Mutation Variations in Improving Local Optima Problem of PSO

Abstract

This paper experiment on various concepts in performing mutation to lessen trap in a local optima problem of Particle swarm optimization (PSO). The first concept is when to perform mutation. The earlier mutation favors exploration more than exploitation and usually leads to slow convergence, while the late mutation tends to have opposite characteristics. The second concept is the reset of a known best position (GBEST) when trapping in local optima. The reset reduces the chance of trapping in the same local optima but may lead to slower convergence. On the other hand, mutations without reset best position exploit previous knowledge and converge faster if the GBEST closes to optima. The performances of each concept are compared using 27 benchmark test functions. The results are mixing, but the early mutation without reset GBEST perform better in many of test function.