that there would be four differentiations, one for each sub&matri#. "owever, for each sub&
matri#, the off&diagonal terms, with Ak, are e#pressed differently than the diagonal terms, with AGk.
Therefore, there are eight differentiations to perform. The student should attempt to obtain a few of these
e#pressions. 0n doing so, the following tips are helpful.
>efore differentiating, it is helpful to pull out from the summation the term that corresponds to the bus
inAection being computed.
Bhen differentiating a sum of terms with respect to a particular unknown, the resulting derivative will
be non&'ero only for those terms in which the unknown appears.
Bhen differentiating with respect to the angles, the chain rule must be properly applied to account for
the derivatives of the trigonometric functions and the arguments of those trigonometric functions.
)ach of the functions appear in the form of f(#)Gg(#)&A. >ecause A is a constant (represented by P,,
X, P8 and =8gH+,X, =8 in e%. (T2.56)), it has no effect on the resulting derivatives.
The resulting e#pressions are given below.
( ) ) cos( ) sin(
) (
k j jk k j jk k j
k
j P
jk
B # V V
x P
)
(T2.52)
,
) (
) (
j jj j
j
j P
jj
V B x Q
x P
)
(T2.5:)
( ) ) sin( ) cos(
) (
k j jk k j jk k j
k
j Q
jk
B # V V
x Q
)
(T2.5;)
,?
;: million
differentiations
for a 6??? bus
model.
; billion
differentiations
for a 6???? bus
model.
: basic
derivative
e#pressionsF for
each of 5
submatrices, we
need diagonal
^ off&diagonal
e#pressions.
Ebserve T2.52 ^ T2.61
are similar. These are
off&diagonal elements
in \
PI
^ \
=.
, bolded
below.
1
]
1

1
]
1
1
1
]
1
Q
P
% V %
&
QV
P
J
J
Q
PV
J
J
"ow many elements is
thisZ Man we do
something smart hereZ
(ismatch
vector.
,
) (
) (
j jj j
k
j Q
jj
V # x P
x Q
)
(T2.6?)
( ) ) sin( ) cos(
) (
k j jk k j jk j
k
j PV
jk
B # V
V
x P
) +
(T2.6+)
j jj
j
j
j
j PV
jj
V #
V
x P
V
x P
) +
) ( ) (
(T2.6,)
( ) ) cos( ) sin(
) (
k j jk k j jk j
k
j QV
jk
B # V
V
x Q
)
(T2.61)
j jj
j
j
j
j QV
jj
V B
V
x Q
V
x Q
)
) ( ) (
(T2.65)
Be are now in a position to provide the algorithm for using 89 to solve the power flow problem. >efore
doing so, it is helpful to more e#plicitly define the mismatch vector, from e%. (T2.,5) or (T2.56) asF
?
) (
) (
) (
) (
) (
) (
) (
) (
) (
+
,
+ +
, ,
+ ,
+
+
1
1
1
]
1
1
1
1
1
1
1
1
1
1
]
1
1
1
1
1
1
1
1
1
1
]
1
1
1
1
1
1
1
1
1
1
]
1

+ + +

Q
P
Q
Q
P
P
Q x Q
Q x Q
P x P
P x P
x f
x f
x f
x f
x f
!
!
!
! !
! !
! !
! !
!
!
# # #
#
(T2.61)
The 89 algorithm, for application to the power flow problem, isF
+. /pecifyF
All admittance data
Pd and =d for all buses
Pg and K.K for all P. buses
K.K for swing bus, with G?
,. $et the iteration counter AG+. Nse one of the following to guess the initial solution.
Flat /tartF .kG+.? ? for all buses.
"ot /tartF Nse the solution to a previously solved case for this network.
1. Mompute the mismatch vector for #
(A)
, denoted as f(#) in e%. (T2.,5) and e%. (T2.56). 0n what follows,
we denote elements of the mismatch vector as Pk and =k corresponding to the real and reactive
power mismatch, respectively, for the k
th
bus (which would not be the k
th
element of the mismatch
vector for two reasonsF one reason pertains to the swing bus and the other reason to the fact that for
type P= buses, there are two e%uations per bus and not one 3 see bo#ed comments ne#t to e%. T2.55).
This computation will also result in all necessary calculated real and reactive power inAections.
5. Perform the following stopping criterion testsF
0f KPkKS P for all type P= and P. buses and
0f KK=kKS = for all type P= buses,
Then go to step 7
Etherwise, go to step 6.
6. Find an improved solution as followsF
)valuate the \acobian \ at #

