Professional Documents
Culture Documents
Tabu Search Algorithms For Water Network Optimization
Tabu Search Algorithms For Water Network Optimization
Tabu Search Algorithms For Water Network Optimization
www.elsevier.com/locate/dsw
O.R. Applications
a,*
, Luisa Ribeiro
Dep. de Engenharia Civil, Universidade de Coimbra, PoloII-Pinhal de Marrocos, 3030-290 Coimbra, Portugal
b
Instituto Superior de Engenharia de Coimbra, Quinta da Nora, 3031-601 Coimbra, Portugal
Received 15 March 2001; accepted 17 March 2003
Available online 12 August 2003
Abstract
In this paper we propose a tabu search algorithm to nd the least-cost design of looped water distribution networks.
The mathematical nature of this optimization problem, a nonlinear mixed integer problem, is at the origin of a multitude of contributions to the literature in the last 25 years. In fact, exact optimization methods have not been found for
this type of problem, and, in the past, classical optimization methods, like linear and nonlinear programming, were tried
at the cost of drastic simplications. Tabu search is a valuable heuristic technique for solving problems cast in combinatorial form. This is based on the human memory process and uses an iterative neighborhood search procedure in an
attempt to avoid becoming trapped in local optima. The use of such a heuristic procedure to solve the aforementioned
problem needs particular tailoring to produce high quality solutions. In this paper we present the essential features of
the algorithm and the results obtained when it is applied to some of the classical water distribution network case studies
appearing in the literature. The results are very promising and demonstrate the usefulness of tabu search algorithms in
solving this kind of optimization problem.
2003 Elsevier B.V. All rights reserved.
Keywords: Heuristics; Tabu search; Water distribution networks design; Optimization
1. Introduction
Water distribution networks are very important urban infrastructures whose realization needs
huge investments. Therefore it is important that
appropriate tools are used to help nd the best
decisions regarding their layout, design and operation. In the past twenty-ve years a great deal of
research has been conducted on the optimization
of water distribution networks.
*
Corresponding author.
E-mail addresses: mccunha@dec.uc.pt (M. da Conceic~
ao
Cunha), luirenco@mail.isec.pt (L. Ribeiro).
0377-2217/$ - see front matter 2003 Elsevier B.V. All rights reserved.
doi:10.1016/S0377-2217(03)00242-X
M. da Conceic~ao Cunha, L. Ribeiro / European Journal of Operational Research 157 (2004) 746758
747
748
M. da Conceic~ao Cunha, L. Ribeiro / European Journal of Operational Research 157 (2004) 746758
s:t:
sgn Qi Sn ;
8n 2 NN ;
i2NPn
DHk w
Lk
Cka Dbk
Hn P Hn min ;
sgnQk jQk j ;
8n 2 NN ;
8k 2 NP ; 3
4
Dk P Dmin ;
8k 2 NP ;
Dk 2 fDg;
8k 2 NP ;
M. da Conceic~ao Cunha, L. Ribeiro / European Journal of Operational Research 157 (2004) 746758
12
11
11
10
31
33
34
32
25
32
26
27
26
27
28
16
16
15
14
15
14
13
10
17
18
24
30
18
31
19
24
19
29 30
29
28
9
8
17
25
23
23
20
20
21
100.00 m
21
22
22
New pipe
91.44 m
15
15
14
New pipe
14
13
3
4
13
12
18
17
12
19
18
4
5
5
6
6
7
10
7
8
210.00 m
12
13
749
17
10
16
11
11
19
9
9
20
21
20
16
3
4
4
New pipe
8
7
5
6
750
M. da Conceic~ao Cunha, L. Ribeiro / European Journal of Operational Research 157 (2004) 746758
1
13
16
4
5
14
9
15
10
Existing pipe
New pipe
9
12
120.701 m
24
21
11
14
23
11
18
13
20
8
10
22
17
19
5
12
15
25
16
17
371.856 m
365.760 m
Existing pipe
New pipe
B jDHk j C
C
Qk sgnDHk B
@
A
Lk
w Ca Db
16
15
5
17
6
9
9
7
10
12
10
13
Neighborhood structures will vary with algorithm behavior. The move considered rst consists
of reducing the diameter of one pipe at a time.
