Extended pso with partial randomization for large scale multimodal problems

Particle swarm optimization (PSO) is a population-based stochastic optimization algorithm inspired by the social behaviors of bird flocking and fish schooling. Each particle searches for a better solution through interaction with other particles. However, PSO tends to prematurely converge to a local minimum, particularly for large-scale multimodal problems. This paper proposes two extensions for avoiding the premature convergence observed in standard PSO algorithms. First, partial randomization is applied on particles in a small probability for performing a continuous global search. Next, PSO is extended to perform an intensive local search around the best solution. This second extension is designed as a mechanism that can prevent partial randomization from causing inordinate divergence and thereby losing the best solution. We conducted computer simulations and analyzed the searching behavior of the PSOs using a set of several standard benchmarks. The results exhibit an improved performance of PSO with our extensions, especially on large-scale multimodal functions.