关键词：小波神经网络； 方差分析； 时间序列； 预测
中图分类号：TP274 文献标志码：A 文章编号：1001-3695(2008)08-2366-03
Research about time-serial prediction based on tight wavelet neural-net
LUO Hang, WANG Hou-jun, LONG Bing
(School of Automation Engineering, University of Electronic Science & Technology of China, Chengdu 610054, China)
Abstract:This paper analyzed the characteristic and construction of wavelet neural-net called tight type. At the same time, it decided the length of sample serial by using a kind of statistic method called variance analysis so that the knot number of input layer could be efficiently decided, which was based on analyzing time-serial prediction by means of neural net. It trained and predicted the time-serial of sunspot’s annual average activity and compared the difference of training and prediction to the time-serial between wavelet-net and BP-net from net itself restriction. Accordingly, analyzed the cause of that difference. The method of combining wavelet’s multi-discrimination with neural-net’s non-linear approaching function exerted their advantages respectively. Thus, the prediction precision was obviously enhanced.......