Novel Methods for Enhancing the Performance of Genetic Algorithms
Esra'a O Alkafaween
DOI: https://doi.org/10.48550/arXiv.1801.02827
2018-01-09
Neural and Evolutionary Computing
Abstract:In this thesis we propose new methods for crossover operator namely: cut on worst gene (COWGC), cut on worst L+R gene (COWLRGC) and Collision Crossovers. And also we propose several types of mutation operator such as: worst gene with random gene mutation (WGWRGM) , worst LR gene with random gene mutation (WLRGWRGM), worst gene with worst gene mutation (WGWWGM), worst gene with nearest neighbour mutation (WGWNNM), worst gene with the worst around the nearest neighbour mutation (WGWWNNM), worst gene inserted beside nearest neighbour mutation (WGIBNNM), random gene inserted beside nearest neighbour mutation (RGIBNNM), Swap worst gene locally mutation (SWGLM), Insert best random gene before worst gene mutation (IBRGBWGM) and Insert best random gene before random gene mutation (IBRGBRGM). In addition to proposing four selection strategies, namely: select any crossover (SAC), select any mutation (SAM), select best crossover (SBC) and select best mutation (SBM). The first two are based on selection of the best crossover and mutation operator respectively, and the other two strategies randomly select any operator. So we investigate the use of more than one crossover/mutation operator (based on the proposed strategies) to enhance the performance of genetic algorithms. Our experiments, conducted on several Travelling Salesman Problems (TSP), show the superiority of some of the proposed methods in crossover and mutation over some of the well-known crossover and mutation operators described in the literature. In addition, using any of the four strategies (SAC, SAM, SBC and SBM), found to be better than using one crossover/mutation operator in general, because those allow the GA to avoid local optima, or the so-called premature convergence. Keywords: GAs, Collision crossover, Multi crossovers, Multi mutations, TSP.