IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS—PART B: CYBERNETICS, VOL. 33, NO.
3, JUNE 2003 531
[6] H. Jürgensen and L. Staiger, “Finite automata encoding geometric fig-
ures,” in Workshop on Implementing Automata WIA 99 Pre-Proceedings
Potsdam 1999, 1999, pp. XII-1–XII-8.
[7] J. R. Büchi, “On a decision method in restricted second order arith-
metic,” in Proceedings of the International Congress on Logic, Method,
and Philosophy of Science. Stanford, CA: Stanford Univ. Press, 1962,
pp. 1–12.
[8] W. Thomas, “Automata on infinite objects,” in Handbook of Theoretical
Computer Science, J. van Leeuwen, Ed. Amsterdam, The Netherlands:
Elsevier, 1990.
[9] H. Samet, Applications of Spatial Data Structures. Reading, MA: Ad-
dison-Wesley, 1993.
[10] B. Boigelot, S. Rassart, and P. Wolper, “On the expressiveness of real and
integer arithmetic automata,” in Proceedings of the 25th International
Colloquium on Automata, Languages and Programming, Lecture Notes
in Computer Science. New York: Springer-Verlag, 1998, vol. 1443, pp.
152–163.
[11] J. Hopcroft and J. Ullman, Introduction to Automata Theory, Languages,
and Computation. Reading, MA: Addison-Wesley, 1979.
[12] B. Nelson, Stochastic Modeling: Analysis and Simulation. New York:
McGraw-Hill, 1995.
[13] L. de Alfaro, “Formal verification of probabilistic systems,” Ph.D. dis-
sertation, Dept. Comput. Sci., Stanford Univ., Stanford, CA, Dec. 1997.
[14] H. Freeman, “Computer processing of line-drawing images,” ACM
Comput. Surv., vol. 6, no. 1, pp. 57–97, 1974.
Optimal Generation Scheduling Based on AHP/ANP
James A. Momoh and Jizhong Zhu
Abstract—This paper proposes an application of the analytic hierarchy
process (AHP) and analytic network process (ANP) for enhancing the selec-
tion of generating power units for appropriate price allocation in a compet-
itive power environment. The scheme addresses adequate ranking, priori-
tizing, and scheduling of units before optimizing the pricing of generation
units to meet a given demand. In the deregulated environment, the clas-
sical optimization techniques will be insufficient for the above-mentioned
purpose. Hence, by incorporating the interaction of factors such as load
demand, generating cost curve, bid/sale price, unit up/down cost, and the
relative importance of different generation units, the scheme can be imple-
mented to address the technical and nontechnical constraints in unit com-
mitment problems. This information is easily augmented with the optimiza-
tion scheme for an effective optimal decision. The scheme proposed is tested
using the IEEE 39-bus test system.
Index Terms—Analytic hierarchy process, analytic network process,
deregulated power industry, optimization, unit commitment.
I. INTRODUCTION
Since generators cannot instantly turn on and produce power, unit
commitment must be planned in advance so that enough generation is
always available to handle system demand with an adequate reserve
margin in the event that generators or transmission lines go out or
load demand increases. Unit commitment handles the unit generation
schedule in a power system for minimizing the operating cost and sat-
isfying the prevailing constraints such as load demand and system re-
Manuscript received June 1, 1999; revised March 26, 2001. This work was
supported by Electric Power Research Institute (EPRI), NSF RIMI, and CESaC.
This paper was recommended by Associate Edtior K. Pattipati.
J. A. Momoh is with the Department of Electrical Engineering, Center for
Energy Systems and Control, Howard University, Washington, DC 20059 USA
(e-mail: jmomoh@nsf.gov).
J. Zhu is with Alstom EAI, Bellevue, WA 98004 USA (e-mail:
jizhong.zhu@alstom.esca.com).
Digital Object Identifier 10.1109/TSMCB.2003.811122
1083-4419/03$17.00 © 2003 IEEE
serve requirements over a set of time periods [1]–[5]. Optimization
techniques such as augmented Lagrangian relaxation, dynamic pro-
gramming, and the branch–and–bound algorithm have been used to
solve the classic unit commitment problem. In the deregulated power
industry, the unit commitment problem is, in general, very complicated
with a large number of possible decision variables such as load demand,
generating cost curve, time-varying bid price of generation power, and
the relative importance of different generation units. The classical op-
timization techniques are insufficient for handling these factors. Thus,
in recent power system operation, it is necessary to modify the conven-
tional unit commitment (UC) planning of thermal units with network
constraints and the value of generator bidding as additional constraints.
The classical UC problem is aimed at determining the startup and shut-
down schedules of thermal units to meet forecasted demand over cer-
tain time periods (24 h to 1 week) and belongs to a class of combina-
torial optimization problems. The methods [10], [11] that have been
studied so far fall into roughly three types: heuristic search, mathemat-
ical programming MP, and hybrid methods.
Although these techniques are effective for problem posed, they do
not handle network constraints and bidding issues. The paper proposes
future UC requirements in a deregulated environment where network
constraints, reliability, value of generation, and variational changes in
demands and other costs may be factors.
The classical UC Lagrange method cannot solve the problem due to
combinatorial explosion. Accordingly, as an initial approach to solve
this complex problem, we attempt to find a method for solving UC
considering network limitation and generation bids as a daily opera-
tional planning problem. This approach supports the decision making
effectively of ranking units in terms of their values by using the ana-
lytic hierarchy process (AHP) and the analytic network process (ANP)
techniques. The scheduled generation over time is studied as input into
the optimal power flow (OPF) problem for optimal dispatch within the
network and generation constraint.
This paper proposes application of the AHP [6], [8] and the ANP [7]
for enhancing the selection of generating power units for appropriate
price allocation in the deregulated power industry. It is different from
the traditional unit commitment problem, which can be solved by di-
rectly using the optimization methods.
Explanation of Proposed Scheme
The basic concept of proposed optimal generation scheduling is as
follows:
First, it is assumed that the ranking of generating units, and their
priority as well as demand is known. As a result, the preferred genera-
tors for competitive scheduling and pricing will be known. Therefore,
the number of generators whose fuel consumption constraints must be
considered can be reduced considerably. This reduces the difficulties
of unit commitment and optimal power flow. The proposed scheme ad-
dresses adequate ranking and prioritizing of units before optimizing the
pricing of generation units to meet a given demand. By incorporating
the interaction of factors, such as load demand, generating cost curve,
bid/sale price, unit up/down cost, and the relative importance of dif-
ferent generation units, the scheme can be implemented to address the
technical and nontechnical constraints in unit commitment problem.
This information is easily augmented with the optimization scheme for
effective optimal decisions. The scheme consists of the three following
stages:
1) ranking of units in terms of their values by AHP/ANP;
2) checking the constraints by rule–based method;
3) solving optimization problem by interior point optimal power
flow.
532 IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS—PART B: CYBERNETICS, VOL. 33, NO. 3, JUNE 2003

