PENALTY-FREE FEASIBILITY BOUNDARY-CONVERGENT MULTI-OBJECTIVE EVOLUTIONARY APPROACH FOR WATER DISTRIBUTION SYSTEMS BASED ON HEAD DEPENDENT MODELLING

Most water companies spend more than half of their total budgets to address the problem of rehabilitation. For example, the U.S. Environmental Protection Agency reported in 2001 that $151 billion would be needed for maintenance and replacement of drinking water systems in the USA over 20 years with 55% of this amount dedicated to pipelines. The proposed research addresses the problem of rehabilitation and long term planning for future upgrading of water distribution systems. Water distribution systems are an invaluable component of the critical infrastructure of urban populations worldwide. These systems need to be managed in a cost effective way while ensuring that key performance indicators and regulatory compliance criteria are not breached. The various requirements are extremely difficult to achieve because of their often conflicting nature and the sheer scale and complexities of water distribution systems. Optimization of water distribution systems is extremely challenging and the development of algorithms that are fast enough for routine use in industry is a most pressing issue.The optimal solutions of constrained optimization problems generally occur at the boundary of the feasible region of the solution space. The project will develop a radically different approach for dealing with constraints in evolutionary optimization algorithms for water distribution systems using head driven analysis. Head driven analysis provides the means to quickly and accurately identify the feasible region of the solution space and, more importantly, locate cost-effective solutions along its boundary without recourse to ad-hoc penalty functions. The application of evolutionary algorithms such as genetic algorithms to water distribution systems requires parameters such as the population size, crossover and mutation probabilities and a few empirical guidelines exist for the proper values of some of these parameters. However, information on how to determine suitable values of the penalty factors used to convert the typical water distribution system constrained optimization problem into an unconstrained optimization problem solvable by evolutionary algorithms is extremely scarce. In general the value of a penalty parameter is currently determined by trial and error. This is the principal issue to be addressed in this project, by developing a procedure which does away with the penalty parameter altogether. The goal is to develop fast evolutionary algorithms for use in industry on a routine basis.