基于回声状态网络的多变量预测模型的研究

摘要

考虑单变量在混沌时间序列预测中的不足,文章利用多变量模型进行混沌时间序列的预测.针对多变量预测过程中的维数过高问题,文章结合主元分析理论(PCA)和回声状态网络(ESN),构建了基于PCA和ESN的多变量混沌时间序列预测模型,将PCA降维后的时间序列数据输入ESN网络进行预测分析.论文对由Lorenz动态方程生成的三变量混沌时间序列进行了仿真实验,结果表明该模型有效地提高了预测的精度和预测的效率,是一种有效的混沌时间序列预测方法.%Considering the shortages in the prediction of chaotic time-series using single variable, this paper studies a new multivariate chaotic time-series prediction model, which is based on the principal components analysis (PCA) and echo state networks (ESN). This new prediction model extracts the joint information of multiple variables by PCA method firstly, then uses the ESN to make predictions. Finally,the simulation example, which is from the typical Lorenz equation, shows that the new model will obtain the well performance both in the precision and speed.