1256 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 18, NO.

4, NOVEMBER 2003

Short-Term Hydrothermal Generation Scheduling

Model Using a Genetic Algorithm
Esteban Gil, Student Member, IEEE, Julian Bustos, Senior Member, IEEE, and Hugh Rudnick, Fellow, IEEE

Abstract—A new model to deal with the short-term generation

scheduling problem for hydrothermal systems is proposed. Using
genetic algorithms (GAs), the model handles simultaneously the
subproblems of short-term hydrothermal coordination, unit com-
mitment, and economic load dispatch. Considering a scheduling
horizon period of a week, hourly generation schedules are obtained
for each of both hydro and thermal units. Future cost curves of
hydro generation, obtained from long and mid-term models, have
been used to optimize the amount of hydro energy to be used during Fig. 1. Time hierarchical decomposition for the HGSP.
the week. In the genetic algorithm (GA) implementation, a new
technique to represent candidate solutions is introduced, and a set sides, it is a large multi modal and nonconvex problem. Most
of expert operators has been incorporated to improve the behavior
of the conventional optimization techniques are unable to pro-
of the algorithm. Results for a real system are presented and dis-
cussed. duce near-optimal solutions for this kind of problem. Moreover,
conventional methods usually require certain suppositions that
Index Terms—Genetic algorithms, hydrothermal systems,
short-term hydrothermal scheduling. force them to work with simplified instead of realistic models.
In order to deal with the HGSP in a more efficient and robust
way, this paper proposes an optimization model using a genetic
I. INTRODUCTION algorithm (GA) to solve it.
A GA is a metaheuristic technique inspired on genetics
T HE efficient scheduling of available energy resources for
satisfying load demand has became an important task in
modern power systems. The generation scheduling problem
and evolution theories [4]. During the last decade, it has
been successfully applied to diverse power systems problems:
consists of determining the optimal operation strategy for the optimal design of control systems [5], [6]; load forecasting
next scheduling period, subject to a variety of constraints. For [7]; OPF in systems with FACTS [8]–[10]; FACTS allocation
hydrothermal systems, the limited energy storage capability [11]; networks expansion [12]–[14]; reactive power planning
of water reservoirs, along with the stochastic nature of their [15]–[17]; maintenance scheduling [18], [19]; economic load
availability, make its solution a more difficult job than for dispatch [20], [21]; generation scheduling and its subproblems
purely thermal systems. The well-timed allocation of hydro en- [22]–[34].
ergy resources is a complicated task that requires probabilistic Section IV presents an overview of GA, and describes the im-
analysis and long-term considerations, because if water is used plementation of the proposed model using a GA, and Section V
in the present period, it will not be available in the future, shows tests results for test systems. Finally, Section VI presents
increasing in this way the future operation costs. the main conclusions of the paper.
So, the hydrothermal generation scheduling problem (HGSP)
is usually decomposed into smaller problems in order to solve it II. PROBLEM FORMULATION
[1]. In this way, the HGSP involves three main decision stages, A. Hydrothermal Coordination Problem (HCP)
usually separated using a time hierarchical decomposition
It is the first stage in the solution of the HGSP. The HCP con-
(Fig. 1): the hydrothermal coordination problem (HCP), the
sists of determining the optimal amounts of hydro and thermal
unit commitment problem (UCP), and the economic load
generation to be used during a scheduling period [1], [2]. The
dispatch problem (ELDP). The model proposed in this paper
HCP is also decomposed in long-, mid-, and short-term models
handles simultaneously the subproblems of short-term HCP,
[35], depending on the reservoirs storage capacity.
UCP, and ELDP.
Decisions in hydrothermal systems are coupled in time. In
The HGSP is a nonlinear optimization problem with high di-
other words, the operating costs in the future depend on the
mensionality, continuous and discrete variables, a nonexplicit
amount of hydro generation during the present period [36]. Ac-
objective function, with equality and inequality constraints. Be-
cording to the kind of output of the model, HCP approaches can
be classified in two principal categories:
Manuscript received November 13, 2002.
E. Gil was and J. Bustos is with the Department of Electrical Engi- • fixed reservoir storage level for each stage: the use of the
neering, Universidad Técnica Federico Santa María, Valparaíso, Chile water in each stage is determined strictly by the model;
(e-mail: julian.bustos@vra.utfsm.cl). • future cost functions (FCF): future or opportunity cost of
H. Rudnick is with the Department of Electrical Engineering, Universidad
Católica de Chile, Santiago, Chile (e-mail: h.rudnick@ieee.org). the water used during the present stage, versus the storage
Digital Object Identifier 10.1109/TPWRS.2003.819877 level at the end of the scheduling period.
0885-8950/03$17.00 © 2003 IEEE

