Parameter Optimization For Wavelet Neural Network And Its Application

Wavelet neural network is the perfect combination of wavelet analysis theory and artificial neural network theory, which has the superiority of wavelet analysis and artificial neural network. On the one hand, the wavelet neural network makes full use of time-frequency localization properties of wavelet transform; and on the other hand, it fully plays the self-learning ability of an artificial neural network, which has strong approximation and fault-tolerant capability. However, because of its own structural problems of wavelet neural network and the problems caused by the combination of wavelet analysis and artificial neural network, the wavelet neural network model can easily fall into local minima or without convergence issues. Vast number of scholars have already study on the wavelet neural network parameter optimization problem, and with these problems, this paper made the following work:1. Based on the achievement of previous study results, and begin with the structure of wavelet neural network, the paper expounded the theory foundation of wavelet neural network, and pointed out the cause of the problem.2. Aimed at the existed problems in the application of wavelet neural network, the paper proposed to use particle swarm algorithm to optimize the parameters of wavelet neural network, with the speediness calculation and easy implementation of particle swarm algorithm.Wavelet neural network prediction model of crop water requirement based on particle swarm optimization algorithm was established, using the data of Fujin City. The prediction results of this model was better than the model of simpler wavelet neural network, so there is a great innovation in the theory.Wavelet neural network prediction model of dynamic change trend of the groundwater based on particle swarm optimization algorithm was established. The simulation result was same to the fact, so the model was feasible and had practical application value.3. Genetic Algorithm is a kind of randomized search method using evolution rule for reference, directly operating object is its main character, there is no the limitation of求导and function continuity; and it has better global optimization ability. Aimed at the existed problems in the application of wavelet neural network, the method of using Genetic Algorithm to optimize the parameters of wavelet neural network was taken.Engineering management performance evaluation model based on genetic algorithm optimized wavelet neural network was established. The experimental results showed the practicality of the method, and it can be used as a new method applied to the practice.Annual runoff prediction model based on genetic algorithm optimized wavelet neural network was established, using the data of Nuomin river. The simulation and prediction results were good, satisfied results was attained.