Therefore neighbor solutions are characterized by
congurations in which all pipes but one have the
same diameter as in the current conguration. The
new current solution will be the one presenting
the least-cost while meeting hydraulic constraints.
The new current solution is selected by evaluating the cost of each conguration and solving the
hydraulic equilibrium equations. For this purpose,
a hydraulic simulator is incorporated into the algorithm. The hydraulic model assumes steady state
ow conditions and the mass conservation law
equations are rewritten in terms of node heads.
The ow is computed by
0
11=a
;
8k 2 NP :
11
11
14
12
B jDHk j C
C
sgnDHk B
@
A
Lk
i2NPn
w C a Db
k
8n 2 NN :
Sn ;
8
This yields a set of nonlinear equations that is
solved by the method proposed in Gessler (1981).
The ow in each pipe and the pressure in each
node are obtained, thereby allowing the evaluation
of head constraints.
When a local optimum is reached, the move
consists of increasing the diameter of one pipe.
The new current solution will be dened according
to the rules established by the diversication procedure, while meeting the hydraulic constraints.
4.3. Diversication procedure
When a local optimal solution is reached, the
neighbor solutions are generated by increasing pipe
diameters. The information given by a long-term
M. da Conceic~ao Cunha, L. Ribeiro / European Journal of Operational Research 157 (2004) 746758
8500000
8000000
Cost (Dollar)
751
7500000
7000000
6500000
6000000
1
11 13 15 17 19 21 23 25 27 29 31 33
Tabu tenure (Iteration)
Cost (Dollar)
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Tabu tenure (Iteration)
Without aspiration criterion
8500000
8000000
Cost (Dollar)
47000000
46000000
45000000
44000000
43000000
42000000
41000000
40000000
39000000
38000000
37000000
7500000
7000000
6500000
6000000
1
11 13 15 17 19 21 23 25 27 29 31 33
Tabu tenure (Iteration)
ure 2 or 4 for network 2; and in the implementation T.S.1 with tabu tenure 9 for network 3.
But when this parameter is dened randomly an
improvement is accomplished. Fig. 10 displays the
solutions obtained by T.S.2 implementation for
network 2 with a random tabu tenure parameter,
using dierent pseudorandom generator seed
numbers. Only in two cases is the best-known
historical solution not found (and even in these
Cost (Dollar)
752
M. da Conceic~ao Cunha, L. Ribeiro / European Journal of Operational Research 157 (2004) 746758
47000000
46000000
45000000
44000000
43000000
42000000
41000000
40000000
39000000
38000000
37000000
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Tabu tenure (Iteration)
Without aspiration criterion
4.6. Results
M. da Conceic~ao Cunha, L. Ribeiro / European Journal of Operational Research 157 (2004) 746758
753
Table 1
Solution costs for the ve networks
Authors
Networks
1
Solution
3
Dmin 2
in.
Dmin 4
in.
