基于小波网络的非线性经济时序预测模型

为对经济时序准确预测，必须先对其数据结构进行分析，相空间重构技术为之提供了理论基础，通过关联维数的计算，区分确定性系统和随机系统．在此基础上确定最佳嵌入维数、最佳采样时间间隔及小波元的个数，并通过带有偏差单元的递归小波网络的学习，进行模型参数的辨识．实验研究表明，模型对非线性经济时序具有良好的逼近能力，因此该模型用于非线性经济时序预测具有可行性。

Wavelet network-based non-linear economic time series prediction model

For accurate prediction of economic time series, the data structure analysis should be made, and the phase space technique provides the theoretical basis for the analysis. Through correlation dimension calculation, the determinate system and random system are classified, and the optimum embedded dimension, the optimum sampling time interval, and the number of wavelet units are further determined. Then by training of recursive wavelet networks with deviation units, the parameters of the model are identified. Experimental study shows that the model is of good behavior in approximation of nonlinear economic time series. The good agreement between experimental results and theoretical analysis validates the feasibility of the model applied to prediction of non-linear economic time series.