The Application Of Time Series Prediction Based Wavelet Neural Network

In medical and health fields, the application of time sequence is very common. Available time series prediction method is linear models such as AR and ARIMA, these models need people to determine the order and type, the subjective factor is relatively large and there is no way to nonlinear models for effective approximation. Through the difference or logarithm transform, the non-stationary sequence could into a stationary sequence, but the capacity of this single function transformation is very limited, or requires a certain skill. Therefore, the traditional method can only deal with some relatively smooth, or using methods can be transformed into stationary time series. But for the data of non-stationary or traditional methods can not be stabilized, the existing method has some limitations.Wavelet neural network is a promising method that can apply to non-stationary data. It is a perfect combinational product of the wavelet analysis and artificial neural network.It has the advantages of wavelet analysis and neural network. On one hand, make full use of the time-frequency localization characteristics of the wavelet transform; on the other hand, give full play to the self-learning ability of neural network. It corresponds to the neural network is introduced two new variables-telescopic factor and translation factor. Not only avoid to inherent defects of neural network, but also together with the local approximation of wavelet analysis. So it has stronger capability of approximation and fault tolerant. Wavelet neural network can be used to solve the data of large non-stationary number, the unknown system that cannot use the formula to describe the mechanism and the problems of the traditional methods can’t solve or have a poor effect.In the medical field, it is rare that the wavelet neural network is applied to forecast the multivariate non-stationary data. The subject use wavelet neural network to solve the multivariate non-stationary time series prediction. By means of the theoretical analysis non-stationary multi factor simulation data, discusses its application conditions, and then be achieved in the software. Prove that wavelet neural network have a good predictive ability for the fluctuation non-stationary time series data and provide new ideas and methods for the time series prediction.The first chapter of this thesis introduced the basic concepts, development and characteristic of wavelet analysis and neural networks. The second chapter describes some traditional methods of time sequence; the development, application and prospect of wavelet neural network in time series. The third chapter simulates examples to illustrate the application of the wavelet neural network.In this paper, Matlab7.0is used to programming and processing the dataset.