(A)
. Qenote this \acobian as \
(A)
/olve for #
(A)
fromF
,+
Ebserve T2.5; ^ T2.6+
are similar. These are
off&diagonal elements
in \
=I
^ \
P.
, bolded
below.
1
]
1

1
]
1
1
1
]
1
Q
P
% V %
&
QV
P
J
J
;<
J
PV
J
"ow many elements is
thisZ Man we do
something smart hereZ
Flat start@hot
startF common
industry terms
Mompute
mismatch
vector.
Mheck
stopping
criterion
"actori=ationF method to
solve linear simultaneous
e%uations. Be will look a
little more at this issue.
[ ]
1
1
1
]
1

1
1
1
]
1

Q
P
) x
Q
P
x )
j j j
or
+ &
(A) ) ( ) ( ) (
where we must use *actori=ation with the left e%uation if the system is large, but if the system is
not large, we may use the right hand e%uation.
Mompute the updated solution vector as #
(AH+)
G #
(A)
H #
(A)
.
9eturn to step 1 with AGAH+.
7. /top.
The above algorithm is applicable as long as all P. buses remain within their reactive limits. To account for
generator reactive limits, we must modify the algorithm so that, at eac0 iteration, we check to ensure P.
bus reactive generation is within its limits (see /ection T2.+ regarding modeling of reactive limits). 0n this
case, steps +&5 remain e#actly as given above, but we need a new step 6 and 7, as followsF
6. Mheck reactive limits for all generator buses as followsF
a. For all type P. buses, perform the following testF
0f =gkT=gk,ma#, then
=gkG=gk,ma# and M"A8D) bus k to a type P= bus (see step 7a)
0f =gkS =gk,min, then
=gkG=gk,min and M"A8D) bus k to a type P= bus (see step 7b)
b. For all type P= generator buses, perform the following testF
0f =gkG=gk,ma# and K.kKTK.k,setK or if =gkG=gk,min and K.kKSK.k,setK, then
M"A8D) this bus back to a type P. bus (see step 7b)
7. 0f there were no M"A8D)/ in /tep 6, then stop. 0f there were one or more M"A8D)/ in step 6, then
modify the solution vector and the mismatch vector as followsF
a. For each M"A8D) made in step 6&a (changing a P. bus to a P= bus)F
8DG8D&+
0nclude the variable .k to the vector # and the variable .k to the vector #.
0nclude the reactive e%uation corresponding to bus k to the vector f(#).
(odify the \acobian by including a column to \
P.
and including a row to \
=
and \
=.
.
b. For each M"A8D) made in /tep 6&b (changing a P= gen bus back to a P. bus)F
8DG8DH+
9emove the variable .k to the vector # and the variable .k from the vector #.
9emove the reactive e%uation corresponding to bus k from the vector f(#).
(odify the \acobian by removing a column to \
P.
and removing a row from \
=
and \
=.
.
After modifications have been made for all M"A8D)/, go back to /tep 5.
Bhen the algorithm stops, then all line flows may be computed using
R R R
- *
jk k j j jk j jk
y V V V I V "
4(am%le T7.1 *6- (used with permission of .. .ittal)
Find *, V
3
, +, /D+, and =D, for the system shown in Fig. T2.:. 0n the transmission system all the shunt
elements are capacitors with an admittance yc G j?.?+, while all the series elements are inductors with an
impedance of ,- G j?.+.
,,
0s there a P. bus
that should be P=Z
0s there a P= gen&bus, which was
a P. bus, that should be now be
P. againZ
Another way is to let algorithm
converge, check limits, make
changes, ^ repeat. >ut checking
every iteration provides increased
convergence robustness.
Qoes this e%uation account for
shunt elements or line chargingZ
/hould itZ