Authorized licensed use limited to: Universidad de La Salle. Downloaded on December 14,2020 at 09:19:24 UTC from IEEE Xplore. Restrictions apply.
GIL et al.: SHORT-TERM HYDROTHERMAL GENERATION SCHEDULING MODEL USING A GENETIC ALGORITHM 1257

timal allocation of power demand among the running thermal

units, satisfying the power balance equations and the unit’s op-
eration constraints.
When the ELDP is solved in the context of the online oper-
ation of the system, transmission losses are usually included in
the optimization process, and sometimes even an optimal power
flow is executed. However, in the context of the selection of an
optimal schedule, there is evidence that losses do not have much
influence and they are not included.

D. Mathematical Formulation for the Short-Term HGSP

The main objective of the short-term HGSP is to determine
the optimal generation level for each hydro and thermal unit for
each hour over an entire period (a day or week), subject to a
Fig. 2. Immediate and future cost functions [36]. large set of equality and inequality constraints.
The objective function of the short-term HGSP is represented
Hydro generation has an opportunity cost associated to the by (1). The objective function is set as to minimize the total
thermal generation displaced. As seen in Fig. 2, if a larger operation costs plus a
amount of water is used during the present stage, the immediate penalty factor (feasibility measure). (Please see the equation at
cost (obtained by solving the UCP and the ELDP for the present the bottom of the page.) Where
stage) decreases, and the water available in the future (final total system operation cost;
water storage) decreases. Hence, if less water will be available, fuel costs for hour obtained from the ELDP;
future costs will increase. number of hours for the time horizon;
The FCF can be obtained by calculating recursively the number of thermal units;
system operation costs in the future (starting from the end of number of hydraulic reservoirs;
the period under analysis) considering different starting values status of thermal unit during the hour (1
of water storage. This can be achieved using stochastic dual for up and 0 for down);
dynamic programming (SDDP) [36]–[39]. power output for the thermal unit during the
The FCF allows uncoupling the long/mid-term from the hour ;
short-term hydrothermal coordination activity. This is the power output for the reservoir during the
approach used in this paper. hour ;
fuel cost for the thermal unit during the hour
B. Unit Commitment Problem (UCP)
with a power output (using a quadratic
Once the hydroelectric generation for each hour is de- cost function);
termined, thermal units must meet the load not covered by and start-up and shut-down costs for the thermal
hydroelectric generation. The UCP deals with the decision unit during the entire scheduling horizon;
on which of the thermal units will be running or not during volume for reservoir during the hour ;
each hour of the scheduling period [1], [3]. The committed volume for reservoir at the end of the
units must be able to meet the system load at minimum op- horizon;
erating cost, subject to a variety of constraints. The UCP is a future cost of thermal units as a function of
NP—complete optimization problem. the volume of reservoir at the end of the
scheduling horizon;
C. Economic Load Dispatch Problem (ELDP) Penalty penalty factor.
Once the running units for an hour have been determined by Immediate costs are calculated as the sum of the energy pro-
the solution of the UCP, it is necessary to distribute the load de- duction costs and the start-up and shutdown costs, during the
mand solving the ELDP. The ELDP consists of finding the op- present time horizon (from 1 to ). The energy production cost

(1)

Authorized licensed use limited to: Universidad de La Salle. Downloaded on December 14,2020 at 09:19:24 UTC from IEEE Xplore. Restrictions apply.
1258 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 18, NO. 4, NOVEMBER 2003

, for a given hour , corresponds to the solution of the ELDP

for that hour, considering only the dispatched units.
On the other hand, the future cost for each reservoir is calcu-
lated using the respective FCF, as a function of the volume at
the end of the time horizon.
The penalty factor is directly proportional to the level of vio-
lation of constraints. In this way, it works as a feasibility mea-
sure.
From (1), it can be appreciated that the objective function
is not explicit, because the value of is obtained through the
solution of the ELDP, instead of a direct function evaluation.
The short-term HGSP presents a large set of units and system
constraints, which are taken into account in this paper, as fol-
lows:
1) Demand satisfaction for each hour :

(2)

where and are the total load demand

forecasted for the hour , total power output of hydraulic
units without water storage capacity during the hour and
the total losses estimated for the system during hour ,
respectively.
In order to accomplish this rule, the model incorporates
a fictitious unit, whose cost function corresponds to the
failure cost for the system.
2) Technical operation limits of each unit Fig. 3 Proposed model.

