The Modeling And Application Of The Wavelet Neural Network On Complicated Nonlinear System

The research of artificial neural network, wavelet analysis and genetic algorithm is the frontier and focus of information science technology, which has great value in theory and application for identification and predicting of nonlinear systems. On the basis of studying wavelet analysis, artificial neural network and genetic algorithm, the author has completed the following research: In the article, the author studied approximation performance of artificial neural network and wavelet network. The advantages of high approximation and fast convergent rate of wavelet are certified through the analysis of theory and in accordance with the result of simulation. On the basis of the application background of identification problems of nonlinear system, adaptive wavelet neural network is obtained by integrating the multiresolution analysis theory of wavelet with artificial neural network. This model consists of the smooth sub-network and details sub-network and can adaptively incorporate new details sub-network into the initial network to improve the accuracy. However, training the new sub-network has no influence on the structure of the initial network. The author proposed the model learning improved algorithm based on the adaptive wavelet network,,hence the problem of determining the number of the nodes in the hidden layer in traditional ways can be solved. The sub-model structure identification algorithm based on genetic algorithm is adopted to determine the number of nodes in each hidden layer. The simulation results shows that this approach has two advantages, high approximation and good generalization. In the data of chromatogram noise exists, which destroies the initial actual signals. And this affects the automatic recognizing of the peak of chromatogram and the precise calculating of the peak area. So chromatogram data need to be denoised before the qualitative analysis and quantitative analysis. The author eliminates noise with deniosing of the wavelet threshold on wavelet analysis. The algorithm of adaptive wavelet network that is mentioned in the article is applied to the modeling of oil field record well chromatographic data and the modeling of oil field well test pressure data, and good results are achieved. The effectiveness of the scheme in the article is certified by simulation and application.