You are on page 1of 11

IEEE Transactions on Power Systems, Vol. 10, No.

1, February 1995 77

STATIC SECURITY I N POWER SYSTEM OPERATION WITH


FUZZY REAL LOAD CONDITIONS

K. H. Abdul-Rahman S. M. Shahidehpour
Department of Electrical and Computer Engineering
Jllinois Institute of Technology
Chicago, Illinois 60616

ABSTRACT point while satisfying active power constraints. The optimal


var problem is concerned with improving the voltage profile and
This paper presents a mathematical formulation for opti- minimizing system losses by proper adjustments of the va.r con-
mal power flow (OPF) taking into account the fuzzy modeling trols.
of static security constraints due to the uncertainty in bus loads.
Uncertainties in MW loads and generations are translated into A common trend in previous OPF techniques has been to-
possibility distribution functions. The fuzzy OPFproblem is de- wards utilizing fixed values for loads which may result in an
composed, via Dantzig-Wolfe decomposition, into subproblems overestimated solution. Some recent studies have applied fuzzy
corresponding to the possibility distributions of loads. The ef- sets to obtain more realistic models in the optimization process
fects of phase shifters are modeled as equivalent real power in- [8-161. Reference [ll]used interval arithmetics, which can be
jections at corresponding system buses, which preserves the Y- viewed as a special case of fuzzy sets, to model load uncertainty
bus symmetry and maintains minimum memory requirements. in formulating the power flow problem. The fuzzy OPF model
Contingency constraints are added to the fuzzy OPF problem. taking real load uncertainty into account was proposed in (121.
Fuzzy sets are utilized to exercise a tighter control on least cost It considered real power generation as the only viable control
real power generation with minimum emission dispatch solution. variable for alleviating line flow violations. Reference 1131 pro-
The final solution is a compromise among cost, static security posed another fuzzy OPF model and a solution strategy based
and emission considerations. Numerical results for the applica- on multi-parametric programming formulation. In our previous
tion of the proposed approach to test systems are discussed. studies [14-161, we proposed a solution to the fuzzy var opti-
mization. In this regard, a conservative range of reactive loads
would have a membership equal to 1 and additional values of
Keywords- Optimal Power Flow, Contingency Analysis, Fuzzy loads are assigned to lower memberships. The uncertainty in
Sets, Phase Shifters, Minimum Emission Dispatch reactive loads resulted in possibility distributions for voltages as
well as system losses. Voltage membership functions were intro-
duced to exercise a tighter control on variables for implementing
1. INTRODUCTION the static security.
The optimal power flow (OPF) problem plays an important This paper presents a method for considering real load un-
role in power systems planning and operation. OPF is defined as certainty in the optimal power scheduling taking into account
a nonlinear optimization problem subject to equality (power flow line flow limits and contingency constraints. Uncertainty in lo-
equations) and inequality (limits on the variables) constraints. ads is translated into trapezoidal possibility distribution. Phase
It determines the control variables such as real power genera- shifters are used for alleviating line overflows. If phase shifters
tions, generator voltages, transformers tap settings and phase control is not sufficient, then power generation will be resched-
shifters angles, for static security while optimizing a given ob- uled; however load sheddmg [21] is not considered in this study.
jective function, e.g., minimum cost of power generation. The fuzzy OPF problem is decomposed into subproblems via
The OPF solution techniques consider linear programming Dantzig-Wolfe decomposition (DWD). Since it is very difficult
(LP) as an effective and reliable tool with a fast rate of conver- to model conflicting objectives, e.g. minimum cost and mini-
gence within a reasonable computation time [l-71. Most of the mum emission, in standard optimization techniques, fuzzy sets
LP-based algorithms described in the literature take advantage are utilized to enforce minimum emission requirements in the
of the loose coupling between the magnitudes and angles of bus fuzzy OPF problem. The final solution is a compromise among
voltages [Z]. The global optimization problem can be apprmti- minimum cost, maximum static security and minimum emis-
mated by separate and successive solutions of an active power sion requirements. As an alternative, one may try to solve the
flow and a reactive power (var) flow problems. The active power problem by considering the minimum and maximum load values
flow optimization is concerned with minimizing either the gen- which result in overestimated resources and a higher operation
eration cost or the deviation of control variables from a desired cost.
In the following section, the phase shifter modelicg is intro-
94 WM 219-6 PWRS A p a p e r recommended and approved duced. The fuzzy optimal formulation is given in Section 3. The
by the IEEE Power System Engineering Committee of the LP formulation for controlling the real power generation with
I E E E Power Engineering Society f o r p r e s e n t a t i o n a t the fuzzy loads and minimum emission requirements are discussed
IEEE/PES 1994 Winter Meeting, New York, N e w York, in Section 4. Results for the application of the proposed method
January 30 - February 3 , 1994. Manuscript submitted on test systems are exhibited in Section 5, and conclusions are
August 2 , 1 9 9 3 ; made available for printing drawn in Section 6.
December 6 , 1993.

List of Symbols:
N B = number of system buses
N G = number of generators
0885-8950/95/$04.00 0 1994 IEEE
78

N V L = number of lines with overflows In the injection model, the effect of phase shifter is modeled
N P L = number of lines with phase shifters as two injections at the corresponding two buses as shown in Fig.
