APPLICATION OF GENETIC
ALGORITHMS FOR OPTIMAL
RESERVOIR OPERATION
D. Nagesh Kumar, Civil Engg. Dept., Indian Institute of Technology, Kharagpur, India
Ashok Kumar, Civil Engg. Dept., Indian Institute of Technology, Kharagpur, India
K. Srinivasa Raju, Civil Engg. Dept., S.E.S.College of Engineering, Kopargaon, India
ABSTRACT
Genetic Algorithms (GAs) application in the field of water resources engineering is of recent
origin. Genetic Algorithms is one of the tools, which handles nonlinear optimization problems in an
efficient manner. Optimal reservoir operation of reservoir for hydropower production involves
constrained nonlinear optimization. The constrained problem is converted into unconstrained
problem by using penalty function method. Genetic algorithm was then used to optimize the
reservoir operation for hydropower production. This apporach was used to develop optimal
operating policy for Hirakud reservoir, a multipurpose project on river Mahanadi, India
(geographical location of the dam: Latitude 210 32 N, Longitude 830 52 E). Hydropower
production from the system is maximized with other demands as constraints. Various steps
involved in deriving the optimal operating policy for the reservoir using GA are discussed in the
paper. For fixing the GA parameters viz. Crossover probability and mutation probability, the model
is run for different values of crossover and mutation probabilities. For a crossover probability of
0.800 and a mutation probability 0.006, the model was found to perform well. After fixing the GA
parameters the model is run for various dependable levels of inflows. The operating policy thus
obtained can be used for optimal operation of the reservoir. Results from the model and possible
extensions were discussed.
KEYWORDS
Reservoir operation, Genetic Algorithms, Hydropower
INTRODUCTION
Application of genetic algorithms (GAs) in the field of water resources engineering is of recent
origin. Genetic Algorithms provided solutions as good as those obtained by other traditional
methods like Linear Programming, Non Linear programming and Dynamic Programming. For
more complicated problems, particularly discontinuous or highly nonlinear and nonconvex type,
genetic algorithm proved to be computationally superior to gradient based methods. McKinney and
Lin (1994) applied GAs in the Management of Ground Water models. Simpson et al, (1994, 1996)
used the GAs in optimization of pipe network and the results obtained compared well with those
obtained by other methods. Savic and Walters (1997) developed a computer model called GANET
for least cost design of water distribution networks. Reddy (1997) developed a nonlinear
optimization model based on genetic algorithms for land grading design of irregular fields. Oliveira
and Loucks (1997) derived multireservoir operating rules using real-valued vectors containing
information needed to define both system release and individual reservoir storage volume targets as
functions of total storage in each of the multiple within-year periods. In this paper genetic
algorithms were applied to determine the optimal reservoir releases for hydropower generation
from a multipurpose reservoir.
GENETIC ALGORITHMS
Genetic Algorithms (GAs) are based on the theory given by Darwin that fittest of the fit will
survive. They belong to a family of combinatorial optimization methods that search for solutions of
complex problems using an analogy between optimization and natural selection. They use a random
search procedure inspired by biological evolution, cross breeding trial designs and allowing only
the Fittest designs to survive and propagate to successive generations. GAs handle nonlinear
optimization problems in an efficient manner and they differ from other traditional methods in
number of ways (Goldberg, 1989) like (i) GAs work with a coding of parameter sets and not with
parameters themselves, thus allowing one to use a wide variety of parameters as decision variables,
(ii) GAs use objective function or fitness information only in contrast to traditional methods which
rely on existence and continuity of derivatives or other auxiliary information.
Working Principle of GAs
Any nonlinear optimization problem without constraints is solved using Genetic Algorithms
involving basically three tasks viz., Coding, Fitness evaluation and Genetic operation. First of all
decision variables are identified from the given optimization problem. These decision variables are
coded into some string like structures. For coding the decision variables binary coding is used. This
coded string is called Chromosome. The length of chromosome depends on the desired accuracy of
the solution. It is not necessary to code all the decision variables in equal substring length.
Generally, the fitness function is first derived from objective function and is used in successive
genetic operations. Genetic operators require that the fitness function should be nonnegative. If the
problem is of maximization, the fitness function is taken as directly proportional to the objective
function. The fitness function value of a string is known as the strings fitness.
Once the fitness of each string is evaluated, the population is operated by three common operators
for creating new population of points. They are Reproduction, Crossover and Mutation.
Reproduction selects good strings in a population and forms a mating pool. In this paper Roulette
wheel simulation is used for the selection of good strings. In cross over operator, two strings are
picked from the mating pool at random and some portions of the strings are exchanged between the
strings. A single point cross over operation is performed by randomly choosing a crossing site
along the string and by exchanging all bits on the right side of the crossing site. The mutation
operator changes 1 to 0 and 0 to 1 with a small mutation probability, pm, within the string. Mutation
creates points in the neighborhood of the current point, which helps in local search around the
current solution. It is also used to maintain diversity in the population.
