OPTIMAL RESCHEDULING OF POWER FOR SYSTEM RELIABILITY s//:-11

A. Thanikachalam J. R. Tudor
/) c--- A a
c 6
,

Department of Electrical Engineering -t/ -/ '

University of Missouri-Columbia
A Columbia, Missouri
ABSTRACT

This paper presents a c omputational procedure for 1 ce of the power flow solution by Newton's method9
the on-line optimal resched uling of power generation. in.conjunction with optimally.ordered triangular fact-
Rescheduling is used to cor rect an unreliable system orizationl0 of the Jacobian matrix, makes it possible
operating state and bring a potentially dangerous con- to obtain a fast load flow solution, optimization
dition into a secure opera ting state. Sensitivity takes many iterations of load flow and therefore this
parameters are the basis of the model, and linear pro- procedure is not suitable-for on-line control of ser-
gramming formulation is deve loped to solve the optimal vice continuity. Having already obtaineda;an optimum
rescheduling problem. The constraints on the variab- dispatching solution for the nominal condition as a
les are naturally included in this practical method. result of solving the fairly sizable power network, it
A small power-network illu .strates the application of would be highly desirable and necessary if the slight-
the method. ly altered solution corresponding to branch (line)cur-
rent variations could be obtained from a simple compu-
INTRODUC TION tation instead of a complete recomputation of the time
consuming nonlinear program.,A solution to the optimal
Ever since Carpentier1 of Electricite' de France rescheduling problem is not guaranteed to exist, espe-
formulated the general and exact economic dispatch cially if the system is highly vulnerable and the con-
problem, there has been con siderable work done in the straints are narrow. If necessary, the constraints may
rigorous study of optimal real and reactive power be suitably relaxed within the limits of not endanger-
flow2,3,4,5. In the past, o ,ptimal scheduling of power ing much the system security, or else some of the in-
was studied separately from the security (reliability) terruptible loads must be curtailed. When a satisfac-
problem of the system. Qu ite justifiably, there has tory solution is reached, a secure operating condition
been efforts recently to c ombine the economic oper- is guaranteed. The method presented in this paper is
ation and the security aspec t of the power system and suitable for accomplishing this task in minimum time.
treat them as an integrated problem. 6,7
The system dispatcher stationed at the centrally
There are so many rea-sons, such as atmospheric operated digital computer would get all the informa-
weather conditions, outageS of components, dynamic tion monitored and would coordinate with the computer
(slowly varying) load changets and spinning reserve av- in obtaining the optimum solution.
ailability, for the reliabi lity of the system to con-
tinuously change. A method for calculating a relia- First,the fundamental power flow equations are
bility criterion for the nLear-term future has been given and the variables involved are classified into
discussed elsewhere8. Vario ,us research groups are now three groups of vectors. After developing the sensi-
actively engaged in develop ,ing mathematical formula- tivity matrix, the method of calculating the linear
tion to analyze the system reliability precisely. It relationships is explained. The objective function
would be implemented in the power-system control cen- and the constraints used in the linear programming
ters by a combination of feaLtures: a) data acquisition problems are formulated. A numerical example is work-
and system control, over re.liable communication chan- ed out to illustrate the application of this method.
nels,b)real-time computer anaLlyses of the state of sys-
tem reliability and system t rouble, c) new techniques POWER FLOW EQUATIONS
for estimating variable syst em data.
The power flow equations in an interconnected
In this paper, system reliability is assumed to system of N nodes is described by a set of 2N real
depend very heavily on the current loading (which is simultaneous equations:
~~ ~in ~
also generally a measure of the Power loading) the ~ ~ P =^ E N E.Y. Cos
(1)
r '

