摘要
提出了一种多变量混沌时间序列的联合熵扩维法(JEED),为多变量时间序列的预测构造了有效的模型输入向量.首先使用互信息法求混沌系统各观测变量的延迟时间;然后使用联合熵确定各分量的嵌入维数,并按最大熵选择重构分量,不断扩张相空间维数,最终使得重构向量空间包含系统的最大信息量.仿真实验表明因为JEED确定的相空间能提供丰富的信息,在其上进行的神经网络预测取得了比单变量预测方法更好的预测效果.
In order to obtain the effective input vector for the prediction,method of joint entropy extending dimension（JEED） is proposed in the reconstructed phase space of multivariate time series.First,it selects delay time of any variable time series by mutual information method,then,determines the embedding dimension of every variable by the joint entropy.It could contain more information of system from low space to high space because it chooses the largest joint entropy.In the end,the authors did chaotic system＇s numerical simulations,simulations of the Lorenz system and power load show that the neutral network prediction of multivariate time series in the phase space,whose dimension is determined by the joint entropy,is much better than the prediction of univariate,because this phase space can provide more information.
出处
《西南师范大学学报（自然科学版）》
CAS
CSCD
北大核心
2011年第4期83-87,共5页
Journal of Southwest China Normal University(Natural Science Edition)
基金
重庆市教委科技项目(KJ111106)
东南大学基本科研业务费创新基金(3207010501)
关键词
多变量时间序列
联合熵
嵌入维
预测
multivariate time series
joint entropy
embedding dimension
prediction
