The Research Of Basic Theory And Improvement On Particle Swarm Optimization

Particle Swarm Optimization is an intelligence algorithm constructed by the simulation of biologic swarm social activity, which imitated cooperative ability of searching optimal location of biologic swarm and is developed as an optimization algorithm. But, because it is a stochastic approximate algorithm, the theory of Particle Swarm Optimization is incomplete, and there exist deficiencies on premature convergence and difficulty of application in discrete problem. So, the research on the theory analysis, improvement and discrete problem of particle swarm optimization is very significant. In this thesis, we have analyzed and improved standard particle swarm optimization and binary particle swarm optimization. We have following conclusions:Particle swarm optimization is a stochastic algorithm with heuristic information, each particle fly following the self optimal location and global optimal location with stochastic factor. It has been analyzed that the particles converge to the global optimal location with stochastic iteration going on. In the condition of adding stochastic factor and the optimal particle updation, it has been proved that the particle trajectories will converge to the swarm optimal location in this thesis.During the running of the algorithm, the particles become more and more similar, and cluster into the best particle in the swarm, which make the swarm premature convergence around the local solution. In this thesis, a new conception, collectivity, is proposed which is based on similarity between the particle and the current global best particle in the swarm. And the collectivity was used to randomly mutate the position of the particles, which make swarm keep diversity in the search space. Experiments on benchmark functions show that the new algorithm outperforms the standrard PSO and other improved PSO.In fact, PSO is a random evolution algorithm. However, during the evolution of the algorithm, the magnitude of inertia weight has impact on the exploration and exploitation of PSO, which is a contradiction. In this thesis, a new PSO algorithm, called as DPSO, is proposed in which the inertia weight of every particle will be changed dynamically with the distance between the particle and the current optimal position. Experiments on benchmark functions show that DPSO outperform standard PSO.The current theory of PSO has constructed a mathematic modal, which can give a clear essence of PSO from mathematic view.In this thesis, the velocity and location updating equation are replaced by the mathematic equation, and get a new evolutionary algorithm. By select appropriate parameters, the performance of new algorithm is not inferior to standard PSO by simulation on benchmark functions. The new algorithm was applied to the real-time dynamic model on multiphase traffic flows. A simulation experiment for the traffic model at a four-phase intersection is also performed in this thesis. The results show that the method is effective.Binary PSO is a new method to solve discrete problem, which is applied in many area. In this thesis, Binary PSO has been analyzed through bit change rate, velocity expected value and genetic algorithm pattern theory. It has been found that binary PSO is not converge to global optimal particle, and the bit is more and more stochastic with iteration going on, so it is lack of local exploration. So, an improved binary PSO are proposed which meet the idea of PSO.