RESEARCH PAPER
International Journal of Recent Trends in Engineering, Vol. 1, No. 6, May 2009
Optimal Operation Scheduling of Irrigation
Canals Using Genetic Algorithm
Y.P. Mathur1, Gunwant Sharma2, A.W.Pawde3
Department of Civil Engineering, Malaviya National Institute of Technology, Jaipur, 302017 India
Email: 1yogeshmathur @yahoo.co.in, 2 gunwantsharma @yahoo.com, 3awpawde@gmail.com (corresponding author)
Abstract— Canal scheduling is an important activity that
significantly influences production of crops compared to
other aspects of agriculture. Irrigation canal scheduling is
the activity of preparing an optimal schedule of outlets on
supply canal as per need of user, subject to canal system
constraints. In this paper, the problem of optimal
operational scheduling of irrigation canal with provision to
open some outlet at specific time slot as per request of user
using Genetic Algorithm is presented. The optimal
operational schedule obtained with Genetic Algorithm
approach is compared with previously published
operational schedule for the Famen secondary canal of
Feng-Jia-Shan Irrigation District, China, obtained using
integer programming. The result shows that Genetic
Algorithm approach gives sufficient flexibility in decision
making regarding the group formation of various outlets. It
is an efficient tool for irrigation manager to prepare optimal
schedule of irrigation canals as per time slot demanded by
users.
Index Terms— canal, canal scheduling, genetic
algorithm, irrigation, integer programming
I. INTRODUCTION
The irrigation canal scheduling problem is the problem
of preparing an operation schedule of outlets on the
distributary canal subject to certain constraints. In
irrigation project water distribution i.e. conveyance of
water from the source to the field head is the most
important link between the supplier of water and farmers
who actually put the water to use. If the water distribution
is well operated, high returns can be obtained from
irrigation project. There are two types of water
distribution methods. (i) flexible method and (ii) rigid
method. In flexible method, water is supplied on demand
of user and may be continuous during entire base period
of crops. In case of rigid method, supply is fixed at
constant frequency with constant or varied amount.
Rotational water distribution in irrigation project is a
common practice throughout the world, which fall under
category of rigid method. The supply is rotational among
individual users in an outlet command area, among
outlets on a distributary canal and among different
distributaries on main canal. Frequency of irrigation in
command area is based on the soil-crop-climate
conditions. The operational schedule of rotation delivery
for distributaries is prepared considering its capacity,
number of outlets on distributary, their duration of supply
and rotation period. In most of the irrigation projects,
cropping pattern is changed with time due to which the
outlets discharge is changed and there is a need to re-
11
© 2009 ACADEMY PUBLISHER
schedule the operation of distributary before each
rotation. Improper water delivery schedules cause
shortage or surplus of water to users which affects
adversely on performance of irrigation projects. Optimal
scheduling of irrigation canal has the potential to increase
the performance of many irrigation projects that are
operated on the basis of rigid delivery schedule such as
rotational delivery of irrigation water (Chamber, 1988).
This paper describes the application of genetic algorithm
(GA) to water delivery of outlets on distributary canal.
The GA model is applied on the secondary canal of Feng-
Jia-Shan, China irrigation project and optimal operation
schedule is presented.
II. PREVIOUS WORKS
Suryavanshi and Reddy (1986) formulated, the outlets
scheduling problem as a mathematical problem and
applied an integer programming model for preparing
optimal operation schedule for irrigation canal outlets.
They considered the supply canal to consist of a number
sub-channels. The discharge of each sub-channel is equal
to that of the outlet along the supply canal. Each outlet is
free to draw water from any sub-channels during rotation
time. A sub-channel cannot supply two or more outlets
simultaneously, but it can supply several outlets
sequentially. Each outlet remains open continuously for a
known duration. The binary decision variable Xij=1, if
particular sub-channel i supplies to a particular outlet j,
else Xij=0. Their model minimizes the total number of
stream tubes used. This was the first mathematical mode
introduced for canal scheduling problems.
Wang et al. (1995) identified a minor shortcoming in
the formulation of objective function by Suryavanshi and
Reddy (1986), which is that, the model minimizes the
sum of all Xij and not the total number of stream tubes
operating simultaneously. They proposed the sequence of
outlets be within a certain group should be decided by the
operators according to specific priorities. They developed
an improved model by introducing dummy variable
called the ‘activation function’. For a given sub-channel,
the activation function assumes a value of one if that sub-
channel is used irrespective of whether the sub-channel
feeds one or several outlets else the activation function
takes a value of zero. The objective function in this model
is to minimize the sum of the activation function.
RESEARCH PAPER
International Journal of Recent Trends in Engineering, Vol. 1, No. 6, May 2009
Reddy et al. (1999) introduced the concept of starting
time block to formulate the problem of canal scheduling
of lateral canals. To control the number of variables in the
problem, a constraint of starting time range to each outlet
was introduced which compels every outlet to start within
