An Extension of Particle Swarm Optimization Based on Partial Initialization (The 1st Report, Performance Evaluation on Test Functions)

Particle swarm optimization (PSO) is a population based stochastic optimization algorithm inspired by social behavior of bird flocking and fish schooling. PSO has proven to be implementable with ease, stable, scalable, and capable of yielding good results in a faster, cheaper way. However, it has been also reported that premature convergence to suboptima often occurs particularly in large scaled multimodal problems. This paper proposes two extensions for avoiding the premature convergence in standard PSO algorithms. Firstly, the partial initialization is applied in a small probability for keeping global search. Secondly, PSO is extended so as to have a function of local search intensively around the best solution. The second extension is designed as a mechanism that can prevent the partial initialization from expanding divergence and loss of the best solution. We conduct computer simulations and analyze searching behavior of the PSOs using a set of standard benchmark functions. The results show the PSO with our extensions surpass a standard technique particularly on large scaled multimodal functions.