K.
,
KG+.?6
/
D+
P
D,
G?.777+
.
+
G+?
.
1

/
Q1
G,.:761HA+.,,55
"i#. T7.>$ T0ree 8us System *or 4(am%le T7.1
Solution$ The admittance matri# for the system shown in Fig. )+?.7 is given by
1
1
1
]
1
;: . +; +? +?
+? ;: . +; +?
+? +? ;: . +;
4
j j j
j j j
j j j
>us + is the swing bus. >us , is a P. bus. >us 1 is a P= bus. Be use the 89 method in the solution. The
unknown variables are ,, 1, and K.1K. Thus, we will need three e%uations, and the \acobian is a 1 # 1
matri#.
Be first write e%. (T2.56) for the case at hand, putting in the known values of K.+K, K.,K, +, and the >iALs.
8ote that since we have neglected line resistance in the problem statement, all DiALs are 'ero.
P
2
(x) V
2
V
1
B
21
sin(
2

1
) + V
2
V
3
B
23
sin(
2

3
)
= 10.5sin
2
+10.5 V
3
sin(
2

3
)
(T2.65a)
P
3
(x) V
3
V
1
B
31
sin(
3

1
) + V
3
V
2
B
32
sin(
3

2
)
= 10.0 V
3
sin
3
+10.5 V
3
sin(
3

2
)
(T2.65b)
The e%uation for =,(#) will not help since we do not know the reactive inAection for bus ,, and its inclusion
would bring in the reactive inAection on the left&hand side as an additional unknown. >ut this loss of an
e%uation is compensated by the fact that we know K.,K (and this will always be the case for a type P. bus).
/o we do not need to write the e%uation for =,(#). 4et, because bus 1 is a type P= bus, we do know its
reactive inAection, and so we will know the left hand side of the reactive power flow e%uation. This is
fortunate, since we do not know K.1K (and this will always be the situation for a type P= bus).
Q
3
(x) V
3
V
1
B
31
cos(
3

1
) + V
3
V
2
B
32
cos(
3

2
) + V
3
2
B
33 [ ]
= - 10 V
3
cos
3
+10.5 V
3
cos(
3

2
) 19.98V
3
2
[ ]
(T2.65c)
,1
EbserveF (+) no real
parts due to neglect of
resistanceC (,) diagonals
are negative (inductive
admittance), (1) off&
diagonals are positive
(negated inductive
admittance)
Bhat are e%uations to useZ
Bhat are unknownsZ
Bhat is \acobian dimensionZ
Bhat would \acobian
dimension be if bus ,
changed to P=Z
The update vector and \acobian matri# isF
1
1
1
]
1

1
1
,
#
V

1
1
1
1
1
1
1
]
1
1
1
1
1
,
1
1
1
1
1
,
1
1
,
1
,
,
,
V
Q Q Q
V
P P P
V
P P P
)
Be obtain the various partial derivatives for the \acobian from e%s. (T2.65a,b,c)F
( ) ( ) ( )
1 , 1 , 1 , ,1 1 , + , ,+ + ,
,
,
cos 6 . +? +?.6cos G cos cos
+ + V B V V B V V
P
( ) ( )
1 , 1 1 , ,1 1 ,
1
,
cos &+?.6 G cos
V B V V
P
( ) ( )
1 , 1 , ,1 ,
1
,
+?.6sin G sin
B V
V
P
( )
, 1 1
,
1
cos 6 . +?
V
P
( )
, 1 1 1 1
1
1
cos 6 . +? cos ? . +?
+ V V
P
( )
, 1 1
1
1
sin 6 . +? sin +?
+
V
P
( ) ( )
, 1 1 , 1 , 1
,
1
sin 6 . +? sin +?
V V V
Q
( ) ( )
, 1 1 1 1 , 1 , 1 1 1
1
1
sin 6 . +? sin +? G sin +? sin +?
+ + V V V V V
Q
( ) ( ) [ ]
( ) [ ]
1 , 1 1
11 1 11 1 1, 1, , 1+ 1+ +
1
1
;7 . 1; cos 6 . +? cos +? G
cos cos
V
B V B V B V B V
V
Q
+
+ +

Be are ready to start iterating using (T2.5?). Be note that the inAections, to be used on the left hand side of
e%s. (T2.65a,b,c) are P, G PD, G ?.777+, P1 G &PQ1 G &,.:761, and =1 G &=Q1 G &+.,,55C these %uantities
,5
0nAections come from
one&line diagram.
remain constant through the entire iterative process. Be use a flat startC therefore our initial guess is
,G1G? and K.1KG+.?.Nsing e%s. (T2.61) and (T2.65a,b,c) we getF
( )
( )
( ) 1
1
1
]
1
1
1
1
]
1

