Studies On Individual Particle Swarm Optimization

Particle swarm optimization (PSO) is a population-based stochastic optimization algorithm simulated with social behavior among animal society such as bird flocking and fish schooling. Due to its simple implementation and fast convergent speed, PSO has been applied into many areas. In this article, a new individual PSO model is introduced from the agent viewpoint, and applied into parameter selection and structure optimization.In the standard particle swarm optimization, only the memory of each particle is used while other characters are omitted. This limitation makes large differences between PSO and its corresponding biological model, and affects the effciency significantly. Therefore, this article proposes an individual particle swarm optimization model, while each particle is viewed as an agent with memory, communication, responsibility, cooperation and self-learning ability. The incorporated interaction such as cooperation and competition makes this new model is more fit for biological background than standard PSO.Parameter selection is an important research topic in PSO literatures. Different from published strategies, for each particle, individual parameter selection strategy utilizes its information and advises a different value. Based on the fitness capability of each particle, in this article, a linear performance index is designed to simulate the self-learning ability, and each particle dynamically adjusts the ratio between global and local search capabilities with cooperative rules. Simulation results show the proposed individual strategies such as inertia weight, cognitive coefficient and social coefficient are effective.As another important research topic, this article discusses two different individual structure optimization manners. Because of the high selection pressure within individual inertia weight selection strategy, a special structure is designed to enlarge the global search capability, as well as shorten the local search capability. However, the exploration of this manner is still weak. Therefore, a new repulsive update manner is used to further improve the exploration capability. Simulation results show it can improve the population diversity significantly.