lines. Therefore, current in the particular lines 1 3i j=l ijij ijij

should be adjusted to the suitable values based upon
the results obtained by somLe reliability evaluation N
methods so that the system would be affected to the Q= Ei jEl E Y Sin(g--G (2)
minimum (least) extent in t:he event of an emergency.
The current in the vulnerabtle lines could be reduced i = 1, 2, ....,N
by diverting part of the cu rrent via some other dura-
ble lines. where
In certain cases, (e.g. , where there are chemical P PG. PCi and Q =QG
QCi
plants), maintaining a relLiable power supply takes
precedence over economic ope ration. This paper is con- PG. = active power generated at node i
cerned with reducing the cu.rrent in potentially dan-
gerous lines by suitably a djusting the power produc- PC. - active power consumed at node i
tions of some generators. T his would result in more
dollars per hour of operatin.g cost of the thermal pow- QGi = reactive power generated at node i
er system than at the most economic operating condi-
tion. Therefore, the probleXm is how to take defensive QCi = reactive power consumed at node i
or preventive action to main tain adequate service in-
definitely, simultaneously minimizing the increase in Ei = voltage at node i
hourly operating cost. One procedure is to start the
computations all over again and obtain the optimum op- =i phase angle of the voltage at node i, with
eration for the new constra ints. Although the excel- Q arbitrarily set equal to zero

Paper 71 TP 61-PWR, recommended and approved by the Power System Engineering Committee of the IEEE Power Engineering Society for presentation at the
IEEE Winter Power Meeting, New York, N.Y., January 31-February 5, 1971. Manuscript submitted September 18, 1970; made available for printing November 16, 1970.
2186
 Y = magnitude of the elements in nodal admit- -l
tance matrix A= gx 9u-A (9)
fij = phase angle of the elements in nodal
admittance matrix
The above may be written in terms of the sensiti-
vity matrix S as
For simplicity, the variables and parameters in- Ax = S Au (10)
volved in (1) and (2) will be classified into
three groups of vectors.11,12,13 where
-1 (11)
1) Independent or Operating Variable Vector u: This is s x
.
u
an M-dimensional vector comprised of the operating
or imposed (manipulated) variables of the system. The sensitivity matrix S is of dimension 2N x M
where M is the number of operating variables. When S
2) Dependent or Controlled Variable Vector x: Vector x is calculated in digital computer, the extreme sparsi-
is comprised of the unknown variables which must be ty of g and g would be exploited by storing and
equal to 2N in number in order for the power flow processing in a Yactored forml. This is necessary as
to be specified uniquely by (1) and (2). otherwise the operations would be prohibitively large
3) Parameter or Uncontrollable Variable Vector p: The and time consuming.
components of the vector p are the variables whose
values are specified in load flow calculation, but CALCULATION OF LINEAR RELATIONS
excluding the operating variables.
Eq. (lO)is written in matrix form as
Recall the set of 2N equations given by (1) and
(2). There are many more than 2N variables involved Ax1 s1l sl2 . . . . sz Au1
in 2N equations. Any 2N variables can be chosen as un- Au. (12)
knowns to be determined and the remaining variables Ax2
Ax s
521
s2
522 * * * 2M 2Au
must be specified. Depending on the problem under dis-
cussion, some variables could lie in one group or in
another. _Ax2NJ s2N1 s2N2- -
2NM AU
By use of the three vectors x, u, and p defined As has been discussed earlier, it is required to
above, (1) and (2) may be expressed compactly by'the control the current in certain lines (branches) of the
simple vector notation as system when the active power generated at specified
nodes are varied. The variations of active powers
g(x, u, p) - 0 (3) cause changes predominantly in the node voltage angles
where g is a 2N - dimensional column vector function. and small changes in the node voltage magnitudes. The
line current being a function of the constant line im-
SENSITIVITY RELATIONS BETWEEN x AND u pedance and the two end voltages, the changes in line
currents may be calculated up to a first order approx-
Assume that a solution x0, corresponding to a imation by the changes in'node voltage angles and mag-
specified control vector u0 and the nominal parameter nitudes. At this point it should be remembered that
vector p has been obtained: not all voltages are allowed to vary freely in magni-
tude. Usually, the voltage magnitudes are specified at
g(x0, u0, p) = 0 (4) generator busses while the load bus voltages are not
restricted, the only constraint being that they should
If the operating variable vector u experiences a lie between maximum and minimum values prescribed at
small change Au from u , the dependent variable vector each load bus.
x would change from x to x + Ax in order to satisfy
the power flow Eq. (3%. Th8n Obtaining a linear relationship between branch
g(x0 + Ax, u0 + Au, p) = 0 (5) current changes and generated power changes is accomp-
lished in two steps.
Expanding (5) by Taylor's series and neglecting Step 1. A linear relation between the changes in
the higher order terms in Ax and Au, node voltage angles and magnitudes is obtained from
the matrix given by (12). Generator bus voltages are
g (x0+Ax,uO+Au,p) = 0 = g(x,u0 p)+g 'ax + g.Au (6) held constant. The vector Ax in (12) is composed of
2N variables, but one might be interested in only K of
where g the Jacobian matrix of g with respect to the variables: In the problem under discussion, the
vector x is given by variables of interest are the changes in voltage ang-
(glp, g2$ s 2N) les of all the busses(excluding the slack bus voltage
.
g..
g (x x2 x2N) (7) angle which is kept constant at zero) and the changes
in voltage magnitudes of the load busses.
where x1, x2,. , x2N are the elements of vector x. The operating (manipul'ated) variables in Au are
the changes in real power productions at M generator
busses. Slack generator is not included in the group
SimilarlY,'Igu-3ul
a
u2 $ )..
. N (8) of M generators manipulated upon. The slack
adjusts itself to maintain conservation of
generator
power in
where ul, u2 . , uM are the elements of vector u. the network.