834
858
498
41.8
38.9
39.2
435
39.0
418
36.6
6116
6319
1884.432
6045
38.8
402
6027
38.0
38.8
419
419
419
6000
6162
6056
6056
6056
420
420
6056
6056
37.1
37.1
37.1
37.1
810
810
808
830
830
830
1750.320
1727.616
1727.616
Split-pipe
Discrete diameters
Discrete diameters
Split-pipe
Discrete diameters
Split-pipe
Split-pipe
Split-pipe
Split-pipe
Continuous diameters
Split-pipe
Discrete diameters
Split-pipe
Discrete diameters
Split-pipe
Split-pipe
Discrete diameters
Continuous diameters
Discrete diameters
Discrete diameters
Discrete diameters
Discrete diameters
Discrete diameters
Discrete diameters
Discrete diameters
754
M. da Conceic~ao Cunha, L. Ribeiro / European Journal of Operational Research 157 (2004) 746758
Table 2
Solution for network 4
Pipe
number
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
Length (m)
365.76
365.76
365.76
365.76
365.76
365.76
365.76
365.76
365.76
365.76
365.76
365.76
243.84
243.84
243.84
243.84
243.84
243.84
243.84
243.84
243.84
243.84
243.84
243.84
914.40
Hazen
Williams
coecients
120
120
120
120
120
120
120
120
120
120
120
120
120
120
120
120
120
120
120
120
120
120
120
120
120
Diameter (mm)
Minimum diameter 2 in. 50.8 mm
Gessler
(1981)
Gessler
(1981)
Sousa and
Cunha
(1999)
152.4
152.4
152.4
50.8
50.8
50.8
50.8
50.8
50.8
203.2
203.2
203.2
152.4
152.4
152.4
50.8
50.8
50.8
50.8
50.8
50.8
152.4
152.4
152.4
254.0
834,000
50.8
50.8
101.6
203.2
50.8
50.8
50.8
203.2
203.2
50.8
101.6
101.6
203.2
50.8
50.8
50.8
203.2
50.8
50.8
50.8
101.6
101.6
101.6
203.2
254.0
810,000
T.S.2
50.8
50.8
50.8
203.2
50.8
101.6
50.8
101.6
50.8
50.8
203.2
203.2
203.2
50.8
50.8
50.8
203.2
203.2
101.6
50.8
101.6
50.8
101.6
101.6
254.0
807,999
101.6
101.6
101.6
101.6
101.6
101.6
101.6
101.6
101.6
203.2
203.2
203.2
152.4
152.4
152.4
101.6
101.6
101.6
101.6
101.6
101.6
152.4
152.4
152.4
254.0
858,000
Sousa and
Cunha
(1999)
T.S.1 and
T.S.2
101.6
101.6
101.6
101.6
101.6
101.6
152.4
152.4
101.6
101.6
152.4
203.2
203.2
152.4
101.6
101.6
101.6
101.6
101.6
101.6
152.4
101.6
101.6
152.4
254.0
101.6
101.6
101.6
101.6
101.6
101.6
152.4
152.4
101.6
101.6
152.4
203.2
203.2
152.4
101.6
101.6
101.6
101.6
101.6
101.6
152.4
101.6
101.6
152.4
254.0
830,000
830,000
M. da Conceic~ao Cunha, L. Ribeiro / European Journal of Operational Research 157 (2004) 746758
755
Table 3
Solution for network 5
Pipe number
Length (m)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
4828.032
1609.344
1609.344
6437.376
1609.344
1609.344
1609.344
1609.344
1609.344
1609.344
1609.344
1609.344
1609.344
1609.344
4828.032
6437.376
1609.344
HazenWilliams
coecient
Diameter (mm)
Walski et al. (1990)
75
80
80
80
80
100
120
120
80
100
120
100
120
120
120
120
120
355.6
254.0
254.0
254.0
254.0
203.2
203.2
304.8
254.0
101.6
304.8
203.2
203.2
203.2
0.0
355.6
0.0
355.6
254.0
254.0
254.0
254.0
203.2
203.2
304.8
254.0
101.6
203.2
203.2
152.4
254.0
0.0
355.6
0.0
355.6
254.0
254.0
254.0
254.0
203.2
203.2
304.8
254.0
101.6
254.0
203.2
101.6
203.2
0.0
355.6
0.0
1,884,432
1,750,320
1,727,616
Table 4
Running times (in seconds)
Networks
T.S.1
T.S.2
4 (Dmin 2 in.)
4 (Dmin 4 in.)
16
16
391
445
54
60
71
94
46
65
46
55
5. Conclusions
A tabu search algorithm has been constructed
to enable the least-cost design of looped water
distribution networks to be found. The ne-tuning
of its parameters permits high quality solutions to
be achieved, and the results obtained reveal the
ability of the algorithm to cope with the combi-
natorial characteristics of this problem. Some improvements in the results achieved with tabu
search algorithms can be indicated.
The small number of case studies available in
the literature means that the comparison of the
solutions provided by the three modern heuristic
techniques (genetic algorithms, simulated annealing and tabu search) does not allow general conclusions to be drawn about the heuristic most
suitable for solving the problem proposed here.
