IEEE Transactions on Power Systems, Vol. 3, No.

1, February 1988 317

MULTIOBJECTIVE OPT I MAL GENERAT I ON DI PATCH BASED ON PROBAB I L I TY SECURITY CR ITER I A

R. YOKOYAMA S.H. BAE T. MORITA H. SASAKI

TOKYO METROPOLITAN UNIVERSITY HIROSHIMA UNIVERSITY
FUKAZAWA. SETAGAYA-KU H IGASH I -HIROSHIMA
TOKYO. 158. JAPAN 724, JAPAN

ABSTRACT

This paper proposes a novel and efficient algo- reliability, security and minimal impact on environ-
rithm to obtain the optimal power flow in power system ments are typical objectives to be satisfied. It may
operation and planning phases by solving a multi- be obvious that trade-offs among these objectives are
objective optimization problem. In deciding the opti- impossible because of their different nature: This is
mal system operation, various objectives, such as stated that objectives are "non-commensurable." In
economy, reliability and minimum influence on environ- other words, it is impossible to treat respective
ments, should simultaneously be attained. However, objectives under the identical criterion. Furthermore,
these objectives are contradictory to each other and there occurs mutual interference among objectives, and
are in trade-off relations, thus making it difficult conditions of attaining a particular objective may
to handle this class of problem by conventional ap- differ depending on system configurations and system
proaches which optimizes a single performance index. states.
In the proposed algorithm, the optimal load flow
problem is first formulated as a multi-objective opti- Unfortunately, conventional optimization tech-
mization problem. €-constrained technique is used to niques are not suitable to obtain the optimal solution
obtain a set of non-inferiority solutions. Then, we which simultaneously optimize a variety of objectives.
introduce the idea of a preference index to decide the One conceivable approach using conventional methods is
optimum solution. In this particular study, security to convert a multi-objective problem into a single ob-
index is chosen as the preference index. jective problem by assigning distinct weights to each
The proposed algorithm has made it possible to objectives, thereby allowing for relative importance
treat the optimal dispatch problems with multiple among goals [ 3 ] . However, this artifice is not totally
performance indices and to grasp trade-off relations satisfactory since different objectives cannot be
between selected indices. The effect of uncertain evaluated under a common measure and there are no
factors pertaining to power systems can also be taken rational basis of determining adequate weights.
into account.
The validity and effectiveness of the proposed The main purpose of the optimal power dispatch
approach is verified through numerical examples on the problems have s o far been confined to minimize the
IEEE 30 node, 6 generator system. total generation cost of a power system. However, in
order to meet environmental regulations enforced in
INTRODUCTION recent years, emission control has become one of im-
portant operational objectives. In this problem, the
In recent years, more stringent requirements have amount of NOx emission, which is in proportion to the
been imposed on electric utilities. This tendency has active power output of a generator, is selected as an
brought about sheer necessity of attaining system evaluation criterion, and the minimum emission is
planning as well as system operation of higher level sought within a small region around an economically
and of greater sophistication. Personnels in charge of feasible operating point [ 4 , 5 1 .
system operations are requested to determine the opti-
mal system state with satisfying many kinds of opera- System security is another essential factor in
tional constraints. A s a means of solving this class power system operation and also in system planning. To
of problems, extensive studies on the "optimal power be specific, it is very important to maintain good
flow" (OPF) have been undertaken [ 1 , 2 ] . Recent studies voltage profiles and to limit line flows within pre-
associated with the OPF have been centered on making scribed upper bounds. In security analysis, a series
it more efficient in algorithm and applicable to on- of anticipated contingencies are assumed to predict
line use. It should be noted, however, that the OPF is possible overloadings or excessive voltage deviations.
to optimize just one specific objective, o r single Then, a security index as a function of overloads
performance index. and/or voltage excursions will be minimized by some
preventive control actions.
In general, a large scale system, typified by an
electric power system, possesses multiple objectives When permissible limits of emission and overloads
to be achieved. In a power system, economic operation, are clearly specified in a power system under study,
these quantities could be incorporated into the OPF as
operational constraints. However, in system planning
studies, these limits posed on emission o r overloads
would be very ambiguous, thus making such treatment
difficult. Also, in actual system operations, it is
necessary to maintain the system at a proper security
and emission levels even when generator or transmis-
sion line trippings do occur. To attain this goal,
system operating points should not be at constraint
limits but needs some operational margin. Furthermore,
This paper was sponsored by the IEEE Power Engineering Society operation indices mentioned herein are in conflicting
for presentation at the IEEE Power Industry Computer Applica- trade-off relations, successful optimization cannot be
tion Conference, Montreal, Canada, May 18-21, 1987. Manuscript attained through any of conventional optimization
was published in the 1987 PICA Conference Record. approaches.
0885-8950/88/0200-03 17$01.000 1988 IEEE
318