g is a nonsingular square matrix. gu is not Eq. (12) now becomes

generally a square matrix.

From (4) and (6), it follows that

g x Ax + g u Au= 0

2187
 AG2 AP2
sll Sl2 '* . .iM increment of the variables can be allowed in the solu-
tion of problems.
s21 s22 *
*
s2M AP3
To check the linearity between I and P, the
system shown in Fig. 1 was studied. 2 and were
AGN1 changed in steps and line (branch) currents ob2ained
AEG+l (13)
) from actual power flow solution were plotted (Fig. 2
and 3). The results reveal that the changes are reas-
AEN onably linear although a few line currents exhibit a
SK1 sK2 . .. . sKMj APM+1 slightly nonlinear change. This implies that one set
of sensitivity parameters would be sufficient if the
changes are not much.
The numbering is such that all the generator bus-
ses are numbered first (slack generator is numberead 1) FORMULATION OF LINEAR PROGRAMMING PROBLEM
and then the load busses. G denotes the total numbbr
of generators (including slack generator) and L the Recall that the problem is to adjust the M gene-
total number of load busses. The left side column rators such that the increase in hourly cost is mini-
vector of (13) is of dimensional K x 1 where K=N--1+L. mum while the predicted vulnerable system is brought
The right side column vector is M-dimensional com- into a secure state by reducing the current loading
prised of the changes in real power of the M man.ipu- in the lines which are likely to fail.
lated generators.
Let the cost function at every generating node
It is
important to note that one of the compon- be a quadratic expressed as ak k + b P2. The cost
ents of vector Ax in (12) gives the changes requ ired per hour for the entire system is given by
in P1 when P2p39 ..., PM+1 are manipulated upon.
G 2
Step 2. The changes in voltage angles and ma gni- F = k=l akpk I bkpk (18)
tudes must be related to the changes in line cur rent
magnitudes. The line current can be expressed as a For small changes in the generated powers, the
function of the line parameters and the voltage s at change in hourly cost
both ends. The magnitude of the current in ij-th line G G
is given by AF (ak+2bkPk)APk
kzlkZl = (19)
=
kpk)Apk= kl ck .AP
c.p

I =1i; (Yijt OiE' EisE i' ; (14) The coefficients ck for k = 1, 2, ..., G are evaluated
at the operating point.
Partial derivatives of Ii. with respect to Ei,
E , G, 0 can be easily calcuiated and the matrix R AP1 is involved in the objective function and in
fArmel:J the constraints. AP1 is not directly known, as only
29 AP3 * * * , AP M+
. are the independent vari-
Ai12
Ai13
r11 r12 ..'. . r1K AO2 ables. One way of determining AP is to assume that
r21 r21 .... r2K
AG3 the transmission losses remain constant so that API +
AP + ... + AP = 0 if the remaining generators and
loads consumedMare unaltered. Transmission losses do
AI23
AGN (15) change, though by a small amount. Another way of ob-
AEG1 taining AP1 is, as mentioned previously, to read it
directly from (12) as a linear function of the oper-
rating variables. Therefore, the objective function
and all the constraints can be expressed in terms of
_ _
~rBI rB2 . . . . rB AEN the operating variables AP? AP APM+l.
where B = number of branches (lines). The left side Suppose it is desired to reduce the present value
column vector denotes the changes in the magnitudes of current I d in line cd to bring a vulnerable system
of currents in all the lines. into a secure operating state. When the current in
line cd changes, currents in other lines would adjust
Combining (13) and (15), themselves in order to supply the load demand and
therefore limits should be set for each line current.
AI12
AI13
t11 t12 * *
* *
tiM AP2 Minimum is obviously zero, if no other values are
specified. Let Imax be the maximum value of permis-
t21 t22 **.* t2M AP3 ible current in ltAe ij without causing insecurity
problem. The set of inequalities may be written as
AI23 (16) Al
ij~ ij
Imax i
ij