a pre-specified time window. The model minimizes the
difference between required capacity and actual capacity
of the supply canal. They demonstrated their model with
practical example.
Anwar and Clarke (2001) developed the model using a
mixed integer programming approach for scheduling
canal irrigation among a group of users where the
duration of flow of each outlet and a target start time is
specified by the user. The objective was to schedule
supplies as close as possible to the water demand
durations requested by users. They developed the single-
period and multiperiod models.
Wardlaw and Bhaktikul (2004) applied the GA
approach to canal scheduling using the concept of 0-1
linear programming approach developed by Suryavanshi
and Reddy (1986) and Reddy et al. (1999) and
demonstrated its efficiency with an example of the Xi Le
submain of the Hetao irrigation project, which is situated
in the Inner Mongolia region of China.
Monem and Namdarian (2005) applied simulated
annealing (SA) technique for optimal water distribution
in irrigation canals. They developed the multi-objective
model called as OWDIS (Optimal Water Distribution in
Irrigation System). One of the objectives of this model is
to minimize the operational stages of canal inlet. The
model was applied on the AMX canal of the Varamin
irrigation network located in south-east of Tehran, Iran.
They also compared their model results with Wang et al.
(1995).
III. GENETIC ALGORITHMS
Genetic algorithm (GA) is heuristic technique for
searching over the solution space of a given problem in
an attempt to find the best solution or solutions (Forrest,
1993). The use of Genetic algorithm was first proposed in
1975, by Holland (1992), who based his research on
Darvin’s principle of evolution. Inspired by the natural
mechanisms of selection and reproduction, Holland
created an ‘intelligent’ form of a random search that
explore the solution more promising region and search
for solution more intensely in those more promising
regions. Since then, GAs has been applied to a variety of
problem. The GAs technique has its roots in the
biological processes of ‘survival of the fittest’ and
adaptation. Genetic algorithms are applicable to a variety
of optimization problems that are not well suited for
standard optimization algorithms, including problems in
which the objective function is discontinuous, non-
differentiable, stochastic, or highly nonlinear. GA
manipulates a string of numbers in a manner similar to
how chromosomes are changed in biological evolution.
An initial population made up of strings of numbers may
be chosen at random or may be specified. Each number
12
© 2009 ACADEMY PUBLISHER
slot is called a gene and an individual string of numbers is
called a chromosome. A set of chromosomes forms a
population. Each chromosome represents a given number
of traits which are the actual parameters that are being
varied to optimize the fitness function. The fitness
function is a performance index that we seek to maximize
or minimize. The GA search starts from a population of
many points rather than starting from one point as is the
case in standard optimization algorithm. This parallelism
means that the search will not become trapped on local
optima. Genetic algorithms manipulate decision or
control variable representations at a string level to exploit
similarities among high-performance strings. This also
helps GA based methodology to tackle functions with
discontinuity. Performance evaluation of candidate
solutions from the population is performed using the
fitness function. The fitness function is evaluated using
information of objective function. This avoids the
necessity of using derivative information. The algorithm
begins with random population of individuals in which
each individual is represented by a binary or real coded
string (chromosome) for one possible solution. Each of
the individuals is assigned a fitness using the objective
function combined with constraints, if any. The optimal
solution is obtained after repetition of generation cycle.
The operations used in GA are: (A) reproduction or
selection, (B) cross-over, and (C) mutation.
A. Reproduction or Selection
During each successive generation, a proportion of the
existing population is selected to breed a new generation.
The fittest strings have the highest probability of being
used in reproduction. There are a number of approaches
to selection like tournament and roulette wheel. All these
approaches control the probability of selection of
chromosome for participation in next generation of
population.
B. Crossover
Crossover is used to vary the programming of a
chromosome from one generation to the next. This
operator of Gas preserves the best genetic material from
two parent strings. The crossover is not usually applied to
all pairs of individuals selected for mating. A random
choice is made, where the probability of crossover being
applied is in the range 0.6 and 0.9. Goldberg (1989) and
Michalewicz (1992), has outlined the various types of
crossovers viz. one point crossover, two-point crossover
and uniform crossover.
C. Mutation Operator
Mutation is used to maintain genetic diversity from
one generation of a population of chromosomes to the
next. The random mutation is controlled by specifying
the probability of mutation. In mutation process, the bits
are changed ‘0’ to ‘1’ and ‘0’ to ‘1’ at the random chosen
position of randomly selected strings. Michalewicz
(1992) has explained uniform and non uniform for real-
value representation, which are two basic approaches to
mutation. The purpose of mutation in GAs is to allow the
algorithm to avoid local minima by preventing the
RESEARCH PAPER
International Journal of Recent Trends in Engineering, Vol. 1, No. 6, May 2009

