Comparison Of Research And Application Of Wavelet Neural Network And Bp Network

As an important branch of nonlinear science, neural net-work provides us a strongtool to research nonlinear problems. Presently, feed-forward neural net-work,especially BP net-work, is most widely used in practice. Wining many successesalthough, they have much shortage at the same time. Wavelet neural-network (WNN),which is put forward recently, has been proved better than BP net-work in many facets,especially in the fields of forecasting.The paper profoundly compared the performance between BP and WNN in theoryand practice, analyzed each network's functional characteristics.Theoretically, the theory contains comparison and analysis of the convergencebetween BP and WNN (including convergence speed, stabilization and so on), causesof "generalization problem" in the application of artificial neural net-work(ANN) andeach generalization characteristic of BP and WNN.Practically, the paper discussed technical problems when ANN is used inforecasting. The paper researched the affect that new sample brings to the old ones,offered a relative guess. Based on these, the paper put forward a method to avoid thisnegative affect, thus improved the quality of net-work's prediction.In the end, combining with the practical examples, the paper elaborated on how toset a mathematic model using WNN, and the model's characteristic, pointed out eachadvantageous field of BP and WNN when used in prediction.