Research Of Wavelet Neural Network Based On Immune Particle Swarm Optimization Algorithm

At present, the wavelet neural network(WNN) based on basic particle swarm optimization algorithm(bPSO) has certain improvement in the convergence speed and the accuracy. But there is no good solution to entrap into the local extremum and improving the global search capability. Therefore, this paper proposes an improved particle swarm algorithm, the immune particle swarm optimization(IPSO) algorithm. Through some simulation experiments, it is proofed that the method presented by this paper can be a good solution to fall into the local extremum, improve the convergence speed of WNN, reduce the error precision and improve the global search capability. In general, there is a great help on improving the whole performance of WNN.Firstly, this paper introduces the developed history of WNN, the background at home and abroad and some training methods. Then, this paper introduces the theory of WNN and some frequently-used mother wavelets. And then it introduces some present frequently-used training methods of WNN. Next, this paper discusses importantly some relevant theories and some commonly modified methods of bPSO. Then, it leads to the modified method used by this paper. Specifically, on the basis of bPSO, the immune system of the artificial immune algorithm is introduced. Then this paper proposed a new algorithm, that is the IPSO algorithm. This algorithm not only helps to improve the diversity of the particle swarm population, but also to improve the convergence speed and accuracy of the particle swarm algorithm. By using the immune particle swarm algorithm to optimize the wavelet neural network, it can solve effectively the developed restriction the WNN, like slow convergence speed, falling easily into the local extremum, etc.Then, this paper uses respectively WNN based on bPSO and WNN based on IPSO to train a function. Through analysising the experimental datas, it is proofed that the performance of IPSO has more advantages than bPSO. Next this paper applies WNN based on bPSO and WNN based on IPSO to the simple motor single target tracking model. At last, through analyzing the results of the contrast experiments, it is proofed that the performance of this method proposed by this paper has more outstanding, greatly improving the network convergence speed and error precision in optimizing WNN. It is more suitable for choice.In the last, it summarizes the full text and analyses the deficiencies of the proposed method and discusses the future direction of the research.