4) Limit storage capacity for each reservoir

(3)
(6)

(4) where and are the minimum and max-

imum feasible volumes for reservoir .
where and are the minimum and max- 5) Spinning reserve requirements
imum power output of the thermal unit and
are the minimum and maximum power output
of the hydro unit and are the up and
down-time at hour for the thermal unit is
the minimum up time for thermal unit and (7)
is the minimum down time for thermal unit .
3) Hydraulic dynamic of each reservoir for each hour
where corresponds to the percentage of the load demand
to be used as reserve ( in this case).

III. PROPOSED MODEL

(5)
A scheme of the proposed model is given in Fig. 3. As input
where is the forecasted inflow; is the information, the proposed model uses the FCF obtained from
discharge for a power output is the filtration; a long/mid-term model, detailed information on the hourly
is the evaporation; and is the spillage. load demand, the reservoir inflows and water losses, models of
Each hydrothermal power system has its own particular the hydro and thermal generating units and initial conditions,
hydraulic restrictions, depending mainly on geographical among others.
and hydrological conditions. Sometimes, water discharge The proposed model uses this input information, handling si-
from one reservoir can affect availability in another reser- multaneously the subproblems of short-term hydrothermal co-
voir, the so-called hydraulically coupled units. ordination, unit commitment, and economic load dispatch. Con-

Authorized licensed use limited to: Universidad de La Salle. Downloaded on December 14,2020 at 09:19:24 UTC from IEEE Xplore. Restrictions apply.
GIL et al.: SHORT-TERM HYDROTHERMAL GENERATION SCHEDULING MODEL USING A GENETIC ALGORITHM 1259

Fig. 5. Future cost calculation process for the hydraulic unit j .

Fig. 4. Candidate solution representation (matrix G ). means of heuristic rules based on the expert knowledge of the
system and using a priority list.
TABLE I
BINARY CODIFICATION EXAMPLE USING 3 b
C. Fitness Evaluation
To compare different solutions, a fitness (or cost) evaluation
of each candidate solution must be done. It is achieved by means
of the decoding of the strings and the evaluation of the objec-
tive function (1) for each candidate solution. In order to achieve
the fitness evaluation, the following steps are executed for each
sidering an analysis horizon period of a week, the proposed candidate solution.
model obtains hourly generation schedules for each of the hydro 1) For each hydro sub-matrix , (from 1 to ),
and thermal units. columns are decoded and final volume for each reservoir
is calculated. Then, weekly FCFs for hydro generation
IV. IMPLEMENTATION OF THE MODEL USING GA have been used to obtain the opportunity cost due to the
use of hydro energy during the week (Fig. 5).
The GA is a search technique inspired on genetics and evolu-
2) Generation of hydro units is discounted from total load
tion theory. They are described in [4], [45]–[48]. The implemen-
demand for each hour. Thermal demand (total minus
tation of the proposed model using a GA includes the following
hydro) must be satisfied by running thermal units at
stages.
least cost. Then, for the running thermal units for each
A. Representation of Candidate Solutions hour (obtained from vectors ), an economic load
dispatch is achieved. The ELDP is solved using Lagrange
Each candidate solution is represented by a binary matrix , multipliers [1]. Production costs for each thermal unit
(Fig. 4), by means of an adequate codification of the decision over the week are calculated.
variables. Each matrix representing a candidate solution must 3) Analyzing each vector , start-up and shutdown costs
contain all of the information necessary to be distinguished from are calculated using (8). As in [23], [25], and [34],
another one, and necessary to evaluate its fitness. The decision is equal to 0 for each thermal unit , and is equal
variables are either to the cold start cost or to the hot start
1) Power output of each hydroelectric unit for each hour: cost , depending of the time that the unit has
it is a continuous variable, which is discretized using a been down
3-bit code. So, there are eight possible discrete power
generation levels for each unit. The generation levels for (8)
each 3-bit combination are assigned arbitrarily, as seen in
Table I. Then, each candidate solution contains a set 4) Specialized subroutines determinates if each constraint is
of binary submatrices with size (3, ) for each hydro violated, and penalty factors are calculated.
unit .
2) Status of each thermoelectric unit for each hour: 1 if the
unit is running, 0 if the unit is down. Then, each candidate D. Offspring Creation
solution contains also a set of binary vectors with
Creation of new individuals is a fitness-dependant activity,
length for each thermoelectric unit .
due to solutions with best fitness have more probabilities to be
selected as parents. The offspring creation process used in this
paper (Fig. 6) involves three groups of genetic operators.
B. Initialization 1) Crossover Operators: The crossover operators select
An initial population of candidate solutions is created randomly (but better solutions have more chances to be se-
randomly, and “seeded” with some good solutions obtained by lected) two parent solutions and then combine their respective