1
1
1
]
1
1
1
1
]
1
1
1
1
]
1
1
1
1
]
1
2?55 . ?
:761 . ,
777+ . ?
,,55 . +
:761 . ,
777+ . ?
6, . ?
?
?
#
#
#
) (
1
1
,
) ? (
1
) ? (
1
) ? (
,
) ? (
1
1
,
) ? (
Q
P
P
Q
P
P
Q
P
P
x f
As e#pected for a flat start, the mismatch is large. 8e#t we calculate the \acobian matri#F
1
1
1
]
1
57 . +; ? ?
? 6 . ,? 6 . +?
? 6 . +? ,+
) (T2.66)
8ote that the \acobian sub&matrices \
P.
and \
=
are both filled with 'eros. This is because when resistance is
neglected, these derivatives depend on sin terms, and because this is the first iteration of a flat start, all
angles are 'ero and therefore the sin terms are all 'ero.
As mentioned, commercial power flow programs normally use $N factori'ation to obtain the update. 0n this
case, however, because of the low dimensionality, we may invert the \acobian. Taking advantage of the
block diagonal structure, we haveF
( )
[ ] [ ]
1
1
1
]
1
1
1
1
]
1
1
]
1
1
]
1
?6+5 . ? ? ?
? ?767 . ? ?1,: . ?
? ?1,: . ? ?75? . ?
57 . +; ? ?
?
?
6 . ,? 6 . +?
6 . +? ,+
+
+
+
)
8ow we computeF
[ ]
1
1
1
]
1
1
1
1
]
1
1
1
1
]
1

1
1
1
1
1
]
1

1
1
1
]
1

?17, . ?
+77? . ?
?6+1 . ?
2?55 . ?
:761 . ,
777+ . ?
?6+5 . ? ? ?
? ?767 . ? ?1,: . ?
? ?1,: . ? ?75? . ?

K K
1
1
,
+ &
1
1
,
) (
Q
P
P
)
V
x
j
The elements of the update vector corresponding to angles are in radians. Be can easily convert them to
degreesF
1
1
1
]
1
1
1
1
]
1
1
1
1
]
1
?.?17, &
;.6+1; &
;1;7 . ,
?.?17, &
rad ?.+77? &
rad ?6+1 . ?
o
o
1
1
,
V
Be now find #
(+)
as followsF
1
1
1
]
1
1
1
1
]
1
+
1
1
1
]
1
+
?.;71:
;.6+1; &
;1;7 . ,
?.?17, &
;.6+1; &
;1;7 . ,
+
?
?
# # #
o
o
o
o
) ? ( ) ? ( ) + (
,6
PN mismatch vector.
Ebserve mismatch is
very large.
\
P.
and \
=
are both
filled with 'eros,
because with 'ero
resistance, they have
only sin terms.
"ere we inverted ) but for larger systems,
you should always use factori'ation.
Angular measures
are in radians.
Be note that the e#act solution is
1
1
1
]
1
;5;; . ?
?+ . +?
?? . 1
o
o
, so this is pretty good progress for one iterationW
Be proceed to the ne#t iteration using the new values ,
(+)
G&,.;1;7, 1
(+)
G&;.6+1;, and K.1K
(+)
G?.;71:.
/ubstituting in e%. (T2.65a), we get P,(#
(+)
) G ?.7,?,, and thus P,
(+)
G ?.7,?,&?.777+G&?.?56;. /imilarly,
using e%s. (T2.65b) and (T2.65c), we get the updated mismatch vectorF
1
1
1
]
1
1
1
1
]
1
,,6+ . ?
++56 . ?
?571 . ?
1
,
+
Q
P
P
8ote that, in Aust one iteration, the mismatch vector has been reduced by a factor of about +?. Malculating
J using the updated values of the variables, we find that
1
1
1
]
1
,+;; . +: 26?: . , +6:, . +

:65+ . , 66:; . +; ?615 . +?
,?+2 . + ?615 . +? 61;7 . ,?
)
The matri# should be compared with \ from the previous iteration. 0t has not changed much. The elements
in the off&diagonal matrices \
P.
and \
=
are no longer 'ero, but their elements are small compared to the
elements in the diagonal matrices \
P
and \
=.
. The diagonal matrices themselves have not changed much. 0t
is also important to note that the upper left&hand \acobian submatri# (\
PI
) is symmetric. This fact allows for
a significant savings in storage when dealing with large systems.
The updated inverse is
( )
1
1
1
]
1
?67+ . ? ??:5 . ? ???; . ?