Since the solution of realistic network problems
would benet greatly from the application of
heuristics, the authors suggest that a database of
networks should be set up. This would enable the
conclusions drawn from studies carried out on
these networks to be more easily compared by
researchers in this eld.
756
M. da Conceic~ao Cunha, L. Ribeiro / European Journal of Operational Research 157 (2004) 746758
Table 5
Morgan and Goulter (1985) and Murphy et al. (1993) solutions
Pipe
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
L (m)
3535.723
6035.113
2225.067
2529.871
2621.312
5821.751
2926.116
3810.046
2926.116
3413.802
4419.654
3718.605
7345.769
6431.358
4724.457
8046.818
9509.876
7315.289
4389.173
11704.462
8046.818
D (in.)
D (mm)
D (in.)
D (mm)
0
0
0
0
0
0
144
0
0
0
0
0
0
0
0
96
96
84
60
0
84
0
0
0
0
0
0
3657.6
0
0
0
0
0
0
0
0
2438.4
2438.4
2133.6
1524
0
2133.6
0
0
0
0
0
0
0
0
0
0
0
0
0
0
120
84
96
84
72
0
72
0
0
0
0
0
0
0
0
0
0
0
0
0
0
3048
2133.6
2438.4
2133.6
1828.8
0
1828.8
39.2
Cost ($ million)
38.8
51000000
Table 6
Gessler (1982) solution
Cost ($ million)
D (in.)
D (mm)
D selected
(mm)
0
0
0
0
0
0
100
100
0
0
0
0
0
0
0
100
100
80
60
0
80
0
0
0
0
0
0
2540
2540
0
0
0
0
0
0
0
2540
2540
2032
1524
0
2032
0
0
0
0
0
0
2438.4
2438.4
0
0
0
0
0
0
0
2438.4
2438.4
2133.6
1524
0
2133.6
41.8
Cost (Dollar)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
Gessler (1982)
47000000
45000000
43000000
41000000
39000000
37000000
0
150
300
450
600
Iteration
750
900
1050
Fig. 11. Algorithm development when the Gessler (1982) solution is the starting solution.
51000000
49000000
Cost (Dollar)
Pipe
49000000
47000000
45000000
43000000
41000000
39000000
37000000
0
40
80
Iteration
120
160
Fig. 12. Initial iterations when the Gessler (1982) solution is the
starting solution.
Cost (Dollar)
M. da Conceic~ao Cunha, L. Ribeiro / European Journal of Operational Research 157 (2004) 746758
46000000
45000000
44000000
43000000
42000000
41000000
40000000
39000000
38000000
37000000
0
10
20
30
40
50 60 70
Iteration
80
Fig. 13. Initial iterations when the Morgan and Goulter (1985)
solution is the starting solution.
53000000
Cost (Dollar)
51000000
49000000
47000000
45000000
43000000
41000000
39000000
37000000
200
220
240
260
280
300
320
340
Iteration
Fig. 14. Initial iterations when the Murphy et al. (1993) is the
starting solution.
The results obtained with tabu search algorithms seem very promising, encouraging us to go
further in future works by including pumps and
other devices, reservoir sizing, etc.
References
Alperovits, E., Shamir, U., 1977. Design of optimal water
distribution systems. Water Resources Research 13 (6), 885
900.
Cunha, M.C., Sousa, J., 1999. Water distribution network
design optimization: Simulated annealing approach. Journal
of Water Resources Planning and Management, ASCE 125
(4), 215221.
Cunha, M.C., Sousa, J., 2001. Hydraulic infrastructures design
using simulated annealing. Journal of Infrastructure Systems, ASCE 7 (1), 3239.
Dandy, G.C., Simpson, A.R., Murphy, L.J., 1996. An improved genetic algorithm for pipe network optimization.
Water Resources Research 32 (2), 449458.
Eiger, G., Shamir, U., Ben-Tal, A., 1994. Optimal design of
water distribution networks. Water Resources Research 30
(9), 26372646.
757
758
M. da Conceic~ao Cunha, L. Ribeiro / European Journal of Operational Research 157 (2004) 746758