Authorized licensed use limited to: Universidad de La Salle. Downloaded on December 14,2020 at 09:19:24 UTC from IEEE Xplore. Restrictions apply.
1260 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 18, NO. 4, NOVEMBER 2003

search space is allowed. To investigate the effect of the repair

operators, a sensibility analysis was performed in a small test
system with four units (not shown in this paper). It could be
seen that the inclusion of repair operators implies a faster con-
vergence and solutions closer to the optimal.

E. Replacement of the Population Members

Fig. 6. Offspring creation process.

In order to create a new and improved population of solutions,
a parents versus descendants competition is achieved, where
best solutions survive and bad solutions disappear. The replace-
strings based in some rules, generating new population mem- ment procedure used is the selection, used successfully
bers. To achieve the parent selection, tournament selection has in [34]. It can be described as:
been used. Step 1) For each solution G of the present population, se-
Three different kinds of crossover operators were used (one at lect randomly (using an uniform distribution) an off-
a time), performed with probabilities , and , respec- spring .
tively, (with ). Step 2) If , then add to the new
Window Crossover: For two selected parents, it selects population. If , then add .
randomly a “window” formed by two rows and two columns, Step 3) Remove and from the selectable offspring and
and interchanges the bits inside the window between the par- repeat the process for the next solution .
ents. The better solution must transfer more information (bits)
This procedure is described in more detail in [48].
to the descendant [34].
2 Points Crossover: It is a particular case of the window
F. Convergence Criterion
crossover. It selects randomly two columns, and the parents in-
terchange the bits between the columns. If a fixed number of generations is reached, the algorithm
Daily Crossover: this specialized operator takes advan- stops, else it goes back to stage . The maximum number of
tage that hourly demand has a similar behavior for different generations depends on the size of the system (number of units).
weekdays. When the scheduling is being achieved for a week,
this operator interchanges 24-h blocks between parents to create V. CASE RESULTS
an offspring. It is a particular case of the two points crossover. The algorithm was programmed using MATLAB 5.3, and the
2) Mutation Operators: They are applied to avoid prema- simulations were performed using a 1-GHz Athlon processor.
ture convergence of the algorithm, and are achieved over the For tuning the parameters of the GA, it was previously tested
created descendants. Two mutation operators has been used in using a purely thermal test system. After, the model was tested
this paper, performed with probabilities and , respec- for a hydrothermal system.
tively.
Standard Mutation: it randomly changes a bit of the ma- A. Test Results for Purely Thermal Systems
trix.
Swap Mutation: this operator selects arbitrary an hour , The generation scheduling for a purely thermal system (or
and search for the most expensive unit that is ON and the most unit commitment problem) is a particular case of the short-term
cheaper unit that is OFF. Then, with probability 0.7, unit HGSP. The simulations were performed over a 24-h demand
is turned OFF while unit is turned ON [34]. schedule for 10, 20, and 40 thermal units systems, which are
3) Repair Operators: The offspring creation process often described in [25]. Probabilities for the GA were set to
produces unfeasible solutions due to violations of restrictions per bit and .
described in (4) and (6). To avoid the creation of too many un- Results were compared with test results reported in [23], [25],
feasible solutions, two repair operators have been included: and [34], as seen in Table II, where the cost for the better, the av-
Repairing of Minimum Up/Downtime Constraints: This erage, and the worst solution over ten runs are shown. It can be
operator goes across each one of the vectors evaluating the seen that the results from the simulations are competitive with
consecutive time that a thermal unit has been up or down. If previously reported results. The convergence process is shown
a minimum up or down time constraint is violated for a given in Fig. 7, with the average, over ten runs, of the minimum pop-
hour, the state of the unit at the hour is changed. An analysis of ulation cost, normalized by the minimum cost known for the
this operator and its benefits can be seen in [34]. system.
Repairing of Storage Capacity Constraints: This operator
tracks each submatrix , decodes it, and recursively calculates B. Test Results for a Hydrothermal System
the water volume for each hour using (5). If at a given hour the The hydrothermal test system is a reduced version of the
constraint is violated, the operator randomly changes a bit of Chilean Central Interconnected System (see Appendix). It con-
for that hour until the violation is fixed. sists of six water reservoirs (11 hydro units, any of them hy-
If more feasible solutions are created on each generation, draulically coupled) and ten thermal units.
the process of replacement of the population members becomes Probabilities for the GA were set to
more competitive, and the exploration of more zones of the per bit and .