ii ij ij
(20)
tBl tB2 ....
tBM for ij e{all lines}
where T = Bis (17) The individual generator powers are constrained by
Linearity of the Curves maximum and minimum limits.
Ii Versus Pk
AP SPmax _ p
Ai-<i
Most continuous systems can be linearized if the 1
changes in the variables are small enough. If the
system is highly nonlinear, linear relations are very
-
APi < Pi - pmin (21)
much dependent on the operating point and only a small

2188
 An.
where Pmax = maximum limit of real power generation at
bus i
P min = minimum limit of real power generation at
bus i
.64
Pi = present real power generation at bus i
APh could be positive, zero, or negative and c 134
therefore, in order to be able to use the linear prog- :
ramming technique, the variables APi should be split
up as i L.48 152
= AP+ - Al'
AP,i i PicX
+0
where APi > 0, AP>i 0 (22) 112
Combining (16) and (19) - (22) and expressing AP1 - 32 124
as a linear combination of AP2, P t,..., P + , he U u e *
linear programming problem can be written as c
Find y that minimizes AF 0 |45
J subject to
.16
Ay Sc
y 0 o (23)
where vector y is comprised of AP2,
..., M+l and matrix A and vector c
APi for i 2, 3,
are trivially de-
termined from (16), and (20) - (22). 4 .5 .6 .7
The solution of (23) yields the desired result. P2 in per unit
If the changes are too much and the solution not sat-
is factory, linearization may be done at the new op-
g 2
erating point and the process repeated until satisfac-
tory results are obtained.
NUMERICAL EXAMPLE .80_
The proposed method was applied to the five-bus
system (Fig. 1) used by Tudor and Lewisl4. All quan-
tities are expressed in per unit. Table I contains 13
the nodal admittance matrix. The cost constants are
a1 = 2.45060 b1 = 0.00410
a2 =2.49218 b2 = 0.00207
a3 = 2.11700 b = 0.00379
3
~~~~~3
2 .C 48 ,D
C

Q Generatorsn_

LiLoads

.4 .5 .6 .7
P3 in per unit
Fig. 1. Example System
Fig. 3. Line Current Variations for Change in P3.

2189
Constraints of the system are Table II-Sensitivity Between x and u
(at economic operation)
0.05 < P1 < 0.50 E1 < 1.10 Dependent Operating Variables
0.15 < P2 < 0.80 E2 < 1.15 Variables P2 P3
- 3- E3< 1.15
Loads at bus P Q
02 0.1784 0.1516
0.1784 1.2359
93
4 0.50 0.15 0.1784 0.4697
94
5 0.70 0.15 0.1784 0.3519
95
The power system was optimized for minimum cost per E4 0.0000 -0.0840
hour by using the method in Ref, (2). E5 0.0000 -0.0511
Minimum cost is 619 dollars per hour (base MVA = p1 -1.0186 -0.8657
200) and the variables are:
P1 = 0.1078 9 = - 0.0316 E = 1.10 Table III-Matrix R (at economic operation)

P2 = 0.6613 93 = 0.2187 E2 = 1.15 Varia- Variables

tion of -
P3 = 0.5576 9 = - 0.1577 E3 = 1.1436 Current 193 93 94 95 E4 E5
0 5= - 0.2224 in Line 2
E4 = 0.9711
E5 = 0.9580 1,2 -3.0350 0.0000 0.0000 0.0000 0.0000 0.0000
The line currents are:
2,4 0.8685 0.0000 -0.8685 0.0000 -1.0498 0.0000
3,4 0.0000 1.1379 -1.1379 0.0000 -0.2579 0.0000
I12 =0.2858 I45 = 0.2600
4,5 0.0000 0.0000 3.8510 -3.8510 0.9617 -0.7033
I24 0.3076 I52 = 05059 5,2 1.3759 0.0000 0.0000 -1.3759 0.0000 -1.1265
I34 =0.5164
It was assumed that the system security required Table IV-Matrix T (at economic operation)
that I34 should not exceed 0.45. The generator volt- Variation of Operating Variables
ages were held constant deliberately, as the objective Current in Line P P
was to study the increase in cost by controlling only 2 3
P2 and P3. With the notation adopted, the following 1,2 -0.5414 -0.4602
quantities are defined:
2,4 0.0000 -0.1880
2' 3' 94' 5' E49 E5 p1' Q1' Q2' Q3] 3,4 0.0000 0.8935
P= [1' E1, E2 E3, P4, P5, Q4, Q5] 4,5 0.0000 0.4088
5,2 0.0000 -0.2180
u = [P] Table V-(New) Sensitivity Between x and u
3
The sensitivity relations between x and u, and Dependent Operating Variables