??:2 . ? ?7;7 . ? ?117 . ?
??+? . ? ?117 . ? ?76+ . ?
+
)
Momparing this inverted \acobian with that of the last iteration, we do not see much change. Nsing the
same procedure as before to calculate the update vector, we obtain
1
1
1
]
1
1
1
1
]
1
?.;6?,
;.;;,5 &
??,1 . 1
#
o
o
1
1
,
) , (
V
This is very close to the correct answer. The largest error is only about ?.?:<. Ef course in the usual
problem we do not know the answer and we would continue into the ne#t iteration. Be would stop the
iterations when the mismatch vector satisfies the re%uired tolerance. Be would findF
1
1
1
]
1
1
1
1
]
1
??1+ . ?
??,1 . ?
??+; . ?
,
1
1
,
Q
P
P
,7
8ow the \
P.
and \
=
terms are non&'ero since angles
are no longer all 'ero. >ut it is interesting that this
\acobian differs very little from initial one, below.
1
1
1
]
1
57 . +; ? ?
? 6 . ,? 6 . +?
? 6 . +? ,+
)
The mismatch has been reduced from that of the last iteration by a factor of +?? and is small enough. En
that basis we can stop here. /o we stop with the values , G &1.??,1
o
, 1 G &;.;;,5
o
, and K.1KG?.;6?,. 0t
remains to calculate the real and reactive power generation at the swing bus (bus +) and the reactive power
generation at the P. bus (bus ,) using the calculated values of ,, 1, and K.1K.

( ) ( )
+;:2 . , sin;.;;,5 6?, . ; ?,1 +?.6sin1.? G
sin sin
o o
1 + +1 1 + , + +, , + + +
+
+ B V V B V V P P
#

( ) ( ) [ ]
[ ]
?.+176 G
;: . +; ;;,5 ;.6?,cos;. H ?,1 +?.6cos1.? & G
cos cos
o o
++
,
+ 1 + +1 1 + , + +, , + + +
+ + B V B V V B V V Q Q
#

( ) ( ) [ ]
( ) ( ) [ ]
+.71;6 G
?,: . ,, 7.;;?+ cos ;22 . ; 1.??,1 & +?.6cos & G
cos cos
o o
,,
,
, 1 , ,1 1 , + , ,+ + , , ,
+
+ + B V B V V B V V Q Q
#