Authorized licensed use limited to: Universidad de La Salle. Downloaded on December 14,2020 at 09:19:24 UTC from IEEE Xplore. Restrictions apply.
GIL et al.: SHORT-TERM HYDROTHERMAL GENERATION SCHEDULING MODEL USING A GENETIC ALGORITHM 1261

TABLE II
TEST RESULTS FOR PURELY THERMAL SYSTEMS

Fig. 8. Convergence process for hydrothermal system.

Fig. 9. Total hydro and thermal hourly generation scheduling for a week.

VI. CONCLUSION

This paper proposes and develops a new model for dealing

with the short-term HGSP, incorporating, as a whole, three prob-
lems traditionally analyzed separately: short-term HCP, UCP,
and ELDP.
Hydrothermal systems are coupled in time. In order to un-
couple the long/mid-term models from the short-term model,
Fig. 7. Convergence process for a purely thermal system. FCFs have been used. In this way, FCFs work as the link be-
tween the short and the mid/long-term models.
The simulation converged (Fig. 8) to a population of feasible The definition of the decision variables, the representation of
solutions. From the analysis of the matrix for the best so- candidate solutions, and the fitness evaluation are the basis for
lution, it could be observed that the two cheaper thermal units implementing a GA. They act like the connection between the
were ON for the entire scheduling period, while the most expen- electric/economic model and the GA. Once these aspects are
sive were only turned on to satisfy demand peaks. Also, running solved, the solution through GAs is fundamentally a program-
thermal units were operating near their respective maximum ef- ming problem.
ficiencies. Promising results obtained from the computational simula-
As seen in Fig. 9, total thermal generation is flattened by the tion have been presented. The proposed GA, using new spe-
effect of the hydro generation. In this way, hydro generation cialized operators, have demonstrated excellent performance in
displaces the most expensive thermal generation. Besides, it can dealing with this kind of problem, obtaining near-optimal solu-
be observed the similar behavior of generation for different days tions in reasonable times and without sacrificing the realism of
of the week, mainly due to the effect of the “daily crossover.” the electric and economic models.

Authorized licensed use limited to: Universidad de La Salle. Downloaded on December 14,2020 at 09:19:24 UTC from IEEE Xplore. Restrictions apply.
1262 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 18, NO. 4, NOVEMBER 2003

TABLE V
FUTURE COST FUNCTION FOR RESERVOIR 1

TABLE VI
THERMAL UNITS CHARACTERISTICS

Fig. 10. Hydraulic configuration of hydro units in Chilean system.

TABLE III
INPUT/OUTPUT CHARACTERISTICS FOR HYDRO UNITS

TABLE IV
RESERVOIRS CHARACTERISTICS

TABLE VII
HOURLY DEMAND FOR A WEEKDAY

the six most important hydraulic reservoirs (and their associated

hydraulic systems) and a reduced set of representative thermal
units were incorporated.
The hydraulic configuration of the hydro units in the test
system is shown in Fig. 10. It can be seen that three of the
hydro units are independent, but the rest are hydraulically cou-
pled. Except for a 2-h time lag between units 7 and 8, no time
APPENDIX
lags were considered. Input/output characteristics (water dis-
HYDROTHERMAL TEST SYSTEM DESCRIPTION
charge/power) for hydro units are given in Tables III and IV.
The hydrothermal test system is a reduced version of the Also, curves for modeling each of the six water reservoirs can
Chilean Central Interconnected System, where a real model of be obtained from Table IV.