matrices R and T (all quantities at the optimum) are Variables p p
2 3
shown in tables II, III, and IV. The coefficients in
(19) evaluated at the optimum are: cl = 2.45149, c = e2 0.1782 0.1681
2.49492 and c3 = 2.12123. 93 0.1782 1.2475
Table I. Nodal Admittance Matrix 94 0.1782 0.5018
A. Magnitude Y: 95 0.1782 0.3863
1 2 3 4 5 En 0.0000 -0.0383
E5 0.0000 -0.0126
1 4.6607 4.6607 0.0000 0.0000 0.0000
P1 -1.0169 -0.9592
2 4.6607 7.8649 0.0000 1.3792 1.8252
3 0.0000 0.0000 1.1999 1.1999 0.0000 Table VI-(New) Matrix R
Varia-
4 0.0000 1.3792 1.1999 6.6593 4.0807 tarion
tion of oVariables
5 0.0000 1.8252 0.0000. 4.0807 5.9056 Current
in Line
922 93 94 5 E4 E5
B. Angle c: 1,2 -2.8444 0.0000 0.0000 0.0000 0.0000 0.0000
1 2 3 4 5 2,4 1.0482 0.0000 -1.0482 0.0000 -0.8690 0.0000
1 -1.2874 1.8542 0.0000 0. 0000 0. 0000 3,4 0.0000 1.0986 -1.0986 0.0000 -0.4239 0.0000
2 1.8542 -1.2826 0.0000 1.8646 1.8671 4,5 0.0000 0.0000 3.7429-3.7429 1.4231 -1.2301
3 0.0000 0.0000 -1.3136 1.8280 0. 0000 5,2 1.4725 0.0000 0.0000-1.4725 0.0000 -0.9943
4 0.0000 1.8646 1.8280 -1.2965 1.8435
5 0.0000 1.8671 0.0000 1.8435 -1.2908
2190
 and M. Cuenod, "OptimumI control of reactive power
Table VII-(New) Matrix T flow", Ibid, vol. PAS-87, pp. 685-689, Jan. 1968.
Variation of Operating Variables 4. A. M. Sasson, "Nonlinear programming solutions for
Current in Line P3 the load flow, minimum loss, and economic dispatch-
P2
ing problems", Ibid, vol. PAS- 88, pp. 399-409,
1,2 -0.5069 -0.4782 April 1969.
2,4 0.0000 -0.3165 5. A. H. E.-Abiad and F. J. Jaimes, "A method for op-
timum scheduling of power and voltage magnitude",
3,4 0.0000 0.8355 Ibid, vol. PAS-88, pp. 413-422, April 1969.
4,5 0.0000 0.3933 6. J. C. Kaltenbach and L. P. Hajdu, "Optimal correc-
5,2 0.0000 -0.3088 tive rescheduling for power system security", pre-
sented at IEEE Summer Power Meeting and EHV Conf.,
Paper 70 TP 691-PWR, July 1970.
Maximum limits were set for other line currents: 7. J. D. Guy, "Security constrained unit commitment",
Imax = 20.3,
112 24 = 03,Imax
0.9Imax 0.35, 45 05adImax
0.5 and I52 = 0 55 Ibid, Paper 70 TP 700-PWR, July 1970.
Linear programming problem was formulated in the for- 8. A. D. Patton, "Short-term reliability calculation",
mat of (23) and using the computer library program IEEE TRANS. POWER APPARATUS AND SYSTEMS, vol. PAS-
the solution was obtained: AP 2= 0.129 and AP3 = -0.086 89, pp. 509-513, April 1970.
For the new generation schedule, a load flow was 9. W. F. Tinney and C. E. Hart, "Power flow solution
carried out and all quantities determined at this new by Newton's method", Ibid, vol. PAS-86, pp. 1449-
operating point. The line currents were: I12= 0.2774 1460, November 1967.
I24 0.3398, 134 = 0.4073, 145 = 0.2095, I52= 0.5393. 10. W. F. Tinney and J. W. Walker, "Direct solutions
The cost F is 622 dollars per hour (base MVA = 200). of network equations by optimally ordered tri-
angular factorization", Presented at the Power-
Linearizing around the new operating point, the Industry Computer Applications Conference, Pitts-
process was repeated. Sensitivity between x and u, burgh, Pennsylvania, May 15-17, 1967.
matrices R and T, evaluated at the new point are given 11. J. Peschon, D. S. Piercy, W. F. Tinney, and 0. J.
in tables V, VI and VII. New values of c are: Tveit, "Sensitivity in power systems", Ibid, Pit-
cl= 2.45135, c2=2.49545, c3 = 2.12027. Lin ar pro- tsburgh, Pennsylvania, May 15-17, 1967.
gramming was applied to get new solution. For the new
schedule of power determined, a power flow study was 12. I. Hano, Y. Tamura, S. Narita, and K. Matsumoto,
carried out. The following are the results: "Real time control of system voltage and reactive
power", IEEE TRANS. POWER APPARATUS AND SYSTEMS,
P1 = 0.1383, P2 = 0.6973, P3= 0.4826 vol. PAS-88, pp. 1544-1559, October 1969.
