基于小波网络的非线性组合预测方法研究

提出了一种基于小波网络的非线性组合预测新方法,以克服线性组合预测方法在解决非平衡时间序列组合建模问题所遇到的困难和存在的不足,并给出了相应的学习算法求解小波函数线性组合的尺度和时延参数以及神经网络权值。理论分析和大量的应用实例表明：本方法具有很强的泛化能力与自适应数据和函数变化的能力,在处理诸如经济时间序列这种具有一定程度不确定性的非线性系统的组合建模和预测方面有较高的应用价值。

Research on nonlinear combination forecasting method based on wavelet network

Inthis paper, a new nonlinear combination forecasting method base d on wavelet network ispresented to overcome the difficulties and drawbacks in combined modeling non-stationarytime serise by using linear combination foreca s ting. Furthermore, corresponding learningalgorithm is put forward to calculate the dilation and shifts which are the parameters ofwavelet function and the we ights of neural network structure. Theoretical analysis andforecasting examples all show that the new technique has strong universalized capabilitiesan d the ability to adapt to the changes of function. With respect to combined modelingand forecasting of economic time series in nonlinear syste ms, which have someuncertainties, the method are feasible and effective.