Authorized licensed use limited to: Universidad de La Salle. Downloaded on December 14,2020 at 09:19:24 UTC from IEEE Xplore. Restrictions apply.
GIL et al.: SHORT-TERM HYDROTHERMAL GENERATION SCHEDULING MODEL USING A GENETIC ALGORITHM 1263

The FCF for reservoir 1 is indicated in Table V. The FCF for [20] H. K. Youssef and K. M. El-Naggar, “Genetic based algorithm for secu-
the others reservoirs can be calculated using the respective rity constrained power system economic dispatch,” Electric Power Syst.
Res., vol. 53, no. 1, pp. 47–51, Jan. 2000.
values given in Table IV. [21] T. Jayabarathi, G. Sadasivam, and V. Ramachandram, “Evolutionary
Parameters for the quadratic cost functions for each thermal programming-based multiarea economic dispatch with tie line con-
unit (with ), along with their technical straints,” Electric Mach. Power Syst., vol. 28, no. 12, pp. 1165–1176,
Dec. 2000.
limits, are summarized in Table VI. [22] Y.-G. Wu, C.-Y. Ho, and D.-Y. Wang, “A diploid genetic approach to
Hourly demand for a weekday is given at Table VII. For Sat- short-term scheduling of hydrothermal system,” IEEE Trans. Power
urday and Sunday, 80 and 70% of a weekday demand have been Syst., vol. 15, pp. 1268–1274, Nov. 2000.
[23] C.-P. Cheng, C.-W. Liu, and C. Liu, “Unit commitment by Lagrangian
used, respectively. relaxation and genetic algorithms,” IEEE Trans. Power Syst., vol. 15, pp.
707–714, May 2000.
[24] C. W. Richter Jr. and G. B. Sheblé, “A profit-based unit commitment
ACKNOWLEDGMENT GA for the competitive environment,” IEEE Trans. Power Syst., vol. 15,
pp. 715–721, May 2000.
The authors acknowledge the helpful comments received [25] S. A. Kazarlis, A. G. Bakirtzis, and V. Petridis, “A genetic algorithm
solution to the unit commitment problem,” IEEE Trans. Power Syst.,
from four anonymous reviewers. Thanks to Universidad Téc- vol. 11, pp. 83–91, Feb. 1996.
nica Federico Santa María. [26] K. P. Wong and Y. W. Wong, “Combined genetic algorithm/simulated
annealing/fuzzy set approach to short-term generation scheduling with
take-or-pay contract,” IEEE Trans. Power Syst., vol. 11, pp. 128–136,
REFERENCES Feb. 1996.
[27] T. T. Maifeld and G. B. Sheble, “Genetic-based unit commitment algo-
[1] A. J. Wood and B. F. Wollenberg, Power Generation, Operation and rithm,” IEEE Trans. Power Syst., vol. 11, pp. 1359–1370, Aug. 1996.
Control. New York: Wiley, 1996. [28] S. O. Orero and M. R. Irving, “A genetic algorithm for generator sched-
[2] G. S. Christensen and S. A. Soliman, Optimal Long-Term Operation of uling in power systems,” Int. J. Elect. Power Energy Syst., vol. 18, no.
Electric Power Systems. New York: Plenum, 1988. 1, pp. 19–26, Jan. 1996.
[3] H. G. Stoll, Least-Cost Electric Utility Planning. New York: Wiley, [29] , “A genetic algorithm modeling framework and solution technique
1989. for short term optimal hydrothermal scheduling,” IEEE Trans. Power
[4] D. E. Goldberg, Genetic Algorithms in Search, Optimization, and Ma- Syst., vol. 13, pp. 501–514, May 1998.
chine Learning. Reading, MA: Addison-Wesley, 1989. [30] A. H. Mantawy, Y. L. Abdel-Magid, and S. Z. Selim, “Integrating ge-
[5] R. Dimeo and K. Y. Lee, “Boiler-turbine control system design using netic algorithms, tabu search, and simulated annealing for the unit com-
a genetic algorithm,” IEEE Trans. Energy Conversion, vol. 10, pp. mitment problem,” IEEE Trans. Power Syst., vol. 14, pp. 829–836, Feb.
752–759, Dec. 1995. 1999.
[6] R. A. F. Saleh and H. R. Bolton, “Genetic algorithm-aided design of [31] D. Dasgupta and D. R. McGregor, “Short term unit commitment using
a fuzzy logic stabilizer for a superconducting generator,” IEEE Trans. genetic algorithms,” Proc. 5th IEEE Int. Conf. Tools with Artificial In-
Power Syst., vol. 15, pp. 1329–1335, Nov. 2000. tell., Nov. 1993.
[7] T. Maifeld and G. Sheblé, “Short-term load forecasting by a neural net- [32] D. Dasgupta, “Unit commitment in thermal power generation using ge-
work and a refined genetic algorithm,” Electric Power Syst. Res., vol. netic algorithms,” presented at the Proc. 6th Int. Conf. Ind. & Eng. Ap-
31, no. 3, pp. 147–152, Dec. 1994. plicat. Artificial Intell. Expert Syst. (IEA/AIE-93), Scotland, U.K., June
[8] N. Li, Y. Xu, and H. Chen, “FACTS-based power flow control in 1993.
