Improving Genetic Algorithm with the Help of Novel Twin Removal Method

M. Imani, E. Pakizeh, and M. Saraee (Iran)

Keywords

Evolutionary Algorithms, Genetic Algorithm, Local Minima, Premature Convergence and Twin Removal (TR).

Abstract

Evolutionary Algorithms is one of the fastest growing areas of computer science. The simple Genetic Algorithm is fairly representative of other EAs. As they all use the same steps, significant researches in this area focus on Genetic Algorithm (GA). Two of the most important problems in EAs, are stalling in local minima and premature convergence. The analysis shows that similarity growth in the population leads to this problems. Twin Removal (TR) has been already investigated to reduce the similarity but most proposed TR methods are problem-specific and tend to gain better result rather than reducing GA runtime as a whole. In this paper, it has been proposed a novel, effective, and general TR method to reduce the negative impact of similarity as well as run time, preventing exploration in the already explored search space, and keeping diversification criteria in GA nearly the same. Results show that, removal of members of initial population having certain percentage of similarity would keep algorithms perform better, having fast convergence property intact as well as avoiding stalling. We discuss that the new generalized approach finds the same or even better solutions and the running time is less than half time of the standard one.

Important Links:

Go Back