基于小波和神经网络模型的邮电业务总量预测

Business Total of Posts and Telecommunications Prediction Based on Wavelet and Artificial Neural Network Model

引入极大重叠离散小波变换的概念,利用极大重叠离散小波变换的多分辨分析特性,对邮电业务总量序列进行分解.然后对分离得到的光滑项和细节项两部分利用小波神经网络模型进行建模和预测,最后再重构得到邮电业务总量序列的预测值.数据测试结果表明:本文方法可实现多步预测,且对邮电业务总量的预测精度比单纯的用小波神经网络模型或BP神经网络模型高.

This paper introduces the concept of Maximal Overlap Discrete Wavelet Transform （MODWT） for business total of posts and telecommunications prediction. The characteristic of MODWT multi - resolution a- nalysis is used to decompose the business total of posts and telecommunications. Then the smooth term and de- tails term are separated from business total of posts and telecommunications series. The smooth term and de- tails term are modeled and predicted by applying wavelet neural network model. Finally, the predictive value of business total of posts and telecommunications series is obtained by restructuring. The results tested with da- ta indicate： this forecasting method introduced in this paper can realize multistep prediction and the forecasting accuracy is higher than the way that business total of posts and telecommunications is predicted just by wavelet neural network model or BP neural network model.