This completes the e#ample.
T7.> 'dvanced issues associated wit0 %ower *low
/ome more advanced issues in relation to the power flow problem are listed below.
\acobian elements as sensitivities and what they tell you about relations between real or reactive power
inAection and voltage magnitude or angle.
/parsityF
o /parse characteristic of \acobian
o /torage implications
o Eptimal ordering
Qifferent types of power flow formulations@algorithmsF
o Qivide voltage magnitude part of update vector to reduce the \acobian storage
re%uirements
o Fast decoupled power flow
o Dovernor power flow
o QM power flow
o Dauss&/eidel
/ome advanced modeling issuesF
o TransformersF regulating transformers
o Area interchange
o /witched shunt capacitors
T7.? In%ut@9ut%ut and Commercial +ro#rams
There are a number of very high %uality commercial power flow programs on the market today, some of
which include those developed by the )lectric Power 9esearch 0nstitute ()P90), Power Technologies
0ncorporated (PT0), Eperation Technology, 0nc., and )Q/A. (ost of todayLs commercial software packages
are menu&driven from a Bindows environment. A predecessor of one of these was interactive from the
command line, and we use it here because its scroll&down recording serves well to illustrate some basic
features of most commercial programs. The following lists of Aust a few of these features. The item number
,2
PD+ and =D+ are swing
bus generation, a
function of the solution.
=D, is reactive generation
at P. bus Y,, also a
function of the solution.
in this list is denoted in the illustration that follows, using large ,? font numbers, to help identify the
corresponding parts in the session.
+. /olving the power flow
,. $imit checks 3 overloads
1. $imit checks 3 undervoltages
5. 9eporting the solution 3 flows (denoted below as bus flows!)
6. 9eporting the solution 3 generator operation information (denoted below as plant data)
A one&line diagram for the system on which this session was run is shown in Figure (T2.;). This system is a
test system developed by a subcommittee of the 0))) Power )ngineering /ociety (P)/) *+,-. 0t is referred
to as the 0))) 9T/L;7, and it consists of ,5 buses, ++ of which are voltage controlled busesC thus, +1 are
type P= at the beginning of a solution procedure.
"i#ure T7.? A T0e I444 elia&ility Test System B1)C
1. MAIN MENU
=========
1 LOAD THE PSF 8 DATA VERIFICATIO
2 IMPORTING UTILITIES 9 LIMIT CHECKS
3 INPUT DATA PROCESSING 10 SENSITIVITY FACTORS
4 STUDY PREPARATION UTILITIES 11 EPORTING UTILITIES
! PO"ER FLO" SOLUTION 12 CLOSE THE PSF
# NET"ORK REDUCTION 13 $UIT
% PO"ER FLO" SOLUTION REPORTING
ENTER MENU CHOICE& !
''' PO"ER FLO" SOLUTION '''
=============================
1 AUTOMATIC SOLUTION
2 FAST DECOUPLED ()*
3 NE"TON RAPHSON
4 FAST DECOUPLED ()*
! LOCALI+ED SOLUTION (FD )*
# DC PO"ER FLO" SOLUTION
% PO"ER FLO" SOLUTION OPTIONS
,:
ITEM ,- EIT 3
''' START OF NE"TON RAPHSON ITERATIONS '''
ITERATION & 1 NE"TON RAPHSON UNSOLVED A)SOLUTE ERROR
.............. )US "ITH LARGEST ERROR ............ )USES SUMMATION
P(P.U. M"* 0.0003%! ( 30 )US 3138. * 0 0.001%
V(RADIANS* 0.0001%9 ( %0 )US %138. * 8 0.001#
$(P.U. MVAR* 0.0018%3 ( 120 )US 12230. * 0 0.00%8
V(P.U. KV* 0.000210 ( 30 )US 3138. * 3 0.0009
ITERATION & 2 NE"TON RAPHSON UNSOLVED A)SOLUTE ERROR