Here, a brief review will be given on the follow- Original problem:

ing three alternative approaches to deal with optimi-
zation problems having multiple objectives that are in .-,f.(z)Ir
min(fi(z),f?(z),
trade-off relations and non-commensurable:
subj. to ~ € 1 ,
g,(z)<O
(a) An approach in which only a specific objective to k= 1, 2, ..., m
be regarded as the most essential is taken as the where, 2 : decision variable vector
performance index, while the others are treated X : permissible set of 2
as constraints. fi(Z) : i-th objective function
(b) An approach in which a scalar composite perfor- gk(Z) : k-th constraint equation
mance index is made up by properly weighting each
objectives [ 5 ] . It is assumed that the multiple objectives given above
(c) An approach in which the most essential objective are in the relation of trade-off, and also is assumed
is taken as the primal performance index and the they are non-commensurable to each other. It should be
remaining objectives are processed in each sub- noted here that both the objective functions and the
problems as respective performance indices. constraints above are, in general, non-linear func-
tions of Z.
Approach (a) works well only when the goal to be
attained is well defined and priority order among ob- For algorithmic convenience of obtaining the non-
jectives is given a pri-o-ri. This treatment, however, inferiority solution set, it is a wise expedient to
encounters a severe difficulty in case trade-off rela- convert original problem (1) into a single-objective
tions among contradictory objectives, such as between optimization problem. A typical and yet very powerful
reliability and economy, are vaguely given. As to method for this is " E -constrained technique". In this
approach (b), it is quite easy to optimize a scalar method, one specific objective function, which is
performance index composed as a linear combination of arbitrarily chosen or preferably corresponding to the
involved objectives. However, it is of serious doubt most important objective, is taken as the scalar per-
to assign weights in homogeneous manner to objectives formance index to be minimized. Assuming the n-th
being not commensurable. Further, the meaning of objective function be selected, the E-constrained
weighting factors is difficult to justify. On the approach may be expressed as
other hand, the main advantage of approach (c) is in
that it can clearly handle trade-off relations among minf. (2)
conflicting objectives. Nevertheless, there is no subj. t o f , ( z ) < s , ( i = l , 2 , ..., n-1)
definite guide to select which one should be selected
as the most desirable from a group of optimal solution XEX
candidates. g4z) 1 0 (k= 1, 2, ...)m)
In power flow optimization problems, there exist where, E is a parameter vector representing the maxi-
a number of objectives to be achieved which inherently mum permissible levels for the remaining n-1 objective
have different characteristics, and hence conflicting functions. The non-inferiority solution set can be
relations hold among these objectives. To be specific, obtained by repeatedly solving eluation ( 2 ) with
indices associated with economy, reliability and envi- E perturbed.
ronment protection are non-commensurable in their
nature. Moreover, there is no invariant priority order Computation Algorithm of €-constrained Technique
among these indices considering drastic changes in
circumstances surrounding electric utilities. In these The efficient procedure of the E-constrained
cases, some personnel must decide which is the optimal technique to obtain the non-inferiority solution set
based on his subjective judgements. Therefore, it is consists of 3 steps described below [ 7 ] .
indispensable to grasp quantitatively these trade-off
relations in order to obtain the optimal solution in Step 1 The original problem with n objectives defined
objective way. in equation (1) is divided into n independent sub-
problems, each of which has respective evaluation
The present study has adopted the tnird approach index, fi(X). Let denote by :E the optimal value of
stated above. An original problem with multiple objec- the i-th subproblem. That is,
tives is formulated as a multi-objective optimization
problem, and then the e-constrained technique is rninfi(z)=&,O ( i = l , 2, ...,n) (3)
utilized to obtain a set of non-inferiority solutions
( o r , Parate-optimal solutions). Here, the concept of
the non-inferiority implies that when an arbitrarily
chosen index is to be improved, some other indices Step 2 Select one objective function arbitrarily, say
deteriorate more than the improvement gained in the fn(Z), out of the n-functions, and designate this as
selected index. Thus, the originally taken performance the main objective. Remaining n-1 objective functions
indices give no clue to choose the optimal solution. fl(2), f2(Z), ...,
fn-l(z) will be treated as inequal-
A specific criterion called a preference index gives a
ity constraints. Then, we can formulate the following
scalar optimization problem:
sound basis for selecting the optimal solution.
In order to demonstrate the effectiveness of the
proposed approach, the IEEE 30 node, 6 generator sys-
tem is taken as the example. Objectives selected are (4)
economy, miminal environmental impact and security.
The selection of the security index as the preference
index has proved successful strategy.
-
where Ei is defined by e
: and a positive value Ei as
FORllULATION OF MULTI-OBJECTIVE OPTIMIZATION PROBLEM the following:

On the basis of the above discussions, we may

formulate a multi-objective optimization problem as a
vector optimization problem:
 319

The solution of single-objective optimization problem FORMULATION OF PLANNING AND OPERATION PROBLFMS WITH
( 4 ) is denoted as NON-COMMENSURABLE PERFORMANCE INDEX

In this section, we shall formulate the optimiza-

e= [&l, 62, ".? En-1lT I tion problems in power system planning and operation
studies which have multiple non-commensurable objec-
tives. In the below, the performance indices together
Step 3 Let us assume E in e uation ( 6 ) as an initial with the equality and inequality constraints pertain-
value and let p(B'o')=p('' .
Repeat the following ing to the power system optimization problems will
substeps for i = 1 to n-1 in order to obtain a non- be described. Symbols appearing in the following
inferiority solution set which forms a hyperplane of discussion are listed in Table 1.
n dimension.
(1) Set p = 1, where p is en iteration count. Evaluation Index
(2) Let ~i' = E--': A E ~: A E >~ 0, and solve
equation (:). If there exists a feasible In optimizing power flows, the 3 most important
solution cp , then go to next step. If there evaluation indices are taken into account here, i.e.,
are no feasible solution any more, terminate economy, environment impact, and security.
the substeps.
(3) If pp < (pP-l , preserve (pp instead of (pp-'. If (a) Economy index: The fuel cost of a thermal unit can
not, discard the new solution. be regarded as an essential criterion for economic
( 4 ) Increase iteration count p by 1, and go back feasibility. The fuel cost curve is assumed to be
to substep ( 2 ) . approximated by a quadratic function of generator
active power output as
It may be appropriate to add a few comments here on
step 3 . Substep (2) is to improve i-th constraint (7)
fi(X) by decreasing the maximum permissible limit Ei
by a small decrement so long as reduced problem ( 4 )
has a feasible solution for a given Ei. Due to the (b) Environmental impact index: Nitrogen-Oxide (NOx)
nonlinearity of the evaluation index, monotonical emission is taken as the index from the viewpoint of
decrease in V P with Ei is never guaranteed. of environment conservation. The amount of NOx emis-
sion is given as a function of generator output, that
The general flow diagram of the E-constrained is, the sum of a quadratic and exponential functions
technique is depicted in Fig. 1. [4,81:
N
Fe= C (6!i-bbiPGi+CiPCs2
i=l (8)
+U!, exp(e,Pci))
(c) Line overload index: Overloading in a transmission
line can lead to system collapse in an extreme case.
Hence, we adopt, as the security index, a weighted sum
of line flow deviations of all transmission lines. The
line flow deviation of a transmission line implies the

I
amount which exceeds its reference transmission capa-
city. Thus, security index is defined as
Selection of main objective from multiple, Setting
of initial parameter,Width of parameter change
E1 , & 2 - - - - &,-I (9)
where,

I
Optimization by non- linear optimization
approach (quasi-Newton approach) I
*
c onver~ e n c ?e Table 1. List of symbols
G Yes
inferiority solution ? PG; : Active output of generator i
QGi : Reactive output of generator i
J. Yes P, : Active power at the bus i
Parameter renewal 01 : Reactive pawer at the bus i
I PI' : Active p e r specified value at the bus i
Qi' : Reactive p e r specified value at the bus i

ave calculations been made f o r all parameter lvil : Voltage magnitude of bus i

1
.
.I.Nn
.-
1
lvals:Power specified value of bus i
: Active p e r flaw of line i
PI;' : Active pawer flow standard value of line i
N : N d e r of generators
L : Number of lines
wi : Weighting coefficients
PL : Total transmission loss
P D : Total load capacity
[ Computation end ) c , f : Real and imaginary part of h s voltage vector
a.8.7 : Coefficients of generator cost characteristic
a,b.c,d,e : Coefficients of generator's NOx emission
characteristic
Fig 1. Flaw diagram of the E-constrainedmthod
320