Therefore, if the number of generators is N G, and the duration of

the period under consideration is N T , the objective function is
NG NT
2
min : (ci + bi Pgi (t) + Pgi (t) )xi (t)
P (t); u (t)
i=1 t=1

2
+(Sgi (t) ui (t) (1 0 u (t 0 1))
i (1)
where xi (t) is the cumulative up-time if xi (t) > 0 and the cumulative
down-time if xi (t) < 0, where ui (t) is 0–1 variable expressing the
state (0: shutdown, 1: startup) of the ith generator in period t.

Constraints

Fig. 1. Scheme for optimal generation scheduling.
Next, for all generators committed, the network availability for transfer
power, the constraints on startup and shutdown, and generated output
and reserve are determined for a daily operational planning. In the
daily UC calculation, a Lagrange method is used without network con-
straints. Since the majority of connected generators include network
constraints, and other equipment limitation to ensure feasibility, an
OPF technique based on the extended quadratic interior point (EQUIP)
method [9] is proposed for solving the resulting optimal generation
scheduling problem. The scheme converges rapidly, and speed-up and
its ability to handle nonlinearity is not sacrificed. This gives the pro-
posed scheme a significant advantage over classical heuristic or La-
grange methods. Further work to evaluate this technique is ongoing for
multiutility areas where reliability and stability constraints on the net-
works are requirements. The proposed scheme is tested on IEEE 39-bus
system.
According to the above discussion, the scheme for optimal genera-
tion scheduling can be represented as Fig. 1.
The constraints can be classified as coupling constraints and local
constraints. The coupling constraints are related to all generators (in
service) under consideration, regardless of age or efficiency, and the
following are considered.
Demand-Supply Balance Constraint
The sum of the generator outputs must be equal to the demand PD (t)
NG
Xgi (t)Pgi (t) = PD (t): (2)
i=1
Again Xi (t) is a 0,1 variable expressing the state i.e., (0: shutdown, 1:
startup) of the ith generator in period t.
Reserve Power Constraint
In order to deal with unpredictable disturbances (interruption of gen-
eration and transmission lines or unexpected increase in demand), the
output of generators in operation must increase, and hence, the instan-
taneous reserve power shown in the equation below must be required
NG