2. Thus, real and reactive power injections at buses t and m are
YI = admittance of line I given as follows,
br = susceptance of line I
51 = reactance of line 1 P+, = t br V, V, sin($l) cos(6, - 6
,
)
y;, = i m f helement of the Y-bus P+m = - t br V, V , sin(&) cos(6, - 6
,
)
$1 = phase shifter angle located on line I
t = control transformer tap position
Q+, = t br V , V, sin(h) sin(6, - 6
,
) + t Z bl xzsinz($r)
L?dm = t br K V, sin(&) sin(6, - 6 ,
)
I , = injected current at bus i
(3)
vi = voltage magnitude at bus i
6;= voltage phase angle at bus i
P+m = injected real power at bus m due to a phase shifter
Q+m = injected var at bus rn due to a phase shifter

Pgi= real power generation at bus i


P,:" = real power generation at bus i for minimum emission
dispatch
F"" = optimal power generation cost without phase shifters
c; = cost coefficient for real power generation at bus i
P, = total injected power at bus m
P d m = real power load at bus m
P" = power flow on line v
4 5 ) = membership function of z
Fig. 2 Injection model of a phase shifter
max = maximum value of a variable
min = minimum value of a variable
For power transmission, phase shifter angles are typically
-10' 5 $1 5 10" and voltage phase angle differences are -20' 5
2. P H A S E SHIFTER MODELING 6;- 6 j 5 20". Using this observation, the equivalent injected re-
The function of an ideal phase shifter is to adjust voltage active power at buses i and m is much smaller than the injected
and current phase angles in order to reroute the power flow, real power. Therefore, we the injected real power p+z and
P+m as approximation to the total equivalent injected power
If the magnitude of the corresponding turns ratio is not unity,
then and current will also be chanced ,221.
~~~~ ~~
due to a phase shifter. It is noted that the partial derivatives of
,--A- 0~~~
In Fig. 1,-a phase shifter isyocaked on line I between buses i (3) with respect to bus voltage magnitudes and angles at buses
and m, and the injected currents at buses and are written i and m are relatively small. Thus, the phase shifter equiva-
as. lent injected power is modeled as constant load (or generation)
connected to buses i and m.
I, = yr V , t Z - yr V, t e-,+(
(1) In the injection model, phase shifter angles will not ap-
I, = - yI V, t $61 + yI v, pear in the Y-bus, thus maintaining its symmetry and minimum
Therefore, the respective elements of the Y-bus are, memory requirements. In addition, if we assume small variations
of bus voltage magnitudes (i.e., DC power flow ) the relations
between the equivalent injected real power at buses i and m and
Y,, = t Z YI phase shifter angle are written as,
Y,, = - t e->+! YI
(2)
Y,, = - t e,+! yI

Ymm = YI (4)

It is seen from (2) that the Y-bus matrix is unsymmetrical and a


function of the phase shifter angle, which will require additional Generation rescheduling may also be used fon adjusting line
memory in OPF studies. Hence, we use a different method called flows. However, scheduled power generation is related to eco-
the injection model to treat phase shifters [17,18]. nomic requirements imposed upon power systems, and it would
be more reasonable to use phase shifters as a first alternative to
control line flows without affecting the existing economic condi-
tions of power systems.

3. P R O P O S E D F U Z Z Y FORMULATION

3.1 Fuzzy Economic Dispatch

4 4 For OPF studies, variations of real loads may be specified


within a conservative range with additional possible values of
load represented as fuzzy values [19]. In this study, we have
Fig. 1 Phase shifter model translated the real load possibilities into trapezoidal possibility
distribution by assigning a degree of membership to each possi-
ble value of the load. The possibility distribution refers to the
19

mapping of a fuzzy variable on the [OJ] interval. Figure 3 rep- where, S is the DC-based sensitivity matrix. Equation ( l l ) ,
resents the possibility distribution for real load at bus m which similar to (5-7), can be decomposed into four equations cor-
is expected to be between P f ) and P p ) , however it is more responding to four break points of the possibility distributions.
likely to be between PPI and P:". It can be argued that, if real These equations are added to (8-10)to assure that the real power
generation will not cause any line flow violations.
loads are modeled as fuzzy then the corresponding real power
generation will be fuzzy. As we use trapezoidal possibility dis- A contingency analysis is used to detect most severe out-
tributions for loads, the real power generation will also have a ages, and contingency constraints are augmented to the base
trapezoidal possibility distributions similar to Fig. 3. Therefore, case to assure a preventive control. The contingency constraints
the economic dispatch with fuzzy loads is stated as follows, are represented similar to ( 1 1 ) except the sensitivity coefficients
NG
are adjusted for the contingency under consideration [23].
min ci Pgi (5) Since we represent phase shifters in terms of equivalent in-
i=l
jected power, phase shifter angle limits are expressed in terms of
subject to: bus power injections as discussed in Section 2. Thus, if a phase
shifter is located on line l which connects buses i and rn, the
related constraint in the fuzzy case is,

4i"'" 6 51 P+m4 4;""" (12)


where, pgi and Pim represent fuzzy real power generation and It is noted that only one variable (injected power) is used for
load, respectively. A trapezoidal possibility distribution is rep- representing a phase shifter, since the injection at corresponding
resented by four break points as given in Fig. 3. Therefore, the buses are equal and in opposite directions as given by (4).
minimization problem described in (5-7) is written as follows,

3.3 T h e Overall Formulation

The overall fuzzy OPF formulation is as follows,


subject to:
NG p(F)
NG
P;:) =
NB
Pi: k = 1,..,4 (9)
min
4
c; +-
i= 1 m=1
Pmin
9:
<
- Pj,?)5 Pjf) 5 P::) 5 Pj:) 5 PZaz i = l , . . , N G (10) subject to:

Equation (8) represents the minimization of fuzzy variable pgi


as defined by (A.2) in the Appendix. The solution of (8-10) will k = 1,..,4
provide the generation possibility distributions correspondingto
fuzzy loads for the minimum cost of operation. However, this NB
solution may not be practical as it may give rise to line overflows
in the base case or in the case of an emergency.
