I. Gupta et al./Environmental Modelling & Software 14 (1999) 437–446
attempted by several researchers (Loganathan et al.,1995, 1990; Fujiwara et al., 1987; Quindry et al., 1979;Alperovits and Shamir, 1977). The optimal cost obtainedfrom the NLP technique was 0.57% higher than the sol-ution achieved by Loganathan et al. (1995). The sol-utions achieved by other researchers are 1.9–18.3%higher than the solution obtained by Loganathan et al.(1995).Further, a comparison between the results of the GAand NLP techniques for augmentation of several mediumsize networks showed that the GA in general provideda lower cost solution, than that obtained from the NLPtechnique. The hydraulic simulator ANALIS (Bassin etal., 1992) which is based on graph theory, was used inboth the NLP and GA solutions to calculate pressureheads, ﬂows and velocities in the design of branched,looped and combined systems.
2. Deterministic optimization techniques
A number of investigators have dealt with the problemof optimization of WDS by applying mathematical pro-gramming techniques.Several researchers employed linear programming tooptimise a WDS. Principal approaches include those of Alperovits and Shamir (1977), Quindry et al. (1981) andKessler and Shamir (1989). The technique given by Alp-erovits and Shamir (1977) requires that a set of variables(pipe ﬂows) be set to particular values before the linearprogramme can be formulated. Information availablefrom the solution of linear programming problem can beused to calculate a gradient which is then used to changepipe ﬂows. Quindry et al. (1981) have decomposed thelooped network problem to branched systems. The limi-tation of such simpliﬁed solution has been critically dis-cussed by Templeman (1982). Kessler and Shamir(1989) also use linear programming gradient procedure.Several non-linear programming packages have beendeveloped for network design problems. These packagesinclude GRG2 (Lasdon and Waren, 1983), MINOS(Murtagh and Saunders, 1987), GINO (Liebman et al.,1986), and GAMS (Brooke et al., 1988) which are allbased on the generalised reduced gradient method. Chi-plunkar et al. (1986) presented an algorithm based oninterior penalty function (IPF) with the Davidon-Fletcher-Powell (DFP) method. Lansey and Mays (1989)used GRG2 to ﬁnd the optimum design and to simulatepumps, tanks and multiple loading cases. Lansey et al.(1989) considered uncertainty in nodal demands, Hazen-Williams coefﬁcients and minimum nodal heads, anddeveloped a methodology for optimal design withrecourse to chance constrained optimization. Duan et al.(1990) extended the work of Lansey and Mays (1989)further and developed a model that (i) identiﬁes the num-bers and locations of pumps and tanks by implicit enu-meration, (ii) uses GRG2 to ﬁnd optimum pipe sizes forthe pump and tank layout speciﬁed in (i), and (iii) usesa separate model to compute various measures of sys-tem reliability.Gupta et al. (1993) developed the software packageWATDIS based on IPF and DFP methods. In thatapproach the problem was formulated as a cost minimiz-ation problem wherein the objective function F(x) com-prised the cost of power and annualised cost of pipes,pumps, and reservoirs satisfying the hydraulic loop lawswith constraints on minimum diameter and residualhead. The non-linear non-convex problem was convertedto an unconstrained problem by appending the con-straints to the objective function through penalty andweighting factors using the IPF method. An independentweighting factor was assigned to each constraint in orderto ensure the normalisation required by the signiﬁcantlydifferent contributions of diameter, reservoir height, andresidual head constraints to the unconstrained objectivefunction.Recently, Loganathan et al. (1995) presented a designheuristic for global cost minima design. That methodwas used to solve a standard eight pipe problem, eachpipe being 1000 m long with a Hazen Williams coef-ﬁcient of 130. The pipe sizes and associated costs usedin the study are presented in Table 1. By assuming aminimum diameter of 1 inch and minimum ﬂow con-straint of 1 m
/hour the method identiﬁed a design forthe network costing US $405 301.The same problem with the same minimum diameterand ﬂow constraints was solved by the authorsemploying WATDIS. Since the single cost equation wasof exponential form and did not show a good ﬁt (thecoefﬁcient of determination
0.932), a piecewise linearfunction was used to represent the cost. The optimal costobtained employing WATDIS is $407 625 which is theactual cost of the network ﬁnally calculated from costof pipes per unit length. This is 0.57% higher than theone reported by Loganathan et al. (1995). The details of the pipe cost and solution are presented in Fig. 1, andTables 1–4. The cost of the same network as determinedin a number of other methods in previous studies is
Table 1Pipe sizes and associated costsDiameter Unit cost Diameter Unit cost(in.) (US$/m) (in.) (US$/m)1 2 12 502 5 14 603 8 16 904 11 18 1306 16 20 1708 23 22 30010 32 24 550Note: 1 in.