The paper is organized as follows. Section II describes the formu- Xsi (t)rsi (t)  R (t) s (3)
lation of optimal generation scheduling including unit commitment i=1

and OPF with UC control. Section III discusses the application of
AHP/ANP to unit commitment. Section IV discusses the implemen-
tation steps of rule-based UC. Section V describes the integrated OPF
and UC scheme. Section VI consists of test examples and analysis,
followed by conclusions.
where rsi (t) is the contribution of unit i to spinning reserve at hour t,
and Rs (t) is the operational reserve requirement at period t.
Generator Output Constraint
When the generator is in the midst of startup, its output must be
between the upper limit Pgi; max and lower limit Pgi; min

II. FORMULATION OF OPTIMAL GENERATION SCHEDULING

xi (t)Pgi; min  x (t)P (t)  x (t)P
i i i gi; max : (4)
For unit ramp rate conditions
Objective Functions
Pgi (t) 0 P (t 0 1)  uP ;
gi gi for unit ramp up of unit i;
In general, in UC problems, the objective function to be minimized
is the sum of the operation and start–up costs. First, the fuel cost of the Pgi (t 0 1) 0 P (t)  DR ;
gi gi for unit ramp down of unit i:
generation is a function of its output Pi .
(5)
For simplicity, we assume
For each selected generator for bid
NG
2
The constraint on bid price for unit i at period t is
Fg (Pgi (t)) = ci + bi Pgi (t) + ai Pgi (t)
i=1 Bgi (t) > BPgi; min (t); i 2 NG (6)

where Pgi (t) is the real power output of the ith generator in period t. where Bgi (t) is bid price of unit i at time t.
Pgi (t) is assumed to be within maintenance schedule i.e., consid-
Network Limitation
ered to be at an acceptable efficiency to meet prescribed load. It should
be noted that machines being committed are not operating at 100% ef- To account for network limitation during UC dispatch, the network
ficiency due to imperfect operating conditions and aging. and operation constraints are specified as additional constraints below.
The start–up cost, on the other hand, increases with shutdown time 1) Power Flow Equation: The power flow equation accounting for
of generator. We assume, for the paper, that the boiler and turbine cool generator i to demand i with losses are given as
down after shutdown and the cost of preheating increases with shut-
down time and is embedded in Sgi (t) (start up cost of generator i at
Pgi (t) 0 P (t) = F
di V; ; T; t)
pi (

time t). Qgi (t) 0 Q (t) = F

di qi ( V; ; T; t) (7)
IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS—PART B: CYBERNETICS, VOL. 33, NO. 3, JUNE 2003 533

where
NG

Fpi = Vi (Vj Yij cos (i 0 0 j ij ) ; Ti ; t)

j =1

NG

Fqi = V (Vj Yij sin (i 0 0 j ; Ti ; t)

ij ) (8)
j =1

where the transformer taps in the circuit within limits to minimum loss
or voltage deviation
Ti; min (t)  T (t)  T
i i; max ( t):
The minimum operation time and minimum shutdown due to fatigue
limit of the generator
Fig. 2. Hierarchical network model of units rank.
tup; min  t  tup max ;

and similarly TABLE I

JUDGMENT MATRIX A–PI
tdown; min  t  tdown max : ; (9)
Limits on flows are defined as
Vi2 + Vj2 0 2Vi Vj cos(i 0 j )
ZL2  IL;2 max: (10)