population of chromosomes from becoming too similar to exceeded has very high value of fitness function and as
each other, thus slowing or even stopping such are highly improbable to be selected for inclusion in
evolution.Following are some advantages of GAs the reproduction process and participate in next
compared with other classical optimization methods. generation.
x GAs work with numerical values and there is no Key algorithm parameters for formulation of GA are:
requirement of objective function to be a linear or the size of initial population, probability of mutation,
differentiable function and it can solve problems probability of crossover and no. of generations. Goldberg
with discrete decision variables. (1989) and Michalewicz (1992) have suggested values for
x GAs generates several near optimal solution and probability of mutation and probability of crossover. The
they can provide sufficient flexibility in decision values set for these parameters for obtaining the results of
making. problem considered are shown in Table I.
x GAs usually does not get trapped to local optimal
TABLE I
solution and they can find a near optimal solution
with a high probability. PARAMETERS OF GENETIC ALGORITHM
Parameter for GA Values selected
IV. OBJECTIVE FUNCTION FOR CANAL SCHEDULING Population size 100
Probability of mutation 0.05
The formulation for canal scheduling given by Probability of crossover 0.8
Wardlaw and Bhaktikul, (2004) using GA is Number of generation 100

V. APPLICATION TO REAL FIELD PROBLEM

2
n ­ª m
º ½ Wang Zhi et al. (1995) demonstrated the application of

MinimizeZ ¦®«Q  ¦ q
ONOFF »
j ij
 Pi ¾ (1) their formulation of the IP using an example of the
i 1 ¯¬ j 1 ¼ ¿ secondary canal No.11 of North main canal; Feng-Jia-
Shah Irrigation District, China. It is a large canal with a
maximum discharge capacity of 2.8 m3/s and a total
Where, Q =supply canal capacity (m3/s) command area of 3930 ha. Administratively, the
qj = capacity of lateral j (m3/s) secondary canal No. 11 is managed by two sectors: the
ONOFFij = 1 if outlet canal operates in time period i Famen sector which is responsible for the upper reach
else it is zero (2000 ha), and the Chenguan sector for the lower reach
m = number of outlets (1930 ha). During the summer irrigation of 1987, the
n = number of time blocks in the rotation period and Famen sector was allowed to run for a maximum of 14
Pi = penalty function for canal capacity constraint days with a maximum flow rate of 1.9 m3/s. There are 26
The value of ONOFFij is determined from the starting outlets in the Famen reach and each outlet has a discharge
time (Sj) of outlet j as follows. capacity of 0.2m3/s. Number of time blocks in the
rotation period of 14 days for 24 hrs. a day are 336
ONOFFij = 1 If i > Sj & i < (Sj + Dj) (14x24) with each time block of 1 hr. According to the
gross field irrigation requirements and water conveyance
and efficiencies associated with each outlet, the running time
ONOFFij = 0 otherwise of each outlet given by Wang Zhi et al. (1995) is as
Where, Dj = operation duration for outlet j. shown in Table II. In this paper the example of above
Sj = starting time of the outlet j, which is a decision mentioned famen canal is taken for demonstration of use
variable of the problem. of GA for optimal scheduling operation.
Wardlaw and Bhaktikul, (2004) formulated the penalty
function as given below; VI. RESULT AND DISCUSSION
Wang et al. (1995) suggested that the operational
m sequence of outlet within a certain group should be
Pi ¦ max ¬ª q j* ONOFFij Q ,0 ¼º
j 1
Where, Pi is added to objective function. Many
constraint-handling techniques have been developed for
the GA optimization. A penalty function is usually used
for handling the nonlinear constraints. It is designed to
penalize the infeasible solutions to reduce their fitness.
The penalty function method converts the constrained
optimization model to an unconstrained one to suit the
application of evolutionary computing techniques. In
present formulation a penalty function ensures that
infeasible solutions in which the main canal capacity is
© 2009 ACADEMY PUBLISHER
(2) decided by operators on specific priorities. In this GA
formulation of optimization problem, starting time (Sj) of
the outlets is the decision variable, the lower and upper
bound of which are to be specified by irrigation manager
during execution of GA program. GA generates several
near optimal solutions which give sufficient flexibility in
decision making regarding the group formation of various
outlets. If it is required to start the specific outlets at any
time slot of rotation period, it can be achieved by
specifying their starting time within specified lower and
upper bound as per requirement of user.
13
 RESEARCH PAPER
International Journal of Recent Trends in Engineering, Vol. 1, No. 6, May 2009