NB

where S' is the sensitivity matrix for the worst contingency case.
The dimension of this problem is large which will be reduced via
the application of DWD [20].
Fig. 3 Possibility distribution of real power load Since we use four sets of variables each describing one break
point of the possibility distributions, DWD is applied to decom-
pose the problem into four subproblems coupled by the last two
3.2 Operation Limits as h z z y Sets
sets of constraints in (14). The dimension of the master problem
Suppose that the solution of (8-10) has resulted in line over- is equal to the number of coupling constraints plus the number
flows. In order to relieve overflows, we will adjust phase shifters, of subproblems, while each subproblem has a dimension equal
or real power generation as a more expensive alternative. Fuzzy to the number of constraints corresponding to each break point.
representation of real loads will result in fuzzy line flows with The solutibn of the master problem generates new simplex mul-
trapezoidal possibility distributions. The total bus injection is tipliers (dual solution) that will adjust the cost function of the
equal to the algebraic s u m of real power generation and load at subproblems. The solution of the subproblems with the ad-
each bus, as well as the injection due to phase shifters. There- justed objective function will provide the master problem with
fore, the fuzzy line flow on line v can be expressed in terms of new columns to enter the master basis matrix. Fig. 4 depicts
bus power injections as, the overall structure of the problem. In the following, we use
fuzzy sets inside each subproblem to exercise a tighter control on
NB the real power generation to keep it close to least cost generation
pv = Svm (pm + P+m) (11) with minimum emission dispatch solution.
m=l
80
exceeding a certain maximum tolerable cost Fws. In our study,
Master problem F"" corresponds to the optimal cost (13-14) if no phase shifters
~

are considered. Thus, the membership of the objective function

-
Objective function
-
d
in each subproblem is written as,
5 Fe'
Coupling constraints + F

- Fw')/(Fec- F"") Fe' < F < F"" (15)

F >F""

A fuzzy modeling for the real power generation at bus i is


Objective function shown in Fig. 6. It is seen that a membership equal to 1 is
-b
assigned to Pi/". Each generation is represented by two linear
Constraints related to constraints for the minimum and maximum limits. The member-
break point (1) ship function for the lower generation limit at bus i, is described
as,

Pgi 2 P;r
I'
Pgi 5 PZ'"
(16)
The upper generation limit membership function is,

{r'(Pgi- Py)/(P;/" P y ) P,'p < Pg:<


Fig. 4 Overall structure of the problem
YZ(P,i)= -
4. SOLUTION OF THE FUZZY OPF
l o
4.1 Emission Control by Fuzzy Sets
The basic function of an electric power system is to pro-
vide an adequate supply of electric energy to its customers as
Y(pgi)t
economically as possible with a reasonable level of quality and
continuity. In this regard, generation dispatch based on least
costs is conflicting with the dispatch based on minimum emis-
sion. Since it is difficult to combine conflicting objective func-
tions into one overall objective, we use fuzzy sets to deal with
conflicting objectives. Let us assume that P;,'" is the minimum
emission dispatch at bus i. Therefore, we specify our satisfac-
tion with a least cost dispatch at bus i as being closer to the
Pir. It is difficult to reflect the degree of satisfaction in the
standard LP formulation, because once the operating limits of
variables are set, all values within limits will be treated equally
in the optimization process. In the following, we will discuss the Fig. 6 Fuzzy membership function for power generation
application of fuzzy sets to direct the power generation dispatch
given in Section 3.3 towards a solution with minimum emission. 4.2 Optimal Solution with Fuzzy Variables
As we concentrate our following discussion on the formulation
of fuzzy LP in solving each subproblem, we will drop index k The decision in the fuzzy environment is defined as the in-
from the equations. tersection of fuzzy sets describing the objective and the con-
straints. If the minimum operator is used to describe the inter-
In Fig. 5 , the satisfaction with the optimal solution is mea- section in the fuzzy case, then the overall membership is,
sured as being close to the unconstrained economic dispatch (5-
7), denoted as F". A zero membership is assigned to solutions
= min [P,Y~,YZ] (18)

where X is in [OJ] and as its d u e increases, as our satisfaction


with the solution increases. The mathematical formulation for
each subproblem is given as follows,
max X (19)
subject to:

NB
Fig. 5 Fuzzy membership function for the objective P P 5 ,s, (Pm + P#m) 5
m=l
81

and Pi”. This extreme range of loads provides a wider range


of line flows than that of the proposed fuzzy model, indicating
that the fixed load interval leads to an overestimateof the system
behavior in an uncertain environment.
Table l b Generators Data (pu)
p - x z o
7 1 - X t O
n-XzO
A 5 1

The first four constraints are related to loads, line flows, con-
tingency and phase shifter requirements. The next three con-
straints represent the fuzzy objective and generation limits. The
last constraint limits X to [0,1]. So, the overall solution will
be a compromise between membership function of the objective
which requires minimum operation costs, and membership func-
tions of the generation constraints, which tend to keep power
generation close to the minimum emission dispatch point (higher
operation costs). The fuzzy LP described in (19-20) treats the
objective function as another constraint, which makes this ap-
proach suitable for problems with multiple objectives where a
combination of objectives may not be relevant.