Equality and Inequality Constraints functions themselves. One conceivable resolution for
this ambiguity may be to entrust a system operator
The static optimization of a power system can be with the final decision; He will choose the optimal
formulated as a mathematical programming problem which solution out of the candidate solutions based on his
minimizes a given evaluation index under equality persocal experience and/or system operation policy.
constraints given below: However, this way of decision can be quite subjective.
At a load bus, In order to select the optimal solution on a more
rational basis, the authors propose to make use of
Pi -Pi'= 0 some "preference index", other than the originally
Qi-Q,'=O involved objectives. In this particular study, a pre-
ference index reflecting static system security is
introduced which takes into account uncertainties
At a generator bus, intrinsic to power systems. Possible uncertain fac-
tors to be considered are (a) load changes, (b) gene-
rator outages and (c) transmission line trippings;
Only generator outages will be considered in this
study.
To insure active power balance,
N The preference index proposed here is defined as
C PGi-(PD+PL)=O N L
i=l
Isec= i = l Ri k = l J?k(Pi)

Here, transmission l o s s PL is expressed as a nonlinear L(Pi)=f(P,h - Pi a )

function of bus voltages to allow for its fluctuation: (19)
{f(xt)=xilxi€$}
~i=(l-~l)(l-~~)...~i...(l-~")
N
It should be noted that the active and reactive power Pik= Pia'+ CSa,d p ,
i=l
outputs of a generator are expressed as nonlinear
functions of bus voltages; Equality constraints (10) where, pi : output of generator i
to (13) can succinctly be expressed in a vector form ri : forced outage rates of generator i
as h(x) = 0 . pia : permissible value of transmission line k
piat : power flow of transmission line k before
In the next, inequality constraints must be im- an outage
posed on generator outputs and bus voltage magnitudes: Ski : power flow sensitivity relating to gene-
rator i to transmission line k
4 : a set of transmission lines whose power
flows exceed their permissible values

In the above, Ri represents the probability of multi-

ple generator outages. However, as this probability
is extremely low in reality, only a single unit outage
Aggregating equations (7) to (17), we may obtain will be considered here. It is assumed further that
the following vector optimization problem in terms of generation rescheduling to compensate the output of an
bus voltage vector 2 : outaged generator is made in proportion to the rated
capacities of generators (distribution factor). Hence,
incremental output of each generator is allocated
according to the following:

where flmf3 correspond to equations (7)-(9), respec-

tively. where, C, : rated capacity of generator i
Pcr : output of outaged generator k
Thus, the static optimization problem of a power
system results in the multi-objective planning problem The above index has the following characteristics.
in which mutually interfering and non-commensurable
indices (7) to (9) will be optimized under equality (1) It is possible to take into account of the effect
constraints (10) t o (13) and inequality constraints of not only the topological structure (node admittance
(14) to (17). matrix) of a power system but also differences in load
levels and voltage profiles.
As is explained earlier, multi-objective optimi-
zation problem (18) can be solved by the E-constrained (2) The introduction of distribution factor (20) has
technique. As this technique gives rise to a set of made possible to redispatch every possible generators
non-inferiority solutions, another criterion is needed instead of absorbing generation deficiency solely by
to select the optimal solution. the slack generator.

(3) The index may seem similar to the line flow index
PREFERENCE INDEX TAKING INTO ACCOUNT UNCERTAINTIES due to that both indices evaluate line flows, but does
clearly differ in that it can take into considerations
The E-constrained technique yields a set of non- outage rates of generators to make preventive control
inferiority solutions, any of which can be a candidate to avoid system emergency.
for the optimal solution. Again, the non-inferiority
means that any attempt to improve any index results in (4) The uSage of the sensitivity matrix in obtaining
deterioration of other indices. As a consequence, the distribution factor results in computational effi-
which solution should be taken as the optimal can no ciency because of not requiring iterative process at
longer be decided by just examining the objective the expense of slight decrease in accuracy 19,131.
 321

In selecting the optimal solution, security

index Isec is first computed with respect to each non-
inferiority solution obtained by the €-constrained
-- Table 4. S p e c i f i e d bus d a t a
Active Reactive Bus
approach. Zk corresponding to the minimum Is,, is the
desired optimal solution. --
Bus