interconnected power system,” IEEE Trans. Power Syst., vol. 15, pp. [33] A. A. El Desouky and M. M. Elkateb, “A hybrid artificial intelligence
257–262, Feb. 2000. and heuristic method to short term generation scheduling,” in Proc. Int.
[9] T. S. Chung and Y. Z. Li, “A hybrid GA approach for OPF with consid- Assoc. Sci. Technol. Develop., Marbella, Spain, Sept. 2000, pp. 147–152.
eration of FACTS devices,” IEEE Power Eng. Rev., vol. 20, pp. 54–57, [34] J. Valenzuela and A. E. Smith, “A seeded memetic algorithm for large
Aug. 2000. unit commitment problems,” J. Heuristics, vol. 8, no. 2, pp. 173–195,
[10] L. L. Lai and J. T. Ma, “Genetic algorithms and UPFC for power flow 2002.
control,” in Int. J. Eng. Intell. Syst., U.K.: CRL Publishing Ltd., Dec. [35] V. Pereira and L. M. V. G. Pinto, “Application of decomposition tech-
1996, vol. 4, pp. 237–242. niques to the mid-and short-term scheduling of hydrothermal systems,”
[11] S. Gerbex, R. Cherkaoui, and A. J. Germond, “Optimal location of IEEE Trans. Power App. Syst., vol. PAS-102, pp. 3611–3618, Nov. 1983.
multi-type FACTS devices in a power system by means of genetic [36] M. V. Pereira, N. Campodónico, and R. Kelman, “Long-term hydro
algorithms,” IEEE Trans. Power Syst., vol. 16, pp. 537–544, Aug. 2001. scheduling based on stochastic models,” presented at the EPSOM,
[12] V. Miranda, J. V. Ranito, and L. M. Proença, “Genetic algorithm in Zurich, Switzerland, Sept. 1998.
optimal multistage distribution network planning,” IEEE Trans. Power [37] R. Kelman and M. V. Pereira, “Application of economic theory in power
Syst., vol. 9, pp. 1927–1933, Nov. 1994. system analysis: Strategic pricing in hydrothermal systems,” presented
[13] H. Rudnick, R. Palma, E. Cura, and C. Silva, “Economically adapted at the VI Symp. Specialists in Electric Oper. Expansion Planning, May
transmission systems in open access schemes: Application of genetic al- 1998.
gorithm,” IEEE Trans. Power Syst., vol. 11, pp. 1427–1440, Aug. 1996. [38] M. V. Pereira, N. Campodónico, and R. Kelman, “Application of
[14] R. A. Gallego, A. J. Monticelli, and R. Romero, “Comparative studies Stochastic Dual DP and Extensions to Hydrothermal Scheduling,”
on nonconvex optimization methods for transmission network expansion Online Rep., http://www.psr-inc.com.br/reports.asp, PSRI Technical
planning,” IEEE Trans. Power Syst., vol. 13, pp. 822–828, Aug. 1998. Rep. 012/99.
[15] K. Iba, “Reactive power optimization by genetic algorithm,” IEEE [39] M. V. Pereira, L. A. Barroso, and R. Kelman, “Market power issues
Trans. Power Syst., vol. 9, pp. 685–692, May 1994. in Bid-based hydrothermal dispatch,” presented at the IEEE Summer
[16] K. Y. Lee, X. Bai, and Y.-M. Park, “Optimization method for reactive Meeting, 2000.
power planning by using a modified simple genetic algorithm,” IEEE [40] H. Habibollahzadeh and J. A. Bubenko, “Application of decomposi-
Trans. Power Syst., vol. 10, pp. 1843–1850, Nov. 1995. tion techniques to short-term operation planning of hydrothermal power
[17] K. Y. Lee and F. F. Yang, “Optimal reactive power planning using evolu- system,” IEEE Trans. Power Syst., vol. PWRS-1, pp. 41–47, Feb. 1986.
tionary algorithms: A comparative study for evolutionary programming, [41] J. A. Muckstadt and R. C. Wilson, “An application of mixed-integer
evolutionary strategy, genetic algorithm, and linear programming,” programming duality to scheduling thermal generating systems,” IEEE
IEEE Trans. Power Syst., vol. 13, pp. 101–108, Feb. 1998. Trans. Power App. Syst., vol. PAS-87, pp. 1968–1977, Dec. 1968.
[18] A. Ahmad and D. P. Kothari, “A practical model for generator mainte- [42] L. L. Garver, “Power generation scheduling by integer program-
nance scheduling with transmission constraints,” Electric Mach. Power ming—Development of theory,” AIEE Trans., Part III: Power App.
Syst., vol. 28, no. 6, pp. 501–513, June 2000. Syst., vol. 81, pp. 730–735, Feb. 1963.
[19] I. El-Amin, S. Duffuaa, and M. Abbas, “A Tabu search algorithm for [43] J. F. Bard, “Short term scheduling of thermal electric generators using
maintenance scheduling of generating units,” Electric Power Syst. Res., Lagrangian relaxation,” Oper. Res., vol. 36, no. 5, pp. 756–766, Aug.
vol. 54, no. 2, pp. 91–99, May 2000. 1988.

