Improved Particle Swarm Optimization Algorithm Under Dynamic Environment

The particle swarm optimization (PSO) method as an evolutionary technique was originally introduced by Kennedy and Eberhart in 1995. The PSO idea was inspired by natural concepts such as bird flocking, fish schooling and human social relations. It has been applied successfully in various optimization problems, neural networks training, pattern classification, fuzzy system control and some engineering domains. This paper summarized the ideas and backgrounds of the PSO method, introduced the standard PSO method and some other improved PSO methods in detains. By combining the idea of PSO and Evolutionary Programming (EP), this paper introduced an improved PSO method under the dynamic environment, the effectiveness of this method in tracking changing optimum was investigated under different changing environments. Furthermore, we introduced the concept of population entropy and analyzed the relations between the population diversity and the effectiveness of tracking changing optimum.This paper contains two important aspects:First, the standard PSO method's population diversity often losses too fast in the evolving process of which resulted in the premature of the algorithm.This makes it can't track changing optimum successfully and promptly. This paper introduced an improved PSO method (IPSO) under the dynamic environments, the effectiveness this method and some other methods in tracking changing optimum were experimented under different changing environments. Experiments indicate that the IPSO method can applied in dynamic environments more successfully than other methods, the speed and precision were improved.Second, this paper analyzed the population diversity of some PSO methods under dynamic environments. Furthermore, relations between the population diversity and the effectiveness were experimented. Experiments indicated that particle's different movements can affect the population diversity and the effectiveness of tracking changing optimum. If the number of divergent particles in the population is too small, the method probably can't track the changing optimum, but if this number is too large, the population's convergence is weakened, tracking error is increased. So, different problems should have different population diversity based on the needs of the problems.