1
Type
P-Q
power
-0.106
power
-0.019
voltage

2 P-Q -0.024 -0.009 -

APPLICATION TO MODEL SYSTEM 3 P-Q 0.0 0.0
4 P-Q 0.0 0.0
The proposed optimization algorithm is applied to 5 P-Q -0.035 -0.023 -
a model system to verify its effectiveness. The model 6 P-Q 0.0 0.0
system shown in Fig. 2 is the IEEE 30 node test system 7 P-Q -0.087 -0.067
which has 6 generators and 41 lines. Generator 8 P-Q -0.032 -0.016
characteristics, that is, the fuel cost, forced outage 9 P-Q 0.0 0.0
rates and NOx emission characteristics are grouped in 10 P-Q -0.175 -0.112
Tables 2 and 3. Tables 4 and 5 show the relevant 11 P-Q -0.022 -0.007
system data. 12 P-Q -0.095 -0.034
13 P-Q -0.032 -0.009
14 P-Q -0.090 -0.058
15 P-Q -0.035 -0.018
16 P-Q -0.082 -0.025 -
17 P-Q -0.062 -0.016
18 P-Q -0.112 -0.075
19 P-Q -0.058 -0.020
20 P-Q 0.0 0.0
21 P-Q -0.228 -0.109
22 P-Q 0.0 0.0 -
23 P-Q -0.076 -0.016
24 P-Q -0.024 -0.012
25 P-v 0.0 0.0 1.071
26 P-v 0.0 0.0 1.082
27 P-v -0.300 1.010
28 P-v -0.942 1.010
P-v 1.045
--
29
30 S
-0.217
0.0 0.0 1.060

Table 5. Line flow capacity

L i n e No. Pis L i n e No. Pi' L i n e No. Pi*

1 0.20 15 0.70 29 0.20

2 0.20 16 0.50 30 0.20
3 0.20 17 0.70 31 0.20
Fig 2. T e s t system 4 0.20 18 0.50 32 0.50
5 0.20 19 0.50 33 0.50
6 0.20 20 0.50 34 0.50
7 0.10 21 0.50 35 0.15
8 0.15 22 0.50 36 0.50
9 0.15 23 0.50 37 0.50
10 0.50 24 0.50 38 0.50
Generator P =a Pc +B Pc +r (Jlh) Po Outages 11 0.50 25 0.20 39 0.50
No. a P r (P.4 ri
12 0.50 26 0.20 40 0.50
13 0.30 27 0.20 41 0.50
1 100 200 10 0.50 0.02 14 0.50 II
28 0.20 [P.U. 1
2 120 150 10
3 40 180 20 L i n e c a p a c i t y i s 110% of s t a n d a r d v a l u e (PI')
4 60 100 10
5 40 180 20 1.00 0.05 Obtaining Non-inferiority Solution Set
6 100 150 10 0.60 0.03
In numerical studies, already introduced genera-
tion cost Fc, NOx emission Fe and line overload index
Fs are taken as the 3 objective functions. Although a
set of non-inferiority solutions is obtained by the E -
constrained technique, the solution process is traced
in some detail to clarify conflicting relations among
Generator F=a+bPc+crca+dexp(ePc) (ton/h) the selected indices.
No. a b C d e
To begin with, the original problem is decomposed
1 4.091 -5.554 6.490 2.OxlO-' 2.857 into 3 subproblems, each having its own scalar objec-
2 2.543 -6.047 5.638 5 . 0 XlO-' 3.333 tive described above. E y,
the independently minimized
3 4.258 -5.094 4.586 l.OxlO-' 8.000 objective of subproblem E i (i=1-3) is shown with
4 5.326 -3.550 3.380 2.0X10-' 2.000 underlines in Table 6 , where,the output of each gene-
5 4.258 -5.094 4.586 l.OxlO-' 8,000 rator is displayed together. These optimized values
6 6.131 -5.555 5.151 l.OxlO-' 6.667 will be used later as the initial parameters in the E-
x10-' x lo-' x10-' constrained technique. A s is clear from this Table,
322

the total generation cost is 606.04 [$/h] under the Determination of Optimal Preference Solution
cost minimization criterion, but it increases to
645.88 [$/h] and 640.62 [$/h] for the cases of the The thus obtained non-inferiority solutions are
emission minimization and line overload minimization Parate-optimal in relation to 3 objective functions
criteria. This is a clear evidence to demonstrate con- Fc, F,, Fs, and hence their comparative evaluation is
flicting nature among the 3 indices. Similar obseva- impossible as it is. This has made it indispensable
tion holds for other indices. the introduction of a preference index in order to
determine the optimal solution out of the candidate
Table 6. Optimal solution of subproblems set. For reason of this, the authors have proposed the