The newly created population using the above operators is further evaluated and tested for
termination. If the termination criterion is not met the population is iteratively operated by the
above three mentioned GA operators and evaluated. This process is continued until termination
criterion is met. One cycle of these operations and the subsequent evaluation procedure is known as
a generation.
Genetic Algorithms for constrained problems:
The constrained problem is converted into unconstrained problem by using penalty function
method. In this process, the solution falling out side the restricted solution region is considered at a
very high penalty. This penalty forces the solution to adjust itself in such a way that after some
generations it will fall into restricted solution space. In penalty function method a penalty term
corresponding to the constraint violation is added to the objective function. Generally bracket
operator penalty term is used
k
Fi = f (x )+ j (j)2
(1)
j =1
where Fi is fitness value, f (x ) is objective function value, k is total numbers of constraints, is -1
for maximization and +1 for minimization, j is penalty co-efficient and j is amount of violation.
Once the problem is converted into unconstrained problem, the rest of the procedure remains the
same.
DESCRIPTION OF THE CASE STUDY AREA
The Hirakud dam is a multipurpose project built across river Mahanadi in Orissa State, India. The
geographical location of this dam is Latitude 210 32 N, Longitude 830 52 E. Hirakud dam was
conceived primarily for flood control in Mahanadi delta with the other purposes of the dam being
irrigation and hydropower. The catchment area of the reservoir upto the dam site is 83,400 sq. km.
The active storage capacity of the reservoir is 5,375 million m3 with the gross storage capacity
being 7,189 million m3. Total installed capacity for power generation is 307.5 MW. Area of
irrigation during the first crop season, Kharif, is 1,556.5 sq.km. and during the second crop season,
Rabi, is 1,084 sq.km.
HYDROPOWER OPTIMIZATION FORMULATION
The objective for optimization problem adopted is to maximize the hydropower generated from the
reservoir releases for power (RP) with the other demands from the reservoir as constraints. If RP is
expressed in million cubic meters (M m3) per month and head causing the flow, h in meters, then
power produced P in kilowatt hours for a 30 day month is given by P = 2725 RP h. The objective
is to maximize total hydropower produced in a year. As can be seen this objective involves
nonlinear optimization. For the demonstration of applicability of GAs for the optimization problem
a courser time interval of one month is chosen which can be further reduced to weekly or daily.
Thus the objective for hydropower optimization is
Maximize
Z =
12
t=1
Pt
(2)
This objective function is subject to the following constraints.
Releases for Power and Turbine Capacity Constraints
The releases into turbines for hydropower production should be less than or equal to the flow
corresponding to the maximum capacity of the turbine. Also the power production in each month
should be greater than or equal to the firm power.
RPt TC
t = 1,2 ,.............12
(3)
t = 1,2 ,.............12
(4)
RPt
FPt
where RPt is release for power in the period t, TC is turbine capacity and FPt is firm power for the
period t.
Irrigation Demand Constraints
The releases for irrigation should be greater than or equal to the minimum irrigation demand to
sustain the crops and also at the same time this should not exceed the maximum irrigation demand
to produce the targeted yield.
RI t
IDMIN t
t = 1,2 ,.............12
(5)
RI t
IDMAX t
t = 1,2 ,.............12
(6)
where RI t is release for irrigation in the period t, IDMIN t is the minimum irrigation demand to
sustain the crops and IDMAX t is the maximum irrigation demand to produce the targeted yield for the
period t.
Reservoir Storage Continuity Constraints
If the evaporation losses are expressed as a function of storage, storage continuity equation is given
by (Loucks et al., 1984) This constraint involves releases for power, releases for irrigation,
overflows, reservoir storage, inflows and the losses through the reservoir during the period t for all
months expressed in volume units.
(1 + at )St +1 = (1 at )St + Qt RI t RPt OVFt Aoet
(7)
where St is storage at the beginning of the period t, Qt is inflow during the period t, OVFt is
overflow for the period t (if any), Ao reservoir water surface area corresponding to the
dead storage volume, et is evaporation rate for that period in depth units, at = 0.5 Aa et and Aa is
the reservoir water spread area per unit volume of active storage.
Reservoir Storage Capacity Constraints
The live storage in the reservoir during the period t should be less than or equal to the maximum
active storage capacity (Smax) of the reservoir.
St
Smax
t = 1,2 ,.............12
(8)
The above optimization model (Equations 3 through 8) is solved using genetic algorithms as
explained in the following steps.
1. Identification of decision variable: Here the decision variable is, release for power in each
month. So there are twelve decision variables.
2. Fixation of upper and lower bound: For fixing the upper and lower bound of the decision
variable, the two constraints given in eqs. 3 and 4 are considered. The lower bound, is the firm
power and the upper bound is the capacity of the turbines.
3. Fixation of binary string length: Based on the difference between the upper and lower bound
of the decision variables the length of the binary string is fixed.
4. Coding of string: Binary coding of the string is done by generating random numbers.
5. Decoding of decision variable: The coded string is decoded by using linear mapping rule.
6. Calculation of effective head: Effective head for hydropower generation is calculated using
storage continuity equation and elevation-storage relation.