This ensures that flows in each line due to outage on line j is within
prescribed limits.
Also, each generator is also required to maintain one of the following
generator limits: III. APPLICATION OF AHP/ANP TO UC
xi (t)Qgi; min  x (t)Q  x (t)Q
i gi i gi; max (11)
The AHP is a decision-making approach [6], [7]. It presents the al-
Vgi; min (t)  V (t)  V max (t)
gi gi; (12) ternatives and criteria, evaluates tradeoff, and performs a synthesis to
arrive at a final decision. AHP is especially appropriate for cases that
and for demand buses (load buses):
involve both qualitative and quantitative analyzes. The ANP is the ex-
Vdi; min (t) V di (t) V di; max ( t): (13) tension of AHP. It makes decisions when alternatives depend on criteria
The problem posed can be solved by many optimization methods with multiple interactions [7].
such as Lagrange relaxation methods [11], heuristic rules, and optimal According to the theory of AHP/ANP, the following AHP/ANP
power flow with decomposition techniques. The Lagrange method uti- model in Fig. 2 is devised to handle ranking of units.
lizes the following primal problem: The hierarchical network model of units ranking consists of three
Given sections:
min F (xi (t)Pi (t) + s(t)
1) the unified ranking of units;
2) the ranking criteria or performance indices, in which the P IC
s.t. (14) reflects the relative importance of units;
3) generating units G1 ; . . . ; Gm .
1) local coupling constraints (2) to (5);
The performance indices P IG , P IS , and P Ib are defined as
2) power flow constraints (7) and (8), given as gi (xi (t), Pgi (t))
 0 i = 1; . . . ; N G. P IG = 1=Fgi (Pgi (t)) (17)
The function F expresses the sum of fuel consumption and start–up P IS = 1=Sgi (t) (18)
cost. Using the Lagrange multiplier, we determine , , and  , which
are introduced in the Lagrangian function as follows: P Ib = 1=BPgi (t): (19)
L [xi (t); Pgi (t); (t); (t);  ] The four ranking criteria P IG , P IS , P Ib , and P IC are interacted.

0 0
The basic principle of AHP/ANP is to calculate the eigenvector of the
= F (xi (t); Pgi (t); s(t)) (t) Pgi (t) PD (t) alternatives for each criterion. For qualitative factors such as the relative
importance of units and criteria, the corresponding eigenvectors can be
+ (t) xi (t)Pgi; max R(t)0 0 0
Pgi (t) gi (i) : (15) obtained by computing the judgment matrix. The rule based method for
This is usually converted to a dual problem where forming the judgment matrix may be found in [6]–[8]. The judgment
max fmin [x(t); Pg (t); (t); (t);  ]g (16) matrix for representing the relative importance of four criteria is given
as Table I.
s.t. The ranking results of units for each time stage will be obtained from

