Wavelet transformation has been used in many fields such as image, communication, earthquake, radar and turbulent because its good time-frequency locality, projection pursuit is a new statistical methods, it can solve the question of "dimension curse" by. On this basis, we first proposed a new type of neural network -projection pursuit wavelet neural network(PPWNN). Then, its mathematical model, topology construction , learning principles and convergence rate are researched.The main contribution of this paper are described as follows:1. mathematical model and topology construction of projection pursuit wavelet neural network (PPWNN) and its non-linear learning principles are obtained.2. the approximation property of projection pursuit wavelet neural network (PPWNN) which is applied to non-linear function is studied, the convergence rate is given in this paper also.3. we demonstrate projection pursuit wavelet neural network(WPPNN) has a good applicability by the approximation of five non-linear functions and the prediction of sunport and chaos time series and the use of edge detection. |