计算机工程 ›› 2007, Vol. 33 ›› Issue (17): 23-25.doi: 10.3969/j.issn.1000-3428.2007.17.008

• 博士论文 • 上一篇    下一篇

基于核方法的非线性时间序列预测建模

基于核方法的非线性时间序列预测建模   

  1. (东北大学信息科学与工程学院,沈阳 110004)
  • 收稿日期:1900-01-01 修回日期:1900-01-01 出版日期:2007-09-05 发布日期:2007-09-05

Nonlinear Time Series Prediction Modeling Based on Kernel Method

Nonlinear Time Series Prediction Modeling Based on Kernel Method   

  1. ( College of Information Science and Engineering, Northeastern University, Shenyang 110004)
  • Received:1900-01-01 Revised:1900-01-01 Online:2007-09-05 Published:2007-09-05

摘要: 提出了一种基于核的非线性时间序列预测建模方法。对非线性时间序列的相空间进行重构以确定其嵌入维数,并提出一种基于核主成分分析的非线性时间序列相空间重构方法,针对时间序列的时序特征,采用一种加权的支持向量回归模型对时间序列预测建模。在不同基准数据集上的实验结果表明,与通常的基于普通支持向量回归的建模方法相比,该文所提出的预测建模方法具有较高的精度,说明所提方法对非线性时间序列的预测建模是有效的。

关键词: 核主成分分析, 支持向量回归, 相空间重构, 时间序列建模

Abstract: The paper presents a kernel based prediction modeling method for nonlinear time series. Phase space of time series is reconstructed to get its embedding dimension, and a method of phase space reconstruction based on kernel principal component analysis(KPCA) is proposed. A weighted support vector regression(SVR) is adopted to set up prediction model according to the characteristics of time series. The experimental results on different benchmark data show that the model based on the proposed method has higher accuracy compared with normal SVR model, proving the efficiency of the method for nonlinear time series prediction modeling.

Key words: kernel principal component analysis(KPCA), support vector regression(SVR), phase space reconstruction, time series modeling

中图分类号: