PCA-GA-LM模型在水库土石坝渗流预测中的应用  

Application of PCA-GA-LM Model for Predicting the Seepage of Reservoir Earth-rockfill Dams

作　　者：缪新颖 单玉鹏 纪建伟[3] MIAO Xin-ying;SHAN Yu-peng;JI Jian-wei(School of Information Engineering,Dalian Ocean University,Dalian Liaoning 16023,China;School of Food Science and Engineering,Dalian Ocean University,Dalian Liaoning 16023,China;College of Information and Electrical Engineering,Shenyang Agricultural University,Shenyang 110161,China)

机构地区：[1]大连海洋大学信息工程学院,辽宁大连116023 [2]大连海洋大学食品科学与工程学院,辽宁大连116023 [3]沈阳农业大学信息与电气工程学院,沈阳110161

出　　处：《沈阳农业大学学报》2021年第3期377-384,共8页Journal of Shenyang Agricultural University

基　　金：辽宁省科技重大专项计划项目(2020JH1/10200002);辽宁省科技计划项目(20180550573);辽宁省教育厅科学研究项目(JL201918)。

摘　　要：水库是农业生产用水、灌溉农作物的重要水利工程之一,水库大坝安全与否决定着水库能否正常运转。针对大坝渗流预测模型需要解决的有效影响因子确定、网络结构优化和预测算法等问题,将主成分分析(principal component analysis,PCA)方法、遗传算法(genetic algorithm,GA)和Levenberg-marquardt(LM)神经网络协同应用于水库大坝渗流预测,提出PCA-GA-LM大坝渗流预测模型。利用PCA确定大坝渗流的影响因子,实现各因素之间的去耦和降维,避免多重共线性,使尽可能少的变量包含尽可能多的信息;利用GA优化网络结构,确定合适的隐层节点数目和权值;利用LM算法训练神经网络,提高神经网络的泛化能力和收敛速度。为了验证新方法的预测效果,以监测渗流的测压管水位为研究对象,以大伙房水库土石坝2018年的365组观测数据为训练样本,选取2019年15组数据作为测试样本,对土石坝渗流进行预测,并将预测结果与未采用PCA的GA-LM模型进行了比较。对比结果表明:PCA-GA-LM模型预测的土石坝渗流值与实测值吻合较好,平均误差、标准偏差和平均相对误差都较小,说明采用该方法预测准确率较高,预测效果优于试验法。研究结果表明了基于PCA-GA-LM模型可以作为水库土石坝渗流预测的一种有效手段,对于进一步研究水库大坝安全具有推动作用。Reservoir is one of the important water conservancy projects for agricultural production and irrigation of crops,and the safety of dams is the basic guarantee for the normal operation of the reservoirs.When establishing the dam seepage prediction model,it is necessary to solve several problems such as the determination of effective influence factors,the optimization of network structure and prediction algorithm.An earth-rockfill dam seepage prediction model called PCA-GA-LM is proposed,which is developed by principal component analysis(PCA)method in coordination with genetic algorithm(GA)and Levenberg-marquardt(LM)neural network.PCA is used to determine the influence factors of the earth-rockfill dam seepage,which can decouple and reduce the dimension of influence factors,avoid multicollinearity and contain as much information as possible with as few variables as possible.GA is utilized to optimize the network structure and determine the appropriate number of hidden layer nodes and weight values.LM is applied to train the neural network,which can improve the generalization ability and convergence speed.In order to verify the prediction effect of the new method,the piezometric levels are taken as the research objects,and PCA-GA-LM has been developed for predicting the seepage of the earth-rockfill dam of Dahuofang reservoir using 380 sets of field data(of which 365 in 2018 were used for training and 15 in 2019 for testing).Then the predictive results of PCA-GA-LM have been compared with that of GA-LM without PCA.The comparison results showed that the piezometric levels predicted by PCA-GA-LM were in good agreement with the measured values,and the mean errors,the standard deviations and the mean relative errors are all smallest.It is proved that the PCAGA-LM can give superior predictions over the testing method,and it is an effective method to predict the seepage of earth-rockfill dams.This research will promote the further study of dam safety.

关 键 词：土石坝渗流 主成分分析(PCA) 遗传算法(GA) Levenberg-marquardt(LM)算法 预测模型 

分 类 号：S279.2[农业科学—农业水土工程] TP183[农业科学—农业工程]