gi (xi (t); Pg (t) 0) : AHP/ANP calculation. The list of unit ranking shows the priority of
To include the network constraints and bidding of generators, a new units to be committed at each time stage. However, it has not considered
UC–based OPF/AHP is proposed. Namely, we solve for the UC the constraints such as system real power balance and system spinning
problem over time using OPF to account for the network voltage, reserve requirement.
transformer, and flow constraints. Application of the EQUIP opti- This paper adopts the rule-based method to solve this problem.
mization method [9] solves for the optimal operating point at each AHP/ANP is used to decide total ranking of all units for each time
time period. The second phase of our algorithm uses AHP/ANP to stage, and the rule-based system decides the commitment state of units
determine the value and merit of each generation bid to be submitted according to the system power balance and system spinning reserve
for commitment. requirement. So the final unit commitment results are obtained through
534 IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS—PART B: CYBERNETICS, VOL. 33, NO. 3, JUNE 2003
TABLE II
DAILY LOAD DEMANDS IN MEGAWATTS
TABLE III
GENERATING UNIT DATA
the communication between AHP/ANP ranking and the rule–based
constraints checking.
IV. IMPLEMENTATION STEPS OF RULE-BASED UC
As mentioned in Section III, the priority ranking of all units for each
time stage can be obtained by AHP/ANP. This priority rank considers
the nontechnical constraints and nonquantitative factors, but it does not
involve the constraints of power balance and reserve requirements in
the unit commitment. Therefore, the rule-based method is used to co-
ordinate this problem. The implementation steps of the rule-based unit
commitment are as follows.
Step 1) Select the number 1 unit from the priority rank of units at
hour t.
Step 2) Check the constraints of the ramp up/down of the unit.
If the constraints are satisfied, go to step 4.
Step 3) If the constraints of the ramp up/down of the unit are not
satisfied, discard this unit at hour t. Select the next unit from
the priority rank of units, and go to step 2.
Step 4) Check the power balance. If system power can be balanced,
go to step 5. Otherwise, add one more unit according to the
priority of units, and go to step 2.
Step 5) Check the spinning reserve at hour t. If the system has
enough spinning reserve, go to the next step. Otherwise, add
one more unit according to the priority rank of units, and go
to step 2.
Step 6) Stop. All units that were not selected as well as those that
have been discarded in the selection will not be committed
at hour t. The other units will be committed at hour t.
V. TEST EXAMPLE
The proposed approach is examined with the IEEE 39-bus test
system. It has ten generators, i.e., G30, G31, G32, G33, G34, G35,
G36, G37, G38, and G39. The daily load demands are given as in
TABLE IV
BID PRICE OF GENERATION POWER OVER A SET OF TIME PERIOD
IN DOLLARS PER MEGAWATT PER HOUR.
TABLE V
UNIT COMMITMENT WITHOUT TRANSMISSION
SECURITY AND VOLTAGE CONSTRAINTS
TABLE VI
UNIT COMMITMENT WITH TRANSMISSION SECURITY
AND VOLTAGE CONSTRAINTS
Table II. The generating unit data are given as in Table III. Table VI
shows the bid price of generation power over a set of time periods.
The calculation results of unit commitment are listed in Tables V
and Table VI. Table V is the unit commitment schedule obtained from
AHP/ANP and rule–base method. It has not considered the voltage
security and transmission security constraints. The corresponding
power flow solution also violates voltage limits and transmission
security limits.
From Table V, we find that power flows at hours 1, 2, 4, 5, 8, 22, and
24 are infeasible. Table VI is the final unit commitment schedule with
OPF corrections. It satisfies the voltage security and transmission secu-
rity constraints. The total generation cost for unit commitment schedule
in Table VI is $11 391.00. If the commitment states of units are taken
as the input of OPF, the total optimal generation cost will be reduced
to $11 159.60.
VI. CONCLUSIONS
This paper presents the application of the analytic hierarchy process
and analytic network process to unit commitment in the deregulated
IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS—PART B: CYBERNETICS, VOL. 33, NO. 3, JUNE 2003
industry. The technical and nontechnical constraints as well as non-
quantitative factors appeared in the deregulated environment are han-
dled by AHP/ANP and rule–based method. The integrated scheme of
unit commitment and optimal power flow is also proposed to consider
the voltage security and transmission security constraints. It ensures
the feasibility of the final solution of the unit commitment. Test results
validate the proposed approach.
REFERENCES
[1] A. I. Cohen and M. Yoshimura, “A branch and bound algorithm for unit
commitment,” IEEE Trans. Power Syst., vol. PAS-101, pp. 444–451,
1982.
[2] A. I. Cohen and S. H. Wan, “A method for solving the fuel constrained
