基于储备池主成分分析的多元时间序列预测研究  

Prediction of multivariate time series based on reservoir principal component analysis

作　　者：韩敏[1] 王亚楠[1] 

机构地区：[1]大连理工大学电信学院,辽宁大连116024

出　　处：《控制与决策》2009年第10期1526-1530,共5页Control and Decision

基　　金：国家自然科学基金项目(60674073);国家863计划项目(2007AA04Z158);国家科技支撑计划项目(2006BAB14B05);国家973计划项目(2006CB403405)

摘　　要：提出一种基于回声状态网络储备池的非线性PCA方法,并将其应用于多元时间序列的预测中.由于多维输入变量间的相关性会影响建模效果,通过储备池将输入在原空间的非线性特征转化成高维空间的线性特征.在其中运用线性PCA技术寻找输入在储备池空间的最大方差方向,提取有效的多元变量综合信息.经储备池主成分分析处理后的输入与预测点呈动态线性映射,可使用线性方法建模.仿真结果表明了该方法的有效性.A nonlinear principal component analysis （PCA） method based on echo state network （ESN） reservoir is proposed for multivariate time series prediction. As the correlations among multivariate inputs have adverse effect on modeling, the reservoir is utilized for translating the inputs ＇ nonlinear features in the original input space to linear features in the high-dimension reservoir state space. Then the linear PCA is performed on mapped data to find the uncorrelated directions of maximum variance and thus to extract the joint information of multiple variables. Furthermore, dynamic linear mapping is obtained from the RPCA outputs to the predictor points, so linear algorithms are investigated for modeling. The simulation results show the effectiveness of the proposed method.

关 键 词：储备池主成分分析 回声状态网络 多元时间序列 预测 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]