基于小波和过程神经网络的时间序列预测模型  

Time series prediction model based on wavelet and process neural network

作　　者：刘修恒[1] 吕鑫[1] 戚荣志[1] 刘璇[1] 马鸿旭[1] 

机构地区：[1]河海大学计算机与信息学院,江苏南京211100

出　　处：《电子设计工程》2016年第1期9-11,15,共4页Electronic Design Engineering

基　　金：国家自然科学基金面上项目(61272543);国家科技支撑计划(2013BAB06B04);中国华能集团公司总部科技项目(HNKJ13-H17-04);江苏省自然科学基金(BK20130852);江苏省博士后科研资助计划(1401001C)

摘　　要：提出基于过程神经网络和自回归模型的组合预测方案。首先,采用二进正交小波变换对原始时间序列分解和重构,分离出原始序列中的高频部分和低频部分;然后对低频部分构建过程神经网络模型,对低频部分采用自回归模型;最后将两种模型的预测值叠加,得到原序列的预测值。A time series prediction method based on AR and process neural network PPN-AR was proposed. By Binary orthogonal wavelet transform, the non-stationary time series were decomposed into a low frequency signal and several high frequency signals. The low frequency was predicted with process neural network, and the high frequency signals were predicted with auto-regression models. The prediction result of the original time series was the superimposition of the respective prediction.

关 键 词：小波分解 非平稳时间序列 过程神经网络 自回归 预测 

分 类 号：TN-9[电子电信]