5. NUMERICAL EXAMPLE
In this section, the application of the proposed method to
the modified IEEE 30-bus system 191 is presented. The system
has six generators, 41 lines and three phase shifters. All phase
shifters have turns ratios equal to 1. Trapezoidal possibility
distributions are used to represent the system fuzzy real power Table 2 Break points of power generation @U)
loads. The break points of the load possibility distribution are for Test case 1
given in Table la. The generators data are given in Table l b in b
No. pm Power gen. range for
which each generator cost function is approximated by piecewise
linear approximation.
z p,” pp) p;) min and max load

Table l a Possibility distributions for loads @U)


corresponding to Fig. 3

O.OO0 0.030
0.020 0.050 Test case 2
0.020 0.070
0.040 0.100
0.100 0.150
0.220 0.270 The fuzzy power generations, given in Table 2, are used to
0.020 0.030
0.060 0.080 compute the corresponding line flows possibility distributions.
0.050 0.080 0.110 0.150 The break points of line 2-6 are 0.2252,0.2808,0.4333and 0.5238
0.030 0.050 0.080 0.100 pu as compared to 0.2248 and 0.5430 pu for the fixed load inter-
0.040 0.070 0.100 0.130 val which indicates once again the overestimated results by the
fixed interval. Line 2-6 has an overflow as its flow.limit is 0.5 pu.
16 0.010 0.030 0.050 0.060
Therefore, the optimal power generation is computed again by
17 0.030 0.070 0.100 0.140 considering line 2-6 flow limit. In this case, the phase shifter on
0.000 0.020 0.040 0.070 line 4-6 alleviates the overflow without any adjustment to the
0.040 0.060 0.090 0.130 optimal power generation given in Table 2. The corresponding
20 0.000 0.010 0.020 0.040 break points for the phase shifter on line 4-6 are 0.0, 0.0, 0.0,
21 0.100 0.150 0.200 0.230 0.56”, whereas the phase shifter range for the fixed load interval
23 0.000 0.020 0.030 0.050 is between O.OO and 1.02O. So, by utilizing a possibility distri-
24 0.050 0.070 0.100 0.120 bution for loads, we will identify a smaller range for the phase
26 0.010 0.030 0.050 0.060 shifter angle. I
29 O.OO0 0.010 0.020 0.030
0.060 0.090 0.110 0.140
Test case 3
Test case 1 In this case, line flows as well as contingency constraints
are introduced in the optimization problem (13.14). The out-
In this case, no line flow constraints are introduced in the age of line 1-2, which has the largest flow compared to other
problem and the optimal power generation that corresponds to lines in the system, is considered as the worst case scenario in
the system fuzzy load is found using (8-10). The break points of our example. h z z y sets are used to discriminate between dif-
the generation possibility distributions are given in Table 2. For ferent values of power generation within its permissible range
the sake of comparison, in Table 2 we have included the power inside each subproblem to reflect emission requirements (Fig.
generation corresponding to the fixed range of load values Pi1) 6). The objective function (minimum operation cost) is treated
82
as another constraint corresponding to the fuzzy membership in 20-36, 1990
Fig. 5. The power generation break points are given in Table
3. The power generation cost in this case is $7736.20/h which is [4] J. Delson and S.M. Shahidehpour, “Linear Programming
higher than $6149.57/h for the first two cases. The power gen- Applications to Power System Economies, Planning and
eration cost for Test case 3 is a compromise between minimum Operations,” IEEE Trans. on Power Systems, Vol. 7, No.
generation cost given by the f i s t two test cases,and the cost of 3, pp. 1155-1163, Aug. 1992
minimum emission dispatch which is $8424.62/h. [5] J.,Qiu and S.M. Shahidehpour, “A New Approach for Mini-
mzing Power Losses and Improving Voltage Profile,” IEEE
The inclusionof emission control is critical in high pollution Trans. on Power Systems, Vol. 2, No. 2, pp. 287-295, May
regions such as the Los Angeles County, where the emission 1987
control is treated as a dispatch objective.
[6] N. Deeb and S.M. Shahidehpour, “Linear Reactive Power
Optimization in a Large Power Network Using The Decom-
Table 3 Break points of power generation (pu)
position Approach,” IEEE Trans. on Power Systems, Vol.