I12 0.3028 24= 0.3248 , 134= 0.4506 13. A. Kishore and E. F. Hill, "Static optimization
45 0.2297 I52= 0.5244
, of reactive power sources by use of sensitivity
The cost F is 620 dollars per hour (base MVA = 200).
parameters", Presented at the IEEE Summer Power
Meeting and EHV Conf., Paper 70 TP 516-PWR, July
1970.
DISCUSSION AND CONCLUSION
14. J. R. Tudor and W. A. Lewis, "Transmission losses
The set of power-flow equations are linearized and economy loading by the use of admittance con-
around the operating point and linear programming ap- stants", IEEE TRANS. POWER APPARATUS AND SYSTEMS,
proach is used to optimally reschedule the power for vol. 68, pp. 676-683, October 1963.
system security. This paper is confined to the adjus-
tment of generator powers, keeping the generator volt-
ages and the load demands constant. The generator
voltages and interruptible load demands could be eas-
ily included as operating variables by direct appli-
cation of the method presented.
Discussion
Many utility companies have recognized the ef-
fectiveness of combining the optimal scheduling of A. S. Debs (Systems Control, Inc., Palo Alto, Calif. 94306): This paper
power and system security into a single problem which presents a valid technique for generation rescheduling in the case of a
considers the interaction of the control effects of system vulnerability. The use of linear programming makes the tech-
each. In the real-time control of economy and secur- nique very appealing since in this case the very efficient Simplex Method
ity, it is necessary to obtain the solution within a is employed to compute the new schedules by minimizng the increase in
limited amount of time. The computer-aided applica- production costs. This, however, is not the first time such an approach
has been suggested. For example, Ref. 6 used linear programming for
tion described in this paper has great potential for corrective rescheduling. This latter approach is even more general than
on-line control and operation of power system econom- that proposed by the authors, since it treats, in addition, the situation
ically and reliably. where postulated line or generator outages may drive the system into a
vulnerable state.
REFERENCES Thus any contribution that can be attributed to this paper should
be related to the use of limits on line currents (rather than on power
1. J. Carpentier, "Contribution a le'tude du dispatch-. flows or voltage phase-angle differences between line ends as in Ref. 6).
ing e'conomique", BULL. SOC. FRANC. ELEC., vol. 3, From an engineering point of view, line-current limits should be im-
Ser. 8, August, 1962. posed on transmission lines,vulnerable to thermal overload and angle-
difference limits should be used where stability considerations enter. A
2. H. W. Dommel and W. F. Tinney, "Optimal power flow point of some significance is the possibility that linearizing about line
solutions", IEEE TRANS. POWER APPARATUS AND SYS- current flows may lead to faster convergence of the associated linear
TEMS, vol. PAS-87, pp. 1866-1876, October 1968.
3. J. Peschon, D.S. Piercy, W. F. Tinney, 0. J. Tveit, Manuscript received February 19, 1971.

2191
program than the case where linearization is about angular differences linearized around the operating point, and the resulting sensitivity
between line ends as in Ref. 6. This point, however, remains to be parameters are used for formulating the linear programming model. The
ascertained in future research. My a priori assessment is that very little authors are of the opinion that this approach is more accurate than that
difference in computational effort exists between the two approaches. of Ref. 6.