7. Calculation of fitness: The values of the decision variable and the effective head are
substituted into fitness function to evaluate fitness of each string.
8. G A operations: All the three steps involved in GA operation viz., Reproduction, Crossover
and Mutation are performed on the strings.
Step 6 to 8 are repeated until termination criterion is met.
Lower and upper bounds for the decision variable and the string length for each month are given in
Table 1. Average inflows into the reservoir and the average irrigation demands are presented in
Table 2. Most of the inflows to the reservoir occur during monsoon months ie., July to October.
TABLE 1. Lower, upper bound and string length for the decision variable
Month
January
Decision
Variable
Lower bound
Upper bound
String length
616.50
2000
10
February
RP1
RP2
370.00
2000
15
March
RP3
370.00
2000
15
April
RP4
245.00
2000
15
May
RP5
370.00
2000
15
June
RP6
370.00
2000
15
July
RP7
615.00
2000
10
August
RP8
1233.00
2000
10
September
RP9
1110.00
2000
10
October
RP10
615.00
2000
10
November
RP11
615.00
2000
10
December
RP12
615.00
2000
10
TABLE 2. Average inflows and Irrigation Demands for different months
Inflows
M.cu.m.
Irrigation Demand
M.cu.m
January
397.026
229.831
February
178.785
268.671
March
33.291
323.046
April
46.854
313.428
May
6.165
57.334
June
431.550
107.518
July
4,151.511
249.683
18,377.865
243.271
September
5,611.383
265.588
October
1,392.057
295.304
November
626.364
45.005
December
556.083
117.135
Month
August
RESULTS AND CONCLUSIONS
For fixing the GA parameters viz. crossover probability and mutation probability, the model is run
for different values of crossover and mutation probabilities. Two values for crossover probability,
0.80 & 0.85 and three values for mutation probability, 0.005, 0.006 & 0.007 are chosen. Results
obtained are compared in terms of total hydropower produced, fitness and number of generations
(Ashok, 1999). From these comparisons it is concluded that for crossover probability 0.800 and
mutation probability 0.006, the hydropower produced is maximum and the solution converged at
moderate number of generations. For values of mutation probability other than 0.006 the solution
converges very rapidly which is not desirable. So the GA parameters are fixed as crossover
probability of 0.800 and mutation probability of 0.006 for the case study made.
Once the GA parameters are fixed the model is run for four different levels of inflows viz., 40% &
20% below average inflows, average inflows and 20% above average inflows. Optimized monthly
releases for hydropower (in million cubic meters, M.cu.m) are shown for different levels in
Figure1.
From these results, reservoir can be operated for optimal hydropower generation for different
expected levels of inflows into the reservoir after meeting the other demands from the reservoir.
Efforts are on to develop operating policy for much smaller time intervals.
From this study it can be concluded that Genetic Algorithms have very strong potential for
application in water resources optimization. In this paper GAs were successfully applied for
optimal hydropower generation from a multipurpose reservoir and is demonstrated through a case
study of an existing multipurpose reservoir in Orissa state, India.
REFERENCES
1. Ashok, K. (1999) Application of Genetic Algorithms for Optimal Reservoir Operation. M.Tech
thesis, IIT, Kharagpur, India.
2. Deb Kalyanmoy (1995) Optimization for Engineering Design. Prentice Hall of India.
3. Dowsland K.A. (1996) Genetic Algorithms - a Tool for OR. J. of the Operational Research
Society, 47, pp. 550 561.
4. Goldberg D.E. (1989) Genetic Algorithms in Search, Optimization and Machine Learning.
Addison-Wesley Publishing Company, Reading, Mass.
5. Loucks, D.P., J. R. Stedinger and D. A. Haith (1981) Water Resources Systems Planning and
Analysis. Prentice Hall.
6. Oliveira R and D.P. Loucks (1997) Operating rules for multireservoir systems. Water Resources
Research, 33(4), pp. 839-852.
7. Mc Kinney D. C. and Lin Min-Der (1994) Genetic Algorithm solution of Groundwater
Management Models. Water Resources Research, 30(6), pp. 1897-1906.
8. Reddy L.S. (1996) Optimal Land Grading Based on Genetic Algorithms. J. of Irrigation and
Drainage Engineering ASCE, 122(4), pp. 183-188.
9. Savic D.A. and G.A. Walters (1997) Genetic Algorithms for Least-Cost Design of Water
Distribution Networks. J. of Water Resources Planning and Management ASCE, 123(2), pp.
67-77.
10. Simpson R.A., D.C. Graeme and M.J. Laurence (1994) Genetic Algorithms Compared to Other
Techniques for Pipe Optimization. J. of Water Resources Planning and Management ASCE,
120(4), pp. 423-443.
11. Simpson R.A., D.C. Graeme and M.J. Laurence (1996) An improved Genetic Algorithm for
pipe network Optimization. Water Resources Research, 32(2), pp. 449458.