for Test case.3 5, No. 2, pp. 428-438, May 1990
(71 N. Deeb and S.M. Shahidehpour, “A Decomposition Ap-
proach for Minimizing Real Power Losses in Power Sys-
tems,” IEE Proceedings, Part C, Vol. 138, No. 1, pp. 27-38,
Jan. 1991
[8] K. Tomsovic, “A Fuzzy Linear Programming Approach to
the Reactive Paver/Voltage Control Problem,” IEEE
?tans. on Power Systems, Vol. 7, No. 1, pp. 287-293,
Feb. 1992
[9] K.H. Abdul-Fhhman and S.M. Shahidehpour, “A Fuzzy -
Based Optimal Reactive Power Control,” IEEE Trans. on
6. CONCLUSIONS Power Systems, Vol. 8, No. 2, pp. 662-670, May 1993

In majority of cases, the available data may include a great [lo] Y. Terasawa and S. Iwamoto, “Optimal Power Flow Solu-
deal of uncertainty, which may be modeled suitably via fuzzy tion Using Fuzzy Mathematical Programming,” Electrical
sets. In this paper, trapezoidal possibility distributions are used Engineering in Japan, Vol. 108, No. 3, pp.46-54, 1988
to model the uncertainty in system loads. The formulation for [ll]2. Wang and F. L. Alvarado, “Interval Arithmetic in Power
the fuzzy OPF problem is considered with real power generation Flow Analysis,” IEEE Trans. on Power Systems, Vol. 7,
and phase shifters as viable control options. The phase shifter NO. 3, pp. 1341-1349, Aug. 1992
injection model is used which preserves the Y-bus symmetry
and maintains minimum memory requirements. Line flow lim- [12] V. Miranda and J.T. Saraiva, “Fuzzy Modeling of Power
its, contingency constrains and minimum emission are wnsid- System Optimal Load Flow,” IEEE Trans. on Power Sys-
ered in the proposed formulation. It is concluded that the fuzzy tems, Vol. 7, No. 2, pp. 843-849, May 1992
approach provides a smaller range of power generation and line
flow limits than that of the fixed range of loads. The proposed [13] J.T. Saraiva, V. Miranda, L.M.V.G. Pinto, “Impact on
fuzzy approach enables power system operators and planners to Some Planning Decisions from a Fuzzy Modelling of Power
operate the system more economically for a given range of loads, Systems,” Proceedings of the IEEE 1993 PICA Conference,
while conflicting objectives such as minimum cost and minimum pp. 327-333, Scottsdale, AZ, May 1993
emission dispatch are modeled easily by using fuzzy sets. The [14] K.H. Abdul-Rahman and S.M. Shahidehpour, “Reactive
fuzzy set approach can discriminate between different values of Power Optimization Using Fuzzy Load Representation,” Pa-
variables within their operating ranges which may be difficult to per # 93 SM 502-8 PWRS, Presented at the IEEE/PES
implement in conventional optimization techniques. As power 1993 Summer Meeting, Vancouver, B.C., Canada, July 1993
systems become more complex and more difficult to operate, the
merits of fuzzy set theory and other emerging technologies must [15] K.H. Abdul-Rahman and S.M. Shahidehpour, “Optimal Re-
be explored to find more viable solutions, and assure proper active Power Dispatch With Fuzzy Variables,” Proccedings
operation of power systems to meet security, economical and of the IEEE 1993 International Symposiumon Circuits and
environmental objectives. Systems, pp. 2188-2191, Chicago, IL, May 1993

ACKNOWLEDGEMENT [16] K.H. Abdul-Rahman and S.M. Shahidehpour, “Application


of Fuzzy Sets to Optimal Reactive Power Planning with
The authors would like to appreciate the financial support Security Constraints,” Proceedings of the 1993 IEEE PICA
provided by the Power Systems Laboratory at Illinois Institute Conference, pp. 124-130, Scottsdale, AZ, May 1993
of Technology.
[17] Z.X.Han, “Phase Shifter and Power Flow Control,” IEEE
Trans. on Power Apparatus and Systems, Vol. 101, No. 10,
REFERENCES pp. 3790-3795
[l] S.N. Talukdar and F.F. Wu, “Computer-AidedDispatch for [18] M. Noroozian and G. Anderson, LLPower Flow Control By
Electric Power Systems,” Proceeding of the IEEE, Vol. 69, Use of Controllable Series Components,” Paper # 92 SM
No. 10, pp. 1212-1231, Oct. 1981 466-3 PWRD, Presented at the IEEE/PES 1992 Summer
Meeting, Seattle, WA, July 1992
[2] 0.Alsac, J . Bright, M. Prais and B. Stott, “Further Devel-
opments in LP-Based Optimal Power Flow,” IEEE ?tans. (191 D. Dubois and H. Prade, Possibility Theory : An Approach
on Power Systems, Vol. 5, No. 3, pp. 697-711, Aug. 1990 to Computerized Processing of Uncertainty, Plenum Press,
New York, 1988
[3] W.L. Snyder Jr., “Linear Programming Adapted for Opti-
mal Power Flow,” IEEE Tutorial # 90EH0328-5-PWR,pp. [ZO] G.B.Dantzig and P. Wolfe, “The Decomposition Algorithm
83
Linear Programs,” Econometrica, Vol. 29, No. 4, pp. 767-
778, Oct. 1961 21 + 52 + 23 + 54
(‘4.2)
4
S. Shah and S.M. Shahidehpour, “A Heuristic Approach to
Load Shedding Scheme,” IEEE Trans. on Power Systems,
Vol. 4, No. 4, pp. 1414-1420, Oct. 1989
Th
A S . Debs, Modern Power Systems Control and Operation,
Kluwer Academic Publishers, Boston, 1988
A.J. Wood and B.F. Wollenberg, Power Generation, Opera-
tion and Control, John Wiley & Sons, New York, 1984
A. Kaufmann and M. M. Gupta, Fuzzy Mathematical Mod-
els in Engineering and Management Science, North-Holl-
and Publishing Company, Amsterdam, 1988
Fig. A1 Possibility distribution of a fuzzy variable

APPENDIX BIOGRAPHIES
The idea of fuzzifying a variable is to replace the concept
that a variable has a precise value by the fuzzy concept indicat- K.H. Abdul-Rahman was born on June 2nd, 1964 in Jordan.
ing that a variable has a degree of membership assigned to each He received his BS and MS degrees in Electrical Engineering
possible value of the variable. In this paper the term possibility from Kuwait University in 1986 and 1990, respectively. He is
distribution refers to the mapping of a set X to [0,1]. We will currently a Ph.D. student in the Electrical and Computer En-
represent the possibility distribution by its break-point values gineering Department at Illinois Institute of Technology. His
(i.e.,xl, x2, x3, x4), as shown in Fig. A. The minimization of a research interests include optimization and control of power sys-
fuzzy variable X , given in Fig. A, is translated into the min- tems.
imization of its distance from the p,y axis. It is shown in [24]
that the distance is given as, S.M. Shahidehpour is the Associate Dean of Engineering for
Research and Graduate Studies and Professor of Electrical and
Computer Engineering at Illinois Institute of Technology. He
was the recipient of 1993 Edison Electric Institute’s Power Engi-
neering Educator Award. Dr. Shahidehpour was Vice Chairman
of the 1992 IEEE Systems, Man,and Cybernetics Conference,
where B1 and Bz are areas given in Fig. A. For the fuzzy variable and Associate Director (1985-1993) of the American Power Con-
X in Fig. A, the distance given by ( A . l ) is computed as, ference.
