引用本文:韩敏,王新迎.多元混沌时间序列的加权极端学习机预测[J].控制理论与应用,2013,30(11):1467~1472.[点击复制]
HAN Min,WNG Xin-ying.Multivariate chaotic time series prediction based on weighted extreme learning machine[J].Control Theory and Technology,2013,30(11):1467~1472.[点击复制]
多元混沌时间序列的加权极端学习机预测
Multivariate chaotic time series prediction based on weighted extreme learning machine
摘要点击 3191  全文点击 2551  投稿时间：2013-03-06  修订日期：2013-05-22
查看全文  查看/发表评论  下载PDF阅读器
DOI编号  10.7641/CTA.2013.30170
  2013,30(11):1467-1472
中文关键词  时间序列  预测  极端学习机  支持向量机  样本加权
英文关键词  time series  prediction  extreme learning machine  support-vector machines  weighted instances
基金项目  国家自然科学基金资助项目(61074096).
作者单位E-mail
韩敏* 大连理工大学 电子信息与电气工程学部 minhan@dlut.edu.cn 
王新迎 大连理工大学 电子信息与电气工程学部  
中文摘要
      针对多元混沌时间序列具有强非线性, 难以建立数学模型进行准确预测的问题, 本文提出一种加权极端学习机预测算法. 首先对多元混沌时间序列进行相空间重构, 并根据相空间中输入数据对预测误差的影响施加不同的权重. 然后, 提出一种支持向量极端学习机预测模型, 具有支持向量机的核映射表达能力以及极端学习机的一步快速训练能力, 因此训练简便且具有较好的泛化性能. 所提算法具有和训练样本三次方成正比的计算复杂度, 因此适用于10^2～10^3样本规模的平稳时间序列. 基于Lorenz混沌时间序列和年太阳黑子和黄河年径流混沌时间序列预测的仿真结果证明所提算法的有效性.
英文摘要
      Considering the strong nonlinear property of the multivariate chaotic time series and difficulties in mathematical model building for accurate prediction, we propose a weighted extreme learning machine (ELM) prediction model. Firstly, the multivariate chaotic time series are reconstructed in the phase space, and the instances in prediction window are weighted according to their influence on the prediction errors. A support-vector extreme learning machine prediction model is proposed, which combines both the kernel mapping ability of support-vector machines and the one-step fast training advantage of extreme learning machine, so that it is easy to conduct and has good generalization performance. The computational complexity of the proposed algorithm is proportional to the cube of the training sample. Therefore, it is suitable for stationary time series with 10^2～10^3 sample size. The effectiveness of the proposed model are demonstrated by simulation results based on Lorenz chaotic time series, annual sunspots, and the runoff of the Yellow River chaotic time series.