A hybrid multi-objective genetic algorithm for bandwidth multi-coloring problem

Genetic Algorithms (GAs) have been successfully applied on different kinds of problems. Multi-objective Genetic Algorithms (MOGAs) are capable of improving different objectives in a parallel manner. Various applications of MOGAs exist for combinatorial optimization problems. However, the MOGA approach yields a limited success rate especially on grouping problems. The crossover operation, one of the reproduction methods in GAs, is the main reason for the low performance. The crossover operation is quite destructive in grouping problems and it is difficult to produce successful offspring with this operator in this domain. In this study, a novel method that can increase the success rate of crossover operation is proposed for grouping problems. The method is a hybridization of MOGA with Artificial Neural Networks (ANNs), where ANNs guide the crossover process in the genetic search. The bandwidth multicoloring problem where standard MOGA yields limited performance has been used as the testbed for the method. The problem is solved using a multi-objective framework that minimizes bandwidth as well as conflict number in a parallel fashion. It has been observed that the crossover operation guided by the trained ANN improves the possibility of producing high fit offspring and the quality of the overall solution obtained at the end of MOGA runs.