1 I I
security based preference index which is defined in
Merat- Errdssicn LiIEWer- equation (19).
. . .
M u - m m d objectives icn lead index
minimm rrdnimm minimm In actual operation, constraints may be imposed
on total emission in some area depending on time, or
Total generaticn cost ($AI> 606.04 645.88 640.62 on line overload. In such cases, if the solution sur-
Total ardssim (ta-dh) 0.2215 0.1952 0.2350 face is cut by a particular constraint value, inferi-
Total flm daiaticn (p.u.1 0.0381 0.77f8 0.0 ority solution is expressed by a curve such as shown
in Fig. 3. In the below, the determination of the
1 0.124 0.381 0.589 optimal preference solution will be explained for 3
2 0.310 0.515 0.300 different cases with emission constraint, line over-
Generator mtwt 3 0.543 0.562 0.525 load constraint and no constraint
(P.U.) 4 1.016 0.399 1.055
5 0.514 0.522 0.355
1 6 I 0.353 I 0.481 I 0.035 Constraint value: 0.206 ( t / h )

Next, the generation cost is taken as the objec-

tive of main concern, and the remaining subproblems
t r\
are incorporated as constrained parameters. In solving N
this constrained optimization problem, the problem is
converted into a unconstrained optimization problem 3
through the well-known Powell’s method, which is in X
W
turn solved by means of the quasi-Newton method due to
its high computational efficiency.
.a I
U
Y
For the sake of clarity, inferiority solution set
obtained by the €-constrained technique are depicted 9.50 8
in 3-dimensional space shown in Fig. 3, where the
3.00
total cost, emission and line overload constitute
coordinate axes. Note that the obtained set of non-
inferiority solutions corresponds to the curved top 3.50 $
surface in the Figure. For instance, points P(0.206,
0.03, 612.9) and Q(0.206, 0.05, 611.5) on the top
3.00 2
P,
surface in Fig. 3 are examples of non-inferiority
solutions. I
,0.4
,, ,
0.6
,,
0.8
,
,
1.0
,,
1.

FcL
, . (0.205,0.05,614.1,0.0842)
Line flaw index ( x 10-lP.U.)

Fig 4 . Generation cost and preference index

under the emission constraint

(1) Case with emission constraint

Here, it is assumed that the total emission from
6 generators be restricted at 0.206 [ton/h]. The non-
inferiority solution under this constraint is expres-
sed by section planea in Fig. 3, which may adequately
be called an emission constrained plane. Fig. 4 plotts
non-inferiority solution and the preference index with
respect to the line overload index. Point A gives the
minimum value for Isec, and hence the corresponding
solution point is judged as the optimal. Generator
outputs under this optimal operating point are given
in Table 7. It should be noted that there are no
feasible solutions in the region left of point A.

( 2 ) Case with line overload constraint

Assume in this case that the maximum permissible
amount of line overload be specified at 0.05 [p.u].
Non-inferiority solution in this case is expressed by
section plane B in Fig. 3 , which may well be called
line overload constraint plane. ‘Total generation cost
(0.22’2,0.05,6% .l,O.O936) and Is,, are plotted with varying total emission in
Fig. 5 . The Figure has shown that operating condition
corresponding to point B gives the optimal solution;
Fig 3. Surface of mn-inferior solution (Fe,Fs,Fc,Isec) Table 8 lists generator outputs at that point.
 323

Table 7. Preferred optimal solution out of non-inferior Significance of the Preference Index
set under the emission constraint
Preference index Isec employed in this study sig-

ranera tnr I : I 0.303

0.300
nifies the security level of a power system. The index
can immediately be calculated at the time generator
outputs are given. The values of preference index
I
--..--I---

output
(n 11
Y.Y.,
1I 52 0.494
0.802
0.577
Is,, for this model system are displayed in Table 9 ,
in which the bottom 3 rows show results corresponding
to the independently optimized subproblems. The Table
6 0.386 has made clear that the individual optimization both
Total generation cost ($/h) 612.87 for generation cost and emission is inferior from the
Total emission (ton/h) 0.206 security's viewpoint. Conventionally, the optimal
Total power flow deviation(p.u.) 0.030 active power dispatch problem has been formulated to
minimize a single objective such as total generation
Preferential index L t C 8.66 x lo-' cost. However, such an approach has not taken into
account the vulnerability of a power system against
contingencies. In addition, the optimization with
respect to deterministic line overload index is not
Constraint value : 0.05( P .U) necessarily superior in the aspect of security. These
facts reveal that conventional approaches are not
sufficient t o guarantee the system security.

