Comparing lbest PSO niching algorithms using different position update rules

Niching is an important technique for multimodal optimization in Evolutionary Computation. Most existing niching algorithms are evaluated using only 1 or 2 dimensional multimodal functions. However, it remains unclear how these niching algorithms perform on higher dimensional multimodal problems. This paper compares several schemes of PSO update rules, and examines the effects of incorporating these schemes into a lbest PSO niching algorithm using a ring topology. Subsequently a new Cauchy and Gaussian distributions based PSO (CGPSO) is proposed. Our experiments suggest that CGPSO seems to be able to locate more global peaks than other PSO variants on multimodal functions which typically have many global peaks but very few local peaks.