An Improved Similarity Based Adaptive Step Size Glowworm Algorithm

Abstract

This paper presents the similarity based adaptive step size glowworm swarm optimization algorithm (SBASS_GSO), an improved version of glowworm swarm optimization algorithm (GSO). The standard GSO algorithm lacks unified metric standard to different problems in the selection of neighbor set, which makes the algorithm converge slowly because of improper selection. Because the step size s is fixed, the oscillation phenomenon may occur in local search space, which leads to inferior search accuracy In SBASS_GSO algorithm, we change neighborhood definition base on the similarity not on the distance. The neighborhood is selected by computing average similarity, which provides priori knowledge for the adaptive size s. The dynamic size s is useful for removing oscillation phenomenon and improving the convergence speed. Experimental results demonstrate the efficacy of the proposed glowworm algorithm in capturing multiple optima of a series of complex test functions, such as Zakharov and Sphere functions. We also provide some comparisons of SBASS_GSO with GSO and verify the superiority in the precision and convergence speed.