.............. )US "ITH LARGEST ERROR ............ )USES SUMMATION
P(P.U. M"* 0.00000# ( 1#0 )US 1#230. * 0 0.0000
$(P.U. MVAR* .0.000008 ( 1%0 )US 1%230. * 0 0.0000
''' PO"ER FLO" SOLUTION IS REACHED IN 2 ITERATIONS '''
2. MAIN MENU
=========
1 LOAD THE PSF 8 DATA VERIFICATION
ENTER MENU CHOICE& 9
'' LOAD FLO" SOLUTION LIMIT CHECK ''
====================================
1 OVERLOADED LINES OR TRANSFORMERS
2 CONTROL PARAMETERS VIOLATION
3 )US VOLTAGE OUTSIDE SPECIFIED )AND
4 LINES ACROSS "HICH THE ANGLE ECEEDS THE SPECIFIED VALUE
! CHECK FOR ALL LIMITS AND VIOLATIONS
# SORT OPTIONS
ITEM ,- SPECIFY SU)SYSTEM- EIT 1
ALL LINES TRANSFORMERS& A
''' LOAD FLO" SOLUTION LIMIT CHECK '''
1 OVERLOADED )RANCHES RATING GROUP = 1 / LOADING = 100.0
0..... FROM )US .....1 0...... TO )US ......1 MVA MVA PERCENT
NUM)ER NAME NUM)ER NAME CKT LOADING RATING LOADING
====== ================ ====== ================ ======== ======= ====== =======
#0 )US #138. 100 )US 10138. 1 1%!.4 1%!.0 100.22
3. '' LOAD FLO" SOLUTION LIMIT CHECK ''
====================================
1 OVERLOADED LINES OR TRANSFORMERS
2 CONTROL PARAMETERS VIOLATION
3 )US VOLTAGE OUTSIDE SPECIFIED )AND
4 LINES ACROSS "HICH THE ANGLE ECEEDS THE SPECIFIED VALUE
! CHECK FOR ALL LIMITS AND VIOLATIONS
# SORT OPTIONS
HI VOLTAGE LIMIT LO VOLTAGE LIMIT& L
ENTER VALUE IN PU& .99
''' LOAD FLO" SOLUTION LIMIT CHECK '''
4 )USES "ITH V 0 0.990
0........ )US ........1 0.. VOLTAGE ..1
NUM)ER NAME TYPE )ASE KV MAG ANGLE AREA +ONE O"NER ID
====== ================ ==== ======= ======= ======= ==== ==== ================
30 )US 3138. 1 138.00 0.98#8 .2#.02 10 1 )LANK
40 )US 4138. 1 138.00 0.9888 .20.2% 10 1 )LANK
90 )US 9138. 1 138.00 0.9824 .1!.83 10 1 )LANK
120 )US 12230. 1 230.00 0.98%# .4.%3 10 1 )LANK
UP DO"N SCREEN DUMP LIST TO FILE CHANGE RATING EIT & 2
4. MAIN MENU
=========
1 LOAD THE PSF 8 DATA VERIFICATION
,;
ENTER MENU CHOICE& %
'' LOAD FLO" SOLUTION REPORTING ''
==================================
1 MISMATCH SUMMARY 11 ULTC3PS SUMMARY
2 SU)SYSTEM SUMMARY 12 SERIES COMPENSATORS
3 )US SHUNT DATA 13 PO"ER FLO" SUMMARY
4 PLANT DATA 14 SECTIONALI+ED )RANCHES
! MACHINE DATA 1! STATIC TAP CHANGERS 3 PHASE SHIFTERS
# AREA INTERCHANGE DATA 1# THREE "INDING TRANSFORMERS
% TIE LINE FLO"S 1% INTERFACE FLO"S
8 )US FLO"S 18 VOLTAGE PROFILE
9 DC CONVERTERS 19 SORT OPTIONS
10 DC FLO"S
SU)SYSTEM )USES PICK )USES EIT& 4
ENTER )US LIST FILE NAME OR )USES TO INCLUDE OR ECLUDE
1 #0
''' SOLUTION REPORTING . )US FLO"S '''
)US& #0 )US #138.
VOLTAGE & 1.001% PU ( 138.2 5V* .20.#%
)US NUM NAME AREA CKT M" MVAR MVA TAP
======== ================ ==== ======== ======== ======== = ======== ========
LOAD 13#.00 28.00 138.8!
S" SHUNT 0.00 100.33 100.33
TO 100 )US 10138. 10 1 .128.#3 .10%.#4 1#%.%3
TO 20 )US 2138. 10 1 .%.3% .20.#9 21.9#
UP DO"N NET )US PREVIOUS )US SCREEN DUMP LIST TO FILE EIT & 2
SU)SYSTEM )USES PICK )USES EIT& 4
ENTER )US LIST FILE NAME OR )USES TO INCLUDE OR ECLUDE
1 100
''' SOLUTION REPORTING . )US FLO"S '''
)US& 100 )US 10138.
VOLTAGE & 1.0132 PU ( 139.8 5V* .1#.1!
)US NUM NAME AREA CKT M" MVAR MVA TAP
======== ================ ==== ======== ======== ======== = ======== ========
LOAD 19!.00 40.00 199.0#
TO 120 )US 12230. 10 1 .234.48 #0.92 242.2% 1.0000LK
