A New Discretization Technique for Enhancing Discrete Particle Swarm Optimization’s Performance

Abstract

In this paper, a new technique of converting continuous values to discrete values for Particle Swarm Optimization (PSO) is proposed. PSO is a powerful metaheuristic search technique that has been deployed to solve many complicated and large optimization problems. However, it has a drawback on deteriorated performance when dealing with a discrete search space. The proposed discretization technique borrowed the idea of a traditional gambling game, Sic Bo game. The new technique allowed us to cultivate the information about the global best of the swarm and the personal best of the particles that the existing discretization techniques do not. The proposed technique was tested on uncapacitated fixed- charge location problems and compared to other three existing techniques: Sigmoid function, Hyperboric Tangent function, and Khanesar et al. ’ s technique. Furthermore, ten benchmark problems were drawn from the OR- library. The experimental results showed that the new technique outperformed the competitive techniques on medium to large size benchmark problems. It yielded better solutions which varied from 0.17% to 4.19%. The comparable technique in this study was Khanesar et al.’s technique. It yielded 0.13% difference from our technique on large size benchmark problem. Nevertheless, by performing statistical testing, our proposed technique yielded the better solutions significantly.