84

Discussion W e feel that, it may be better to adopt a membership


function similar to the one described in (19,i.e.,
Xiaobong Guan (Consultant), W.-E. Edwin Liu and Alex D.
Papalexopoulos (Pacific Gas and Electric Company) The authors
have presented an interesting paper on OPF applications with
uncertain constraints, and should be commended for applying the
fuzzy optimization method to this area We agree that this method
may have some potential in modeling the uncertainties encountered
in power systems We would appreciate the authors' comments on
the following issues
where E, is the actual emission, & the best emission
level, and,E the maximum tolerable emission. I& and
1 In the paper, the membership function for real loads is equal to
,E are much easier to determine than Pa,- and
1 between PA2)and P j 3 ) as shown in Fig 3 However, it is
reasonable to assume that the OPF solution would gravitate Pa,-.
2. As to the results presented in Tables 2 and 3, break
toward satisfjmg Pjz),since the lower the load, the lower the point values are obtained in terms of fuzzy real loads by
total cost will be In other words, the OPF solution would be at the proposed method. In power system operations, we
the lower load if the load variable can be shedded without are more accustomed to crisp values. Would the authors
penalty Therefore, a triangle function, instead of a trapezoidal explain which of the four break point values should be
function, may be more appropriate as a membership function for applied for control purposes, such as preventive
real loads controls. in the fuzzy environment?
Again, the authors are wm mded for an excellent paper.
2. Based on our understanding of fuzzy decision problems, the
decision variables are generally defuzzified. When a decision is Manuscript received March , 1994.
made, the actions should be crisp although the environment may
be fuzzy. For the OPF problem described in the paper, the
decision variables are the outputs of generators and phase
shifters. In practice, these variables are usually measured and
controlled in a power system. Could the authors explain why
these variables are considered fuzzy in the paper? In a S i a r
(S.K. Joshi, A.Chandramouli, S.N.Singh and K.N. Srivastava, Indian
development, the discussers have decided to use this method to Institute ofTechnology,Kanpur. INDIA)The Authors are commendesd
model only the limits of transmission lines. We will eventually for the interesting work on Fuzzy based static security operation of
expand the fuzzy formulation to include other uncertain power systems. We would appreciate the authols' comments on the
constraints. However, we felt that decision variables such as following:
generator outputs should not be fuzzified. 1) It mms the authors have not considered the transmission losses
in the proposed formulation. If the losses are included, how would be
the possibility distribution of losses related to the possibility
3. In a practical OPF problem, the cost hnction and constraints distributions of real power generations and loads?
are nonlinear. The problem considered in the paper is an
approximation with a linear cost function and linear constraints. 2) The possibility distribution maps the specific values of real power
As presented in the paper, the method has been tested on the generation ne well ne load to the membership values on the [0,1]
IEEE 30 bus system. What problems do the authors foresee in interval. Where are these membership values used in the formulation?
Why the authors considered trapezoidal distribution?This distribution
implementing a fuzzy OPF formulation on practical power seems very simplistic for load uncertainties. Were any other
systems? distributions considered?

Manuscript received February 25. 1994 3) The power balance constraint equation (9) needs more
substantiation. a) How can one say that the sum of generations for
each break point should be equal to the sum of the loads of the
corresponding break point? b) Is it not correct, as in equation (6), that
the power balance should be satisfied over all the break points rather
than individual break points? c) In view of this constraint (9), how can
the individual possibility distributions be determined for generator
A.A. El-Keib and Xingwang Ma (Department of Electrical buses? d) If losses were considered, how would they be incorporated?
Engineering, The University of Alabama, Tuscaloosa,
Alabama): W e congratulate the authors for an interesting 4) In the case of numerical example . the break points of loads and
generators do not seem to satisfy the above power balance constraint
paper on static security in power system operation with (9).Why is it so?
tiuzy real load conditions.W e would appreciate the authors'
comments on the following: 5) The authors have given emphasis on phase shifters, however the use
1. Equations (16) and (17) are used in consideration of the of variable series capacitors (FACTS) will be reasonable where no
emission constmn' t."bismi&esthedetemmn ' tionofthe approximation is required to maintain symmetry of Y-bus matrix.
minimum and maximum emission-constrained
6) What is the time required for this analysis in comparison with any
generations of the units, pa* and P#,-, important conventional approach?
but somewhat elusive. In addition, these values can also
vary with changes in loads and system configuration. Manuscript received April 4, 1994.
