Experimental study of seeding in genetic algorithms with non-binary genetic representation

Abstract

Seeding is a technique used to leverage population diversity in genetic algorithms. This paper presents a quick survey of different seeding approaches, and evaluates one of the promising ones called the Seeding Genetic Algorithm. The Seeding GA does not include mutation, and it has been shown to work well on some GA-hard problems with binary representation, such as the Hierarchical If-and-Only-If or Deceptive Trap. This paper investigates the effectiveness of the Seeding GA on two problems with more complex non-binary representations: capacitated lot-sizing and single-machine scheduling. The results show, with statistical significance, that the new GA is consistently outperformed by the conventional GA, and that not including mutation is the main reason why. A detailed analysis of the results is presented and suggestions are made to enhance and improve the method.