While preparing the operational schedule of any canal,

many times irrigation manager has to consider the time
slot demand of user in the rotation period. We have 2

applied GA for three different assumed situations in 1.8

which certain canals outlet are opened at specified times 1.6

Inflow Rate (m3/s)

as per demand by users. In the first situation it is assumed 1.4

that upstream group of users i.e. outlet no. 1 to 8 have 1.2

1
demanded supply of water in initial period of rotation so,
0.8
initial sequence of outlets is arranged from the upstream
0.6
end of the canal by specifying the lower and upper bound
0.4
value of time slot as 1. In second situation canal outlet 0.2
no.1, 2, 3, 8, 9, 10, 19, 21 and 25 are assumed to operate 0

in first time slot of rotation period. In third situation 0 12 24 36 48 60 72 84 96 108 120 132 144 156 168 180 192 204 216 228 240 252 264 276 288 300 312 324 336

Flow Time (h)

outlet no.3, 8, 12, 16, 20, and 25 are specified to start in
time slot 195, 140, 242, 104, 72 and 172 respectively. Fig. 2. Computed inflow hydrograph into canal
The result of optimal starting time period of outlet using genetic algorithm (GA 1).
obtained by GA for first, secondly and third situation are
denoted by GA1, GA2 and GA3 respectively in table III.
The results show that irrigation manager can plan the
2
optimal schedule of canal which satisfies the time slot
1.8
demand of some users in rotation period. The inflow
1.6
hydrograph of Famen secondary canal for proposed
1.4
sequence of outlets arranged from the downstream end of
Infloe Rate (m3/s)

the canal using IP by Wang Zhi et al. (1995) is shown in 1.2

fig 1. Canal hydrograph using IP shown in fig.1 indicates 1

that headgate setting is needed at the end of the rotation 0.8

period to avoid the wastage of water. The computed 0.6

inflow hydrograph of the canal in the present paper, as 0.4

shown in fig.2, 3, and 4 indicate that a small head gate 0.2

regulation is required at intermediate time within rotation 0

0 12 24 36 48 60 72 84 96 108 120 132 144 156 168 180 192 204 216 228 240 252 264 276 288 300 312 324 336

period of canal to avoid wastage of water. Flow Time (h)

Fig. 3. Computed inflow hydrograph into canal

2
using genetic algorithm (GA 2)
1.8

1.6

1.4
Inflow Rate (m3/s)

1.2 2

1.8
1

1.6
0.8

1.4
0.6

1.2
0.4
1
0.2
0.8
0
0 12 24 36 48 60 72 84 96 108 120 132 144 156 168 180 192 204 216 228 240 252 264 276 288 300 312 324 336
0.6
Flow Time (h)
0.4

0.2
Fig. 1 Canal inflow hydrograph into canal using integer programming
(IP). WANG (1995) 0

0 12 24 36 48 60 72 84 96 108 120 132 144 156 168 180 192 204 216 228 240 252 264 276 288 300 312 324 336

Flow Yime (hr)

Fig. 4. Computed inflow hydrograph into

canal using genetic algorithm (GA 3).