85

Vladimim Miranda, J. T. Saraiva (INESC, Porto, Portu- very easy to prove and it happens whenever limits in line
gal): This paper is interesting in several ways, and we wish flows or in generations are reached.
to congratulate the authors for submitting it to the judg- Besides, the mathematical model described in equa-
ment of the scientific community. This gives the opportu- tions (8) to (10) requires further research effort in order
nity to discuss some basic assumptions behind the adop- to be validated as an acceptable approach to what may
tion of fuzzy set techniques in Power System analysis, and happen in a real power system. Namely, it is not difficult
in this way clarify some matters still in the shade. to build toy examples where the model leads to infeasibil-
This discussion will be oriented in three directions: ity, although obvious solutions may be detected by inspec-
establishing the meaning of the definition of uncertain tion.
loads, in the fuzzy sense, challenging some assumptions The model actually leads to results whose consequences
accepted in the paper as to the fuzzy behavior of the the authors have correctly identified in their paper: it may
variables, and questioning the mathematical model de- produce narrower possibility ranges than the ones ob-
rived. tained with only studies for the min and max values of the
If a load is defined by a fuzzy number, such a trape- loads. However, this is not an advantage of any fuzzy
zoidal number like in Fig. 3 of the paper, this means that model: it derives from the approximations inherent to the
every value between Pj’) and Pj4)is a possible value, i.e., adoption of equations (8) to (10). Furthermore, previously
a load value that may occur according to the expectation published examples [Al, A21 have shown that one of the
of the planner that defined that fuzzy load. The interval merits of a fuzzy load flow study is precisely to detect
[PA’), PY)] represents the largest uncertainty range admit- larger possible ranges of values than the ones obtained by
ted (to which a membership value of 0 is associated); the performing only studies for the min and max values of the
interval [ PJ*), represents the narrowest uncertainty loads.
range to which the planner is able to reduce the impreci- Considerations similar to these have led us and other
sion in his definition of the load value (and therefore colleagues to abandon the model proposed in [A31 and
receives a membership or possibility value of 1). replace it by the model described in [A4]. This latter
The fuzzy number may be therefore interpreted as a set seems to be the only “correct” approach published, so far.
of continuously nested intervals, defining larger and larger But we would be very interested in cooperating with the
ranges of uncertainty, with decreasing membership or authors in a research for the domains of applicability of
possibility degree. At any level a,the associated interval the approach implicit in equations (8) to (101, as it might
representation of the load means that any value within also lead to some rehabilitation of the model presented in
that interval must be considered possible to happen (with [A31.
possibility a).This is why some arithmetic operations with In any case, the paper here discussed introduces a
fuzzy sets may be viewed as an extension of interval modeling of phase shifters which deserves attention and
arithmetics. may possibly be added as an extension of the formulation
When one defines several fuzzy loads in a Power Sys- of the Fuzzy Optimal Power Flow in [A4].
tem, this must have the interpretation that, at each and We hope the authors may find an adequate frame and
every level a, all load scenarios, resulting from the com- set of assumptions (of course, more restricted those indi-
bination of all values within the uncertainty ranges de- cated in their paper) for which their model may retain
fined at level a,are possible and valid (this set of scenar- some validity and also that the authors may accept our
ios results from the Cartesian product of the intervals at friendly challenge for establishing some degree of cooper-
level a defined for each load). Therefore, when defining a ation in this subject. In any case, the authors must be
dispatch policy in a fuzzy load environment, one must commended for their attempt in trying to include (with
accept that fuzzy generations will come as a result; fur- the help of the fuzzy set theory) qualitative aspects in
thermore, these fuzzy generations must be such that, for numerical models of power systems.
every possible load scenario, at level a,there must exist a
dispatch scenario (defined at the same level of possibility References
a) with all constraint satisfied and with the dispatch
criterion met. In other words, we cannot have a possible [All V. Miranda, M. A. Matos, J. T. Saraiva, “Fuzzy
load scenario without having also a possible dispatch sce- Load Flow-New Algorithms Incorporating Uncer-
nario, and at least at the same degree of possibility. tain Generation and Load Representation,” 10th
The authors have repeatedly stated in the paper that PSCC, Graz, August 1990; in Proc. of the 10th
defining trapezoidal fuzzy loads would imply having trape- PSCC, Buttenvorths, London, 1990.
zoidal generations and trapezoidal line flows. This may be [A21 J. T. Saraiva, V. Miranda, M. A. Matos, “Genera-
the case in some particular situations or systems, but it is tion and Load Uncertainties Incorporated in Load
not possible to generalize this idea without specifying Flow Studies,” Proc. of MELECON91 (IEEE s.8),
clearly under which conditions this assumption holds. In Ljubljana, Youg., May, 1991.
fact, under only very general assumptions, even if the [A31 V. Miranda, J. T. Saraiva, “Fuzzy Modeling of
fuzzy loads are defined as trapezoidal, neither the genera- Power System Optimal Load Flow,” ZEEE Transac-
tions nor the line flows are necessarily trapezoidal-this is tions on Power Systems, Vol. 7, No. 2, May 1992.
86

[A41 J. T. Saraiva, V. Miranda, L. M. V. G. Pinto, function of power generation and may not be easily mod-
"Impact on Some Planning Decisions From a Fuzzy eled in conjunction with a piece-wise linear formulation of
Modeling of Power Systems," Proc. of 11th PICA, the heat rate curve of generators.
Scottsdale, Arizona, May 1993; to be published in 2. In the operational planning of power systems, we are inter-
IEEE Trans. on Power Systems. ested in being able to adjust control variables within a given
range of loads. In OPF calculations, if control variables are
Manuscript received May '17, 1994. not widely adjustable within the given range of loads, then
we have to consider other means of control for maintaining
static security. So the critical issue is to determine whether
the given range of controls is wide enough to accommodate
K.H. A b d u l - R a h m a n and S.M. Shahidehpour- We thank all possible values of load. The wider the range, the more
the respective discussers for their interest in our paper and their expensive the cost of power operation will be. The fuzzy
valuable comments. In the following, we respond to their com- representation of loads will result in a conservative range
ments. for variables with higher memberships and a wider range
with smaller memberships by minimizing the risk in the de-
cision making process. The exact value of control variables
will then be determined within the given range for a given
Miranda and Saraiva
value of the load. Our earlier study was more specifically
related to determining the crisp value of control variables
In maintaining the economic dispatch, we assume phase for a given load value in a fuzzy environment [A].
