基于样本数据重抽样的时序数据预报方法  

Using Resampling Technique to Make Prediction of Time Series

作　　者：郝善勇[1] 刘玉树[1] 

机构地区：[1]北京理工大学计算机科学与工程系,北京100081

出　　处：《北京理工大学学报》2000年第5期581-584,共4页Transactions of Beijing Institute of Technology

基　　金：部级预研项目

摘　　要：研究时序数据预报和提高预报精度的方法 .提出一种新的利用误差项对时序数据样本进行 Boot Strap重抽样的方法 .该方法采用神经网络技术建立时序数据预报模型 ,并通过重抽样技术提高预报精度 .通过 Boot Strap算法与 BP算法的预报偏差平方和比较说明 Boot Strap算法提高了预报精度 .将提出的重抽样技术引入时序数据预测中 ,可提高神经网络的预测精度 。The method of predicting time series and the method of improving the accuracy of prediction were studied. It proposed a new way to resample time series based on residues. ANN model was used to analyze time series and BootStrap method was used to improve the precision of prediction. By comparing the results of BootStrap method with those of BP algorithm it was elucidated that BootStrap is applicable. If BootStrap method is introduced into time series analysis, the precision of ANN can be improved. In addition, this method could be utilized in the fields of foreign exchange trading and the prediction of stock price.

关 键 词：时序数据预报 重抽样 神经网络 样本数据 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]