Research Of Wavelet Neural Network Optimization And It's Application

Wavelet neural network (WNN) is a novel neural network based on wavelet analysis theory and artificial neural network. Because the WNN inherits the time-frequency localization of wavelet analysis and self-learning ability of neural network, it is applied extensively in nonlinear function approximation, nonlinear system identification and dynamic model building, signal approximation and classification fields.In this paper, by studying the constructing and learning algorithm of WNN, WNN is applied for nonlinear system identification and signal disposal. The major contents and contributions of this dissertation are as follows:1,The constructing and learning algorithm of WNN are studied. The WNN based on time-frequency and adaptive WNN are studied respectively. The structure and learning algorithm of adaptive WNN are researched in detail, and a new hybrid learning algorithm is proposed. The new algorithm combines the Levenberg-Marquart (LM) algorithm with Grads-Descend(GD) algorithm to improve the tracking accuracy and speed the learning process. The process flow chart of WNN is given, and the simulation results demonstrate the validity of the proposed methods.2,The application of WNN for nonlinear system identification is studied. The WNN proposed in this paper is used to identificate a one-dimension nonlinear system and a two-dimension nonlinear system based on the LM-GD algorithm. A local linear model is adopted instead of the straightforward connection weights between the hidden layer and output layer for more parsimonious hidden units. The simulation results demonstrate the validity of the proposed WNN model and its learning algorithm.3,The application of WNN for electrocardiogram(ECG) beat estimation and classification is studied. A WNN based on sampling theory is introduced. The original ECG signals are first estimated by this WNN in order to remove the high frequency noise, and the estimation results are given. A BP neural network based classifier then follows to classify the ECG signals. The experimental results prove that the proposed method is excellent for computer-aided diagnosis of ECG signals.