shifters are utilized for alleviating overflows in the case of
contingencies. If phase shifters do not provide an adequate
control, generators will be adjusted as well which will af-
Guan, Liu and Papalesopoulos
fect the economic dispatch, and other alternatives such as
load shedding will be considered if previous preventive con-
trol actions fail to provide proper margins for static secu- 1. The proposed OPF solution for determining the range of
rity. The key assumption in our study is that phase shifters control variables uses four independent subproblems for break
would be capable of providing the adequate margin within points via the Dantzig-Wolfe decomposition. Since we are
the given range of loads, and since line flows are specified as not interested in a single OPF solution, the range that we
a linear function of loads, line flows will increase as we in- use for representing loads can be either trapezoidal or tri-
crease loads. Hence, as we use trapezoidal possibility distri- angular; however a triangular membership function is will
bution for loads, the real power generation will also be rep- be a special form of the trapezoidal function.
resented with a trapezoidal distribution function. In case
phase shifters cannot provide the adequate margin, then we 2. Once we consider a fuzzy representation of loads, then all
have t o consider more expensive options such as the redis- system variables will have a fuzzy representation. As ex-
patch of power generation as a preventive control means plained earlier, we consider a fuzzy (non-crisp) range of
[Al-A4], which may violate our assumption of trapezoidal values for loads and determine the corresponding range of
membership function for line flows, power generations and possibilities for control variables. However, as always for ev-
other variables. ery measured (crisp) value of load within the given range,
there will be a crisp set of control variables representing the
As stated in the paper, the adoption of a fuzzy model for OPF solution.
representing load possibilities in an uncertain environment
will lead to a fuzzy representation for the range control 3. As we increase the size of the power system, the number
variables which will help power system planners minimize of variables will naturally increase. Since we adopt the
the risk of overestimating resources for the operation of a Dantzig-Wolfe decomposition in our solution, within each
large power system. It would be rather inconceivable that subproblem we will be dealing with a traditional approach
the outcome of fuzzy OPF for minimizing the possible range to OPF. Except for the additional computational require-
of control variables would result in a larger range than that ments in the iterative process, there were no additional dif-
of the crisp model. ficulties in applying the proposed method.

El - Keib and M a Joshi Chandramouli Singh and Srivastava

1. In our approach, the bus power generation P '" is deter- 1. Since we were primarily concerned with the control aspect
mined by the minimum emission dispatch reqEirement; in of power systems, we did not incorporate MW losses in
this regard, the heat rate curve of individual generators is our formulation. hrthermore, in a decoupled formulation,
replaced by the corresponding emission curve of generating losses are generally given as a function of bus voltage mag-
units. Most studies consider emission as a quadratic func- nitudes and are minimized by controlling the reactive power
tion of power generation (i.e., similar to the heat rate curve flow [1416].
with different coefticients),where PF'" and P r =in Fig. 6
are the unit i generation limits. 2. As line flows and power generation are linearly related to
loads, we consider four break points of the trapezoidal mem-
Considering the respective discussers' equation for describ- bership function to represent the possibilities of loads and
ing emission constraints is an interesting suggestion which the corresponding control variables. The membership of
may be incorporated in our model knowing that the emis- the solution is optimized by (18-20). Moreover, other mem-
sion is a function of power generation. The difficulty with bership functions may be used (e.g., triangular membership
incorporating the suggested model is that E, is a quadratic function), while a triangular function remains to be a spe-
87

cial case of the trapezoidal function. Nonlinear membership 6. We did not compare fuzzy set results with those of con-
functions may be used alternatively but they will render the ventional (e.g., deterministic, probabilistic) approaches, be-
final formulation to be very complicated and will no longer cause the underlying assumptions are very different. Deter-
be solved via linear programming. ministic approaches deal with situations were uncertainty
is not associated with system variables, while probabilistic
3. Each break point of the load membership function serves approaches assume uncertainties represented by probability
as an operating point which must satisfy the load balance density functions and complex convolutions are likely to be
equation. It is stated in the paper that as long as the load considered. Therefore, a direct comparison is not fair in
has a trapezoidal distribution function, the representation this situation. However, we believe that many of the exist-
of all variables will also have a trapezoidal distribution; this ing methods in power systems can be enhanced with fuzzy
assumption is valid while the corresponding sensitivity coef- set methods, and shortcomings in modeling and analysis of
ficients of matrix S are positive and we are able to eliminate large scale power systems can be addresses by fuzzy sets.
line overloads by adjusting phase shifters. Accordingly, if
we know the break points for variables we will be able to
determine the possibility distribution for fuzzy variables.
4. The load possibility distribution are given in Table l a for
[A] K.H. Abdul Rahman, S.M. Shahidehpour and M. Danesh-
dmst, ‘AI Approach to Optimal Var Control with h z y Reac-
load buses with fuzzy representations. The assumption is tive Loads,’presented at the 1994 IEEE PES Winter Meeting,
that the local loads at generator buses (see the standard New York, NY.
IEEE 30-bus system data in [B]) remain as crisp values and
therefore are not listed in Table la.
[B] 0. Alsac and B. Stott, ‘Optimal Load Flow with Steady-
5. As discussers pointed out, the use of FACTS will provide State Security,’ IEEE Trans. on PAS, Vol. 93, No. 3, pp.
a versatile control over real power flow in power systems. 745-751, May 1974.
However at this time the use of FACTS in power systems
is not extensive which may be due to high implementation Manuscript received June 7, 1994.
costs. In many utilities, phase shifters are still being con-
sidered a viable option for real power flow control and OPF
studies.

You might also like