14

© 2009 ACADEMY PUBLISHER

 RESEARCH PAPER
International Journal of Recent Trends in Engineering, Vol. 1, No. 6, May 2009

TABLE II
AREA AND RUNNING TIME FOR EACH OUTLET CANAL. WANG (1995)

Outtlet Area Time ‘tj’ Outlet Area Time ‘tj’ Outlet Area Time ‘tj’ Outlet Area Time ‘tj’
No.j (ha) (hrs) No.j (ha) (hrs) No.j (ha) (hrs) No.j (ha) (hrs)

1 83 138 8 74 102 15 213 281 22 36 49

2 34 47 9 63 98 16 26 36 23 109 171
3 102 132 10 28 40 17 183 264 24 36 51
4 73 98 11 61 89 18 35 47 25 8 11
5 34 48 12 53 94 19 24 33 26 48 76
6 97 161 13 105 155 20 171 264 -- -- --
7 46 65 14 53 75 21 201 333 -- -- --

TABLE III
OPTIMAL STARTING TIME PERIOD OF OUTLET OBTAINED BY GA AND GIVEN BY WANG (1995) USING IP.

Outlet Outlet Outlet

Start time (hrs) Start time (hrs) Start time (hrs)
No. j No. j No. j

GA1 GA2 GA3 IP GA1 GA2 GA3 IP GA1 GA2 GA3 IP

1 1 1 95 194 10 162 1 291 49 19 300 1 303 1
2 1 1 1 281 11 198 152 58 94 20 73 71 72 1
3 1 1 195 191 12 132 242 242 1 21 1 1 1 1
4 1 102 79 183 13 100 134 181 76 22 262 17 47 1
5 1 288 233 282 14 123 202 1 207 23 138 88 1 1
6 1 98 147 33 15 55 55 55 47 24 283 40 1 11
7 1 268 271 264 16 300 34 104 171 25 65 1 172 1
8 1 1 140 89 17 665 69 1 62 26 202 260 1 1
9 237 1 1 231 18 289 288 1 1 -- - - -- -

[4] Goldberg D.E, “Genetic algorithms in search optimization

VII. CONCLUSIONS
The methodology based on the GA approach to
optimal irrigation canal scheduling problem with
provision to open some outlet at specific time slot as per
request of user has been explained through an example of
Famen secondary canal of Feng-Jia-Shan Irrigation
District, China. In present formulation, application of
GA enables the operator to specify lower and upper
bound of starting time of some outlets as requested by the
user in rotation period which is not the case of the
classical optimization technique like integer
programming. GA model could be an efficient tool to
irrigation manager for planning water delivery schedule
for irrigation canal system as per the time slot demand of
user.
REFERENCES
[1] Anwar A. A, Clarke D, “Irrigation scheduling using
mixed-integer linear programming.” Journal of Irrigation
and Drainage Engineering ASCE 127(2), 2001, pp. 63–69.
[2] Chambers R. (Ed.), “Managing canal irrigation: practical
analysis from South Asia.” Oxford Publishing, New Delhi,
1988, pp. 20-25.
[3] Forrest S, “Genetic algorithms: Principles of natural
selection applied to computation.” Science 261(13): pp
872-878.
and machine learning.” 1989, Addison-Wesley, Reading,
Mass,
[5] Holland, J. H, “Adaptation in natural and Artificial
System.” 2nd ed. 1992, Mass. Inst. Of Technology.
Cambridge.
[6] Monem M. J. and Namdarian R., “Application of Simulated
Annealing (SA) techniques for optimal water distribution
in irrigation canals.” Irrigation and Drainage, 54, 2005, pp.
365-373
[7] Michalewicz Z, “Genetic algorithms + data structures =
evolution programs.” 1992, Springer, New York.
[8 Reddy, J. M, Wilamowski B. ,Sharmasarakar F.C,
“Optimal scheduling of irrigation for lateral canals.” ICID
Journal Vol. 48(3), 1999, pp.1-12.
[9] Suryavanshi A. R. and Reddy, J. M, “Optimal operation
scheduling of irrigation distribution systems”, Agricultural
Water Management., 11, 1986, pp. 23-30.
[10] Wang Zhi, Reddy J. M, Feyen J, “Improved 0-1
programming model for optimal flow scheduling in
irrigation canals.” Irrigation Drainage Systems 9, 1995,
pp.105-116.
[11 Wardlaw R. B, and Bhaktikul K, “Comparison of a genetic
algorithm and Linear Programming Approach for Lateral
Canal Scheduling.” Journal of Irrigation and Drainage
Engineering ASCE Vol. 130(4), 2004, pp. 311–317.

15
© 2009 ACADEMY PUBLISHER