unit commitment,” IEEE Trans. Power Syst., vol. 1, pp. 608–614, Feb.
1987.
[3] S. Vemuri and L. Lemonidis, “Fuel constrained unit commitment,” IEEE
Trans. Power Syst., vol. 7, pp. 410–415, Feb. 1992.
[4] S. Ruzic and N. Rajakovic, “A new approach for solving extended unit
commitment problem,” IEEE Trans. Power Syst., vol. 6, pp. 269–277,
Feb. 1991.
[5] E. H. Allen and M. D. Ilic, “Stochastic unit commitment in a deregu-
lated utility industry,” in Proc. 29th North Amer. Power Symp., Laramie,
Wyoming, Oct. 1997, pp. 105–112.
[6] T. L. Satty, Multicriteria Decision Making: The Analytic Hierarchy
Process. Pittsburgh, PA, USA: RWS, 1990.
[7] , Decision Making With Dependence and Feedback: The Analytic
Network Process. Pittsburgh, PA: RWS, 1996.
[8] J. Z. Zhu and J. A. Momoh, “Optimal VAr pricing and VAr placement
using analytic hierarchy process,” Elec. Power Syst. Res., vol. 48, pp.
11–17, Dec. 1998.
[9] J. A. Momoh, S. X. Guo, and A. R. Ogbuobiri, “The quadratic interior
point method for solving power system problems,” IEEE Trans. Power
Syst., vol. 9, pp. 1327–1336, 1993.
[10] G. S. Lauer, D. P. Bertsekas, N. R. Sandell, Jr., and T. A. Posbergh, “So-
lution of large-scale optimal unit commitment problems,” IEEE Trans.
Automat. Contr., vol. AC-28, pp. 1–11, 1982.
[11] A. Merlin and P. Sandrin, “A new method for commitment at electricité
de France,” IEEE Trans. Power Syst., vol. PAS-102, pp. 1218–1255,
May 1983.
1083-4419/03$17.00 © 2003 IEEE
535
Flexible Computation of Shape Symmetries Within the
Maximal Disk Paradigm
Gert J. van Tonder and Yoshimichi Ejima
Abstract—Shape symmetries, like medial axes [1], contain a wealth of
information useful in engineering and many different models for sym-
metry computation have been proposed. We present the hybrid symmetry
transform (HST) as a compactly reformulated version of the maximal disk
methods. HST is inspired by shunting inhibition networks [2] and general
concepts in wave dynamics. Through adjusting a shunting coefficient, the
strength of adherence to the maximal disk paradigm becomes tunable.
Each image location is scanned with a set of concentric rings which are
then combined via weighted (shunting) summation, as opposed to the
winner-takes-all approach in strict maximal disk methods. The new model
is simple yet operates flexibly to compute maximal disk symmetries in a
variety of conditions, interpreted as different wave propagation modes.
Index Terms—Concentric rings, HST, inhibition networks, maximal disk
symmetry, medial axes, shunting.
I. INTRODUCTION
Mosts of us know the pattern of ripples that follow when a handful of
pebbles is cast into a pond. Ripples emanating from two point sources
move through each other [Fig. 1(A)] in a familiar interference pattern,
because waves in air and water are spatially superimposed, diffused
and damped, obeying classical superposition wave dynamics. But other
modes of wave propagation exist that have potential benefits to engi-
neering applications like image processing.
Consider a burning grass fire. Burning wave fronts in grass block
each other where they meet because there simply is no more grass left to
burn—the medium in which waves propagate is itself consumed by the
waves. Harry Blum [3], [4] was intrigued by blocking waves of grass-
fires because they trace out a line of points equidistant to wave sources
[Fig. 1(B)]. Calling this the symmetry axis of the “input” pattern he
pioneered a powerful image processing method. For example, the tri-
angular shape [Fig. 1(C)] has an inverted “Y”-shaped symmetry axis
which can be converted into an abstract two-dimensional (2-D) graph
of shape topology to be used in pattern recognition. Leyton presented
a detailed discussion on the elegant duality between contour curvature
and symmetry axes [5].
Blum showed that grassfires are equivalent to maximal included
disks [4]. A maximal disk is defined as an included disk (i.e., a disk of
which all points overlap with the given shape) which touches the shape
outline at more than one point while not being included in any other
included disk, as shown in Fig. 1(D). Superposition of the centres of
all maximal disks gives the symmetry axis of a shape, a method now
known as the medial axis transformation (MAT) [1].
For the purposes of this paper mainly three models are considered
from the multitude of existing methods. Annular symmetry operators
[6] begin with multi scale oriented filtering of the image, followed by
symmetry point detection via scale dependent annular symmetry op-
erators. Skeletons are then extracted from symmetry points and char-
acterized as either blobs or limbs (see the angular profile described in
[6]) upon which an abstract object graph describing part topology is
constructed.
Manuscript received June 1, 1999; revised April 21, 2002. This paper was
recommended by Associate Editor B. J. Oommen.
The authors are with the Graduate School for Human and Environmental
Studies, Kyoto University, Kyoto Prefecture 606-8501, Japan (e-mail:
gvtonder@yahoo.co.uk).
Digital Object Identifier 10.1109/TSMCB.2003.810952