The Wavelet Neural Network Based On An Improved Particle Swarm Optimization Algorithm And Its Application

Particle swarm optimization (PSO) algorithm is an intelligent global optimization algorithm. In PSO, the particle which is no mass and no volume is seeing as individuals. For each particle, it must conduct the rules, so the entire population of particles exhibit complex features and it can be used to solving complex optimization problems. Wavelet neural network is a forward artificial neural network which is proposed based on the wavelet transform. It both has the localization property of wavelet transform and the strong nonlinear approximation ability of neural network. Because it inherited the gradient descent method from neural network, so it is inevitably shows some defects such as slow convergence and so on. In the paper, the author use PSO optimizes each parameter of wavelet neural network, it can avoid the gradient descent process, and also it can guide the process of calculation. Furthermore, the PSO is simple, faster, more accurate, and all parameters can easily jump out of local optimal values through the iteration. In this paper, on the basis of expounding the main idea of artificial neural networks and particle swarm optimization, the author proposed an improved particle swarm optimization of wavelet neural network model, and then applied it to the people's livelihood in real estate and automotive industries projections. The main work can be briefly summarized as follows:(1) Sums up the results of previous studies, briefly described the development of artificial neural network, introduced the learning process of BP neural network and three improved methods; introduce the learning process of wavelet neural network, analyze its advantages and disadvantages.(2) According to the current research of particle swarm optimization, for its easy to fall into the local optimum, weak local search capability and other shortcomings, the author introduce a stochastic disturbance particle swarm optimization algorithm, that is, introduce the opinion of stochastic disturbance and an attractive operator into the SPSO.(3) Using four classic testing functions test the performance of SPSO and SDPSO respectively. The experimental result shows SDPSO is apt to getting the best option and its speed is faster than SPSO.(4) Using SDPSO optimizes the WNN, and then gets the SDPSO-WNN forecasting model; introduce the learning process of the SDPSO-WNN.(5) Analysis the important factors which impact the market of the real estate market and the sale for passenger cars in many aspect. Then put these factors as input variables. The experimental result shows the SDPSO is faster and more precise.