A questionable assumption of Ref. 6 is, in my view, the objective The approach of this paper does not suffer from generality limita-
of the minimization. When attempting to bring a vulnerable system into tions. The line or generator outage contingency could be easily included
a secure operating state, it may be a more practical objective to mini- by direct extension of the method. While the inclusion of generator
mize first the number of corrective actions (generator-schedule adjust- outage contingency is very straightforward, the line outage requires the
ments) required, rather than the cost of production. After a secure calculation of sensitivity parameters for postulated line outages. By
state has been reached, the minimization of cost becomes a problem of using the procedure described in this paper, the sensitivity parameters
economic dispatching which can be performed on a much more relaxed can be computed for the modified network resulting from line outage
time scale. Since the authors chose to use an objective function basically considerations and the necessary constraints formulated to be included
similar to that of Ref. 6, could they comment whether this approach in the linear programming model.
indeed reflects the operating philosophy of most power companies? Two ways of determining APl have been discussed in this paper
In conclusion, I feel that this paper does not contribute many and the second way was actually chosen. The necessity of using the
new ideas to the already existing references. The authors, however, second way is amply evident from the sensitivity relations between PI
have to be commended for a clear and well written paper and for their and the operating variables P2 and P3 shown in Tables II and V for the
interest in this important problem. sample problem studied. Ref. 6 suggests a relation similar to the first
way which, in the authors' opinion, is questionable.
The line current is very closely related to the line power flows in
a practical power system under normal operating conditions and, there-
fore, it is not unreasonable to base the vulnerability of the system on
A. Thanikachalam and J. R. Tudor: The authors wish to thank Dr. Debs line currents which would take care of both thermal limitation and
for his discussion of this paper. stability consideration. The authors agree with the discusser that much
Linear programming is being used increasingly for optimization of research needs to be done to ascertain whether this procedure would
many engineering problems. It is only natural that researchers are trying lead to faster convergence. This point merits thorough investigation due
to employ this technique for many power system optimization problems. to the fact that very fast computation is a fundamental and paramount
While both Ref. 6 and this paper use linear programming tech- requirement for on-line application.
nique for optimal rescheduling, there is a basic difference between Ref. The objective function was chosen to demonstrate the new ap-
6 and this paper in arriving at the linear programming model. The proach proposed in this paper; the authors are not aware of the current
mathematical model in Ref. 6 is "based on some important simplifica- operating philosophy of most electric utilities.
tions of the real-life situation, such as the use of linearized power-flow In conclusion, the authors would like to state that this paper
%snuations". In this paper, the rigorous A-C power-flow equations are presents quite a 'few new ideas for optimal rescheduling of power.
Which approach (the one in Ref. 6 or the one presented in this paper) is
more suitable for on-line application can be ascertained only by ex-
Manuscript received March 23, 1971. tensive investigation on large power systems.

2192