TO 110 )US 11230. 10 1 .23!.99 !2.21 241.%0 1.0000LK
TO 80 )US 8138. 10 1 #9.%% .3.%4 #9.8%
TO #0 )US #138. 10 1 130.9! .131.80 18!.80
TO !0 )US !138. 10 1 %4.%# .1%.!9 %#.80
UP DO"N NET )US PREVIOUS )US SCREEN DUMP LIST TO FILE EIT & 2
!. '' LOAD FLO" SOLUTION REPORTING ''
==================================
1 MISMATCH SUMMARY 11 ULTC3PS SUMMARY
2 SU)SYSTEM SUMMARY 12 SERIES COMPENSATORS
3 )US SHUNT DATA 13 PO"ER FLO" SUMMARY
4 PLANT DATA 14 SECTIONALI+ED )RANCHES
! MACHINE DATA 1! STATIC TAP CHANGERS 3 PHASE SHIFTERS
# AREA INTERCHANGE DATA 1# THREE "INDING TRANSFORMERS
% TIE LINE FLO"S 1% INTERFACE FLO"S
8 )US FLO"S 18 VOLTAGE PROFILE
9 DC CONVERTERS 19 SORT OPTIONS
10 DC FLO"S
''' SOLUTION REPORTING . PLANT DATA '''
=======================================
1 ALL PLANTS
2 ON LINE PLANTS
3 PLANTS AT VAR LIMIT "ITH UNE$UAL VAR LIMITS
4 PLANTS "ITH UNSCHEDULED REACTIVE PO"ER
! PLANTS CONTROLLING A REMOTE )US
# PLANTS "ITH RESERVE REACTIVE PO"ER
ITEM ,- EIT 1
''' SOLUTION REPORTING '''
10 PLANTS
1?
PLANT )US MACHINES
NUM NAME TYPE I3S O3S M" MVAR $MA $MIN
======== ================ ==== ======== ======== ======== ======== ========
R 10 )US 1138. 2 4 0 134.%0 41.!3 280.00 .!0.00
R 20 )US 2138. 2 4 0 18%.00 30.!0 140.00 .100.00
R %0 )US %138. 2 3 0 1#!.00 91.#3 280.00 0.00
R 130 )US 13230. 3 3 0 33%.!3 1%0.#4 !40.00 .300.00
R 1!0 )US 1!230. 2 # 0 18!.00 40.3# 310.00 .!!.00
R 1#0 )US 1#230. 2 1 0 1!!.00 10#.41 280.00 .!!.00
R 180 )US 18230. 2 1 0 400.00 100.%3 200.00 .90.00
R 210 )US 21230. 2 1 0 400.00 2!.2! 200.00 .90.00
R 220 )US 22230. 2 # 0 300.00 .80.!! #9#.00 .!#0.00
R 230 )US 23230. 2 3 0 ##0.00 33.34 310.00 .12!.00
UP DO"N OTHER SCREEN SCREEN DUMP LIST TO FILE EIT & 2
e*erences
*+- \. .lach and O. /inghal, Momputer (ethods for Mircuit Analysis and Qesign,! Oluwer Academic
Publications, +;;1.
*,- $. Mhua and P. $in, Momputer Aided Analysis of )lectronic MircuitsF Algorithms and
Momputational Techni%ues,! Prentice "all, +;26.
*1- E. )lgerd, )lectric )nergy /ystems Theory,! (cDraw&"ill, 8ew 4ork, +;:,.
*5- \. Drainger and B. /tevenson, Power /ystem Analysis,! (cDraw&"ill, 8ew 4ork, +;;5.
*6- A. >ergen and .. .ittal, Power /ystems Analysis,! second edition, Prentice "all, Npper /addle
9iver, 8ew \ersey, ,???.
*7- M. Dross, Power /ystem Analysis,! \ohn Biley ^ /ons, 8ew 4ork, +;2;.
*2- >. Beedy, )lectric Power /ystems,! third edition, \ohn Biley ^ /ons, $ondon, +;2;.
*:- ". /aadat, Power /ystem Analysis,! (cDraw&"ill, 8ew 4ork, +;;;.
*;- D. /tagg and A. )l&Abiad, Momputer (ethods in Power /ystem Analysis,! (cDraw&"ill, 8ew
4ork, +;7:.
*+?- D. "eydt, Momputer Analysis (ethods for Power /ystems,! (acmillan Publishing, 8ew 4ork,
+;:7.
*++- .. Qel Toro, )lectric Power /ystems,! Printice&"all, )nglewood Mliffs, 8ew \ersey, +;;,.
*+,- The reliability test system task force of the application of probability methods subcommittee,
The 0))) reliability test system 3 +;;7!, I... /rans. Power "ystems, .ol. +5, 8o. 1, pp. +?+?&
+?,?, +;;;.
1+

Power Flow

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Power Flow

Uploaded by

Copyright:

Available Formats

The Power Flow Problem

The Power Flow Problem

T7.1 T0e %ower *low e6uations

)valuate the \acobian \ at #

,+;; . +: 26?: . , +6:, . +

?67+ . ? ??:5 . ? ???; . ?

You might also like