An efficient heuristic-based evolutionary algorithm for solving constraint satisfaction problems

Abstract

GENET and EGENET are artificial neural networks with remarkable success in solving hard constraint satisfaction problems (CSPs) such as car sequencing problems. (E)GENET uses the min-conflict heuristic in variable updating to find local minima, and then applies heuristic learning rule(s) to escape the local minima not representing solution(s). In this paper we describe a micro-genetic algorithm (MGA) which generalizes the (E)GENET approach for solving CSPs efficiently. Our proposed MGA integrates the min-conflict heuristic into mutation for reassigning allels (values) to genes (variables). In addition, we derive two methods, based on general principles from evolutionary algorithms, for escaping local minima: population based learning, and look forward. Our preliminary experimental results showed that this evolutionary approach improved on EGENET in solving certain hard instances of CSPs.