Using a genetic programming–based hyper-heuristic strategy, we automatically build evolutionary algorithm operators to solve the bi-objective water distribution network design issue

Abstract

Finding the appropriate pipe diameters that give the greatest service at the lowest cost is at the heart of the water distribution network (WDN) design challenge, which is of ongoing relevance in the UK and across the world. As a result, a plethora of solutions to this issue have been presented in the literature, with many of them taking a more bespoke, artisanal approach. In this research, we look into a new hyper-heuristic technique that use genetic programming (GP) to develop mutation operators for evolutionary algorithms (EAs) tailored to a dual-goal formulation of the WDN design issue (minimizing WDN cost and head deficit). The evolved operators, once developed, may be employed indefinitely across all EAs on all WDNs to boost performance. We show that it is possible to develop a set of mutation operators for a single training WDN using a unique multi-objective approach. The top operators are rigorously tested on three different, more difficult test networks. In this experiment, we develop a set of 83 operators. Ten that made the cut are dissected here. While GP5 exhibits the method's capacity to locate well-known operators like a Gaussian, GP1 is proven to be very successful and adds important domain-specific learning (pipe smoothing).