Authorized licensed use limited to: Universidad de La Salle. Downloaded on December 14,2020 at 09:19:24 UTC from IEEE Xplore. Restrictions apply.
1264 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 18, NO. 4, NOVEMBER 2003

[44] F. Zhuang and F. D. Galiana, “Toward a more vigorous and practical unit Julian Bustos (SM’94) was born in Santiago, Chile. He graduated as an elec-
commitment by Lagrangian relaxation,” IEEE Trans. Power Syst., vol. trical engineer from Universidad Técnica Federico Santa María (UTFSM), Val-
PWRS-3, pp. 763–770, May 1988. paraíso, Chile, and received the M.Sc. and Ph.D. degrees from the University of
[45] D. Whitley, “A Genetic Algorithm Tutorial,” Colorado State Univ., Tech. Pittsburgh, PA.
Rep. CS-93-103, Mar. 1993. Currently, he is a Professor at UTFSM and is Vice President for Academic
[46] D. Beasley, D. R. Bull, and R. R. Martin, “An overview of genetic algo- Activities. His research activities include the economic operation, planning, and
rithms: Part 1, fundamentals,” University Computing, vol. 15, no. 2, pp. analysis of electric power systems.
58–69, 1993.
[47] D. Beasley, D. R. Bull, and R. R. Martin, “An overview of genetic al-
gorithms: Part 2, research topics,” University Computing, vol. 15, no. 4,
pp. 170–181, 1993.
[48] W. Banzhaf, P. Nordin, R. E. Keller, and F. D. Francone, Genetic
Programming: An Introduction. San Mateo, CA: Morgan Kaufmann,
1998.
[49] E. Gil, “Programación de la Generación de Corto Plazo en Sistemas
Hidrotérmicos Usando Algoritmos Genéticos,” M.Sc. dissertation. Uni-
versidad Técnica Federico Santa María.

Esteban Gil (S’00) was born in Santiago, Chile. He received the B.Sc. and Hugh Rudnick (F’00) was born in Santiago, Chile. He graduated as an electrical
M.Sc. degrees in electrical engineering from Universidad Técnica Federico engineer from University of Chile, Santiago, Chile. He received the M.Sc. and
Santa María (UTFSM), Valparaíso, Chile, in 1997 and 2001, respectively. He Ph.D. degrees from the Victoria University of Manchester, Manchester, U.K.
is currently pursuing the Ph.D. degree at Iowa State University, Ames. Currently, he is a Professor at Catholic University of Chile, Santiago, Chile.
His research interests focus mainly on economics, optimization, operation, His research activities focus on the economic operation, planning, and regula-
and planning of electric power systems. tion of electric power systems.

Authorized licensed use limited to: Universidad de La Salle. Downloaded on December 14,2020 at 09:19:24 UTC from IEEE Xplore. Restrictions apply.