A knowledge-based technique for initializing a genetic algorithm

Abstract

Genetic Algorithms are efficient for the travelling salesman problem, but they have a premature convergence problem resulting in suboptimal solutions. As the initialization step has a profound impact on the algorithm’s performance, this study proposes a knowledge-based initialization technique to l earn the patterns of evolved populations, and integrates four heuristic strategies to generate the initial population. Advanced initial solutions and high quality gene blocks can be quickly created with this method. Instances in the TSPLIB library are used to set the parameters and test different initialization methods. The results show that this proposed technique can improve the initial population and optimization performance of genetic algorithms.