-
614.
The preference index proposed in this paper has a
612. great advantage in that it can qualitatively grasp the
r\
system security in case of forced generator outages.
5 610.
W
?
2
X
2 608. W Table 9. Preference index at each states
sti 606. L)

(0

0.00 Preferential irdex Isec

.rl
604.
N .50
2
a Preferable solutim (1) 8.655 x lo-?
GM .?I Preferable solution ( 2 ) , (3)
.oo J k r a t i m m s t mininmn 9.201
4
U *,;\/ G Fkissim minimm 10.17
8 .50 Pmer flm i d e x minimrn 7.789
Preference index w
.oo
I I I I
0.205 0.210 0.215 0.220 CONCLUDING REMARKS

Total mission (t/h) This paper has presented a novel approach to deal
with the optimal operation and planning problems of a
Fig 5. Generation cost and preference index power system having multiple performance indices.
Having formulated these problems as multi-objective
under the line flow constraint
optimization problems, non-inferiority ( o r , Parate-
optimal) solution set has been obtained through the E-
constrained technique. Furthermore, in order to select
the optimal solution out of the non-inferiority set,
Table 8. Preferred optimal solution out of non-inferior the security-based preference index is proposed which
set under the line flow constraint takes into consideration generator outage probability.

I l l 0.325 The main results obtained through the present

may well be summarized as follows:
study

Generator
-~ -
coutput 1 2 2
3
4
1 0.300
0.521
0.782 (1) In power system operation and planning problems,
there exist multiple objectives to be attained such as
(P.U.) I 5 I 0.555 economy, minimum environment impact, security, etc.,
6 0.378 which are non-commensurable and subject to mutual
Total generation cost interference. The formulation as a multi-objective
Total emission optimization problem has made it possible to quanta-
tively grasp trade-off relations among conflicting
Preferential index 8.42x lo-' objectives.

(2) The solution set of the thus formulated problems

are non-inferior due to contradictions among objec-
tives taken; There are no way of determining the
optimal solution out of the candidate set. Thus, the
(3) Case with no given constraints preference index, which allows for contingencies in a
In case no emission and line overload constraints power system, is introduced to determine the optimal
be imposed, the minimum value of Is,, must be sought solution in a rational manner.
with respect to a set of operating points correspond-
ing to the non-inferiority solution surface in Fig. 3. (3) Numerical tests on the model system have verified
Point 9 in Fig. 3 gives the minimum value of Is,, = the existence of trade-off relations among economy,
0.08415. This value incidentally coincides with that security, and environment protection. The results also
in the previous case. have clearly shown that the optimal preference solu-
324

