基于多变量相空间重构的地下水动态神经网络预测模型  

Neural network model for groundwater prediction based on multivariate phase space reconstruction

作　　者：刘卫林[1] 董增川[1] 陈南祥[2] 董贵明[1] 

机构地区：[1]河海大学水文水资源与水利工程科学国家重点实验室 [2]华北水利水电学院岩土工程系,河南郑州450008

出　　处：《河海大学学报（自然科学版）》2008年第1期14-18,共5页Journal of Hohai University(Natural Sciences)

基　　金：教育部科学技术研究重点项目(104197)

摘　　要：从多元变量时间序列出发,以相空间重构理论为基础,结合偏最小二乘回归与神经网络方法,提出了基于多变量相空间重构地下水动态预测的神经网络模型.通过分别确定各个时间序列的时间延迟和嵌入维数的方法对地下水动态的多变量时间序列进行相空间重构,采用偏最小二乘回归法提取对系统解释最强的成分作为网络输入,利用神经网络模型进行各成分之间的非线性拟合.将基于多变量相空间重构地下水动态预测的神经网络模型应用于黑龙洞泉域地下水位预测,并通过分析与比较验证了该方法的有效性.Based on the phase space reconstruction theory for multivariate time series, a neural network model for groundwater prediction was put forward by combination with the neural network and partial least square regression methods. With the model, the phase space reconstruction for multivariate time series of groundwater was performed by determination of the time delay and embedded dimensions of each time series. By use of the partial least square method, the components which could reflect the characteristics of the system to the greatest extent were extracted and taken as the input of the network, and the neural network model was used for nonlinear fitting of each component. The validity of the model was verified by its application to groundwater prediction for Heilongdong spring area.

关 键 词：地下水动态 相空间重构 神经网络 偏最小二乘回归 预测模型 

分 类 号：TV211.1[水利工程—水文学及水资源]