Using Fitness Landscape to Improve the Performance of Particle Swarm Optimization

Particle swarm optimization (PSO) is a swarm intelligent stochastic optimization algorithm inspired by social behavior. Due to the fast convergent speed, it gets easily trapped into a local optimum when solving high-dimensional multi-modal numerical optimization problems. To avoid this phenomenon, a new strategy fitness landscape modification is introduced to reduce the number of local optima. In this new strategy, the fitness values of some particles are estimated, while others are evaluated with the truly objective functions. To provide a precise estimation, the estimated fitnesses are weighted combination where each weight of one parent individual is proportional to the distance among estimated particle and itself. With this manner, many potential local optima which may guide a wrong searching direction departing from global optimum are removed. To testify the performance of this new strategy, several famous multi-modal benchmarks are chosen, and simulation results show fitness landscape modification is an effective strategy which improves the performance significantly when compared with the standard PSO. Furthermore, this new modification is used to solve stochastic expected model, the performance increases 4.29%similar to 5.37% when compared with the known best results.