Genetic Algorithms for Least-Cost Design of Water Distribution Networks

Abstract

The paper describes the development of a computer model GANET that involves the application of an area of evolutionary computing, better known as genetic algorithms, to the problem of least-cost design of water distribution networks. Genetic algorithms represent an efficient search method for nonlinear optimization problems; this method is gaining acceptance among water resources managers/planners. These algorithms share the favorable attributes of Monte Carlo techniques over local optimization methods in that they do not require linearizing assumptions nor the calculation of partial derivatives, and they avoid numerical instabilities associated with matrix inversion. In addition, their sampling is global, rather than local, thus reducing the tendency to become entrapped in local minima and avoiding dependency on a starting point. Genetic algorithms are introduced in their original form followed by different improvements that were found to be necessary for their effective implementation in the optimization of water distribution networks. An example taken from the literature illustrates the approach used for the formulation of the problem. To illustrate the capability of GANET to efficiently identify good designs, three previously published problems have been solved. This led to the discovery of inconsistencies in predictions of network performance caused by different interpretations of the widely adopted Hazen-Williams pipe flow equation in the past studies. As well as being very efficient for network optimization, GANET is also easy to use, having almost the same input requirements as hydraulic simulation models. The only additional data requirements are a few genetic algorithm parameters that take values recommended in the literature. Two network examples, one of a new network design and one of parallel network expansion, illustrate the potential of GANET as a tool for water distribution network planning and management.