tion is obtained by menas of proposed preference index University, Tokyo, Japan in 1968, 1970, 1973
Is,,. This index has capability of accounting for load respectively. He joined Mitsubishi Research Institute
level, voltage profile, and outage rates as well as from 1974 to 1978 a d worked as a system researcher in
changes in system configurations. Energy Economics Division. Presently he is an associate
professor of Faculty of Technology at Tokyo
(4) Conventionally, optimization problems with multi- Metropolitan University, Tokyo, Japan. He has been
ple objectives have benn treated so that a particular concerned mainly with planning and operation of large
objective is selected as an optimization index while scale systems and optimization in power systems.
others are processed as security and/or environment He is a member of the Institute of Electrical
constraints. However, solution points obtained through Engineering of Japan, the Society of Instrument and
this artifice are frequently at upper or lower limits Control Engineemg of Japan and the Operation Research
of constrained variables, thus causing instant limit Society of Japan. He is also a member of the IEEE.
violations due to a fault. The proposed method has no
such disadvantages.
Sang Hyun. Bae was born in Naju, Korea on
There is, of course, some room for further re- April 7, 1958. He received B.S., M.S in Electrical
finement associated with the present studies: Engineering from Chosun Universitv. K w a m- Ju.
, Korea in
1982, 1986respectively. He joined Tokyo Institute of
The adequacy of adopting preference index ISec as Technology from 1984 to 1985 and studied as a system
evaluation index researcher in Super Conductor Store. He is currently
Inclusion of other reliability indices not inves- a doctoral course in Electrical Engineering at Tokyo
tigated in this studies Metropolitan University, Faculty of Technology,
Improvement in computational efficiency. Tokyo, Japan. His current research interests are in
Power System Engineering Field with using Planning
and Operation of Large Scale Systems, Optimization
REFERENCES and Mathematical Programing in Power Systems, and
Application of Artifical Intelligence is his recently
A.M. Sasson, "Nonliaear programming solutions for research.
load flow, minimum-loss and economic dispatchin2 He is a member of the IASTED and the Institute of
problems, IEEE Transactions on Power Apparatus Electrical Engineers of Japan and the Institute of
and Systems, vol. PAS-88, No.4, pp.399-409, April Electrical Engineers of Korea. He is also a member of
1969 IEEE.
H.H. Happ, "Optimal power dispatch," IEEE Trans-
actions on Power Apparatus and Systems, vol. PAS-
93, No. 3, plq.820-830, May/June 1974 Tatsuo. Morita
~~ was born in Tokyo, Japan, on
F. Beglari, Optimal economic-environmental gene- July 14,1925. He received the degree of B.S.(in 1951),
ration expansion planning," 5th PSCC Proceedings, M.S.(in 1953) and Dr. Enn.(in 1971) in Electrical
Paper 1.117, vol. 1, September 1975 Engineering from Waseda LhGersity, Japan. Since 1954,
M.R. Gent and J.W. Lamont, "Minimum-emission dis- he has worked at the Tokyo Metropolitan University, and
patch," IEEE Transactions on Power Apparatus and is a Professor of Electrical Engineeriqg at the same
Systems, vol. PAS-90, No. 6, pp. 2650-2660, Nov./ university from 1975. From 1979 to 1983, he was Dean of
Dec. 1971 Faculty of Technology. His fields of interest are
Y. Nakamura, S. Yamashiro and T. Koike, "A method ionizes gases and nuclear engineering. He has written
for startup and shutdown problem of thermal gene- some twenty papers on these fields.
rating units with the total emission constraints" He is a member of the Institute of Electrical
Transaction of IEE of Japan, Vol.l02-B, No. 7, Engineering of Japan, the physical society of Japan and
pp.415-422, July 1972 (to be translated) the Japan Radioisotope Association.
F.W. Gembicki and Y.Y. Haimes, "Approach to per-
formance and sensitivity multi-objective optimi-
zation: The goal attainment method," IEEE Trans-
actions on Automatic Control, vol. AC-20, No. 12, -___
Hiroshi.Sasaki was born in Hiroshima, Japan
pp.769-771, (Short Paper) December 1975
on March 10, 1941. He received B.S., M.S. and Ph.D.
degree all in Electrical Engineering from Waseda
J.G. Lin, "Multiple-objective problems: Parate-
optimal soluticns by method of proper equality University, Tokyo, Japan in 1963, 1965 and 1979,
constraints," IEEE Transactions on Automatic respectively. He is presently a n Associate
Control, vol. AC-21, No. 5, pp.640-650, October Professor of Department of Electrical Engineering,
1976 University of Hiroshima. He was given Visiting
J.K. Delson, "Controlled emission dispatch," IEEE Lecturership from the University of Salford,
Salford, England from 1971 to 1972.
Transactions on Power Apparatus and Systems, vol. He has been interested in various problems in
PAS-93, No. 5, pp.1359-1366, Sept./Oct. 1974
R. Yokoyama and Y.Tamura, "Multi-stage generation the power system engineering field, especially in
transient stability analysis by Lyapunov's direct
shifting in preventive and restorative states in
power systems ," CIGRE, SC-32, 32-73-CE-08, 1973 method, state estimation, micro - processor based
S.H. Bae and R. Yokoyama, "Optimal generation protective devices, etc. Dr.Sasaki is a member
rescheduling based on multi-objective programming of the Institute of Electrical Engineerings of
and probability security index," IASTED Power Japan, and the Atomic Energy Society of Japan.
High Tech'86, 1986 I

Ryuichi. Yokoyama was born in Nagasaki, Japan

on August 25, 1944. He received B.S., M.S. and Dr.Eng.
degree in Electrical Engineering from Waseda