Optimal (short-term) pump schedule detection for water distribution systems by neutral evolutionary search

The application of neutrality is a straightforward tool to preserve population diversity since it allows the genotype (on the represented search space) to be changed without affecting the corresponding fitness. To implement neutrality the literature suggests representational redundancy (more to one correspondence in genotype–phenotype mapping) although using it as a source of neutrality researchers uniformly reported better or worse results. Instead of applying representational redundancy here the utilization of pseudo redundancy as the source of neutrality is proposed, that is, neutrality is achieved by simple objective-fitness transformation while pseudo redundancy (as another redundancy interpretation) denotes more to one correspondence between objective-fitness domains by objective-fitness mapping. The contribution of this work is specified by the dynamic generational gap model introduced for evolutionary algorithms which appears when elitist strategy is used under neutrality by pseudo redundancy. This paper investigates the influence of dynamic generational gap model on the performance of a micro-genetic algorithm framework applied to achieve least cost water pump control policy for an industrial size water network distribution system. The presented constrained mixed-integer optimization problem is originated from the regional water network of the city of Sopron (60,000 citizens) located in Hungary. Here, the goal is to obtain intra-day pump schedule which minimizes the cost required for operation while satisfies the system constraints (water reservoir level limitations, pump flow and delivery regulations, pump energy consumption limitations) and fulfills the water requirement by the users.

Figure optionsDownload as PowerPoint slideHighlights
► A dynamic generational gap GA model is proposed implementing the theory of neutrality.
► Neutrality is achieved by objective-fitness transformations.
► GA is applied and benchmarked on a (single-objective constrained) scheduling problem.
► Remarkable improvement in success rate and convergence speed is reported.