Modification And Application Of Particle Swarm Optimization Algorithm

Particle swarm optimization (PSO) algorithm is a swarm intelligent algorithm which is based on imitating the bird flock's preying behavior, and it is a new branch of the evolutionary computation technique. The main trait of PSO is simple in principle, few in tuning parameters, speedy in convergence and easy in implementation. So, the PSO algorithm immediately attracts the scholars' extensive attention when it is proposed. Gradually, it becomes a hotspot in the evolutionary computation field. Now, the PSO is used for the training of neural networks, optimization of functions and multi-target. It obtains good effect, and its applied foreground is wide.The main works of the dissertation can be summarized as follows:(1) This paper makes a brief introduction of the basic theories and present situation of PSO research. The PSO algorithm principles and flow are analyzed in detail. In the paper, the parameters of PSO are researched and the corresponding emulator experiments are carried out.(2) The problems of PSO algorithm are analyzed. It contains the setting of the parameters, the premature of PSO and the stability of PSO. In PSO algorithm, the settings of parameter have much influence on the result. So, how to choose parameters to achieve the best solutions is the problem that should be solved. The premature problem is prevalent in the optimization algorithm. If particles converge early, the algorithm may get into local minimum value and can't find the global optimization. Because particle's position, speed and some parameters are random initialized, the result will be different at every turn and sometimes it differs a lot. In this way, it leads to the unstable results.(3) Aiming at the problems of PSO, the particle swarms optimization based on particle evolution is proposed in the paper. It uses local version of particle swarm optimization algorithm to improve the standard PSO in the two aspects of particle evolution and swarms. The independent searching in the solution space of particle swarms makes it diversified, and then enhances the global searching ability. The evolutional particle will be reset in order to force it getting out of locally minimum. It makes the particle escaped from the premature convergence and increases the stability of algorithm. Comparative experiments on testing functions indicate that the improved algorithm is effective.(4) The particle swarms optimization based on particle evolution is applied to blind source separation. Compared to standard PSO, the former results of emulator experiments need fewer iterative to separate the mixed signals and its stability is higher.(5) The particle swarms optimization based on particle evolution is used for solving nonlinear equations. The algorithm has good convergence reliability and can get the solving results quickly and accurately. This method overcomes the problems of some algorithms' sensitivity to the initial value and their need for differential of functions, and it can find the best solution of nonlinear equations quickly. Simulation results show that the algorithm is feasible and effective. This algorithm provides a practical method for solving nonlinear equations.