经济时间序列的连续参数小波网络预测模型  

Continuous Parameter Wavelet Networks-Based Forecasting Model for Economic Time Series

作　　者：张新红[1] 

机构地区：[1]华侨大学商学院,福建泉州362021

出　　处：《运筹与管理》2007年第2期72-77,共6页Operations Research and Management Science

基　　金：国家自然科学基金资助项目(40271054);国务院侨办基金资助项目(04QSK05)

摘　　要：本文利用连续小波变换方法,给出一种连续参数小波网络。网络参数的学习采用一种类似神经网络的后向传播学习算法的随机梯度算法。另外,提出了一种借助小波分析理论指导网络参数赋初值的方法。进一步,通过对中国进出口贸易额时间序列预测建模的研究和仿真预测,提出了用连续参数小波网络建立经济时间序列预测模型的一般步骤和方法。预测结果表明,此模型具有较好的泛化、学习能力,可以有效地在数值上逼近时间序列难以定量描述的相互关系,所以利用连续参数小波网络建立的时间序列预测模型有较高的预测精度。Based on continuous wavelet transform theory, the paper presents continuous parameter wavelet networks, whose activation function is continuous wavelet function. Parameters of networks are calculated by stochastic gradient algorithm, and initializations of network parameters are made according to wavelet analysis theory. Furthermore, general methods and procedure for economic time series forecasting models are proposed based on continuous parameter wavelet networks, which are used in forecast simulation of the time series for the import-export trade in China. The result from simulation shows that the wavelet networks can effectively approximate to a mutual relationship in value,which is difficult to be formulated quantitatively by time series. The result is encouraging.

关 键 词：数量经济 经济预测 连续参数小波网络 经济时间序列 

分 类 号：G303[文化科学]