五强溪水电站大坝变形预测模型研究  

Study on Deformation Prediction Models of Wuqiangxi Dam

作　　者：韩行进 杨松林 匡楚丰 陈冰 张健飞[3] 余天堂[3] HAN Xingjin;YANG Songlin;KUANG Chufeng;CHEN Bing;ZHANG Jianfei;YU Tiantang(Hunan Wuling Power Corporation,Changsha 410004,Hunan,China;Wuling Power Corporation Ltd.,Changsha 410004,Hunan,China;Department of Engineering Mechanics,Hohai University,Nanjing 211100,Jiangsu,China)

机构地区：[1]湖南五凌电力科技有限公司,湖南长沙410004 [2]五凌电力有限公司,湖南长沙410004 [3]河海大学工程力学系,江苏南京211100

出　　处：《水力发电》2022年第4期111-116,共6页Water Power

摘　　要：基于逐步回归法、偏最小二乘回归法和长短期记忆(LSTM)循环神经网络,构建了五强溪水电站大坝变形预测模型。采用拉伊特准则确定可靠的监测数据,基于可靠的监测数据,构建考虑水压、温度、时效因素的混凝土重力坝变形预测逐步回归和偏最小二乘回归模型,根据五强溪大坝坝顶J23测点2006年~2020年的监测资料获得该测点的沉陷曲线逐步回归和偏最小二乘回归预测模型。根据数值试验,选定的LSTM模型包括2个LSTM层,激活函数采用整流线性单元函数,输入序列长度为20。训练集数据取2006年~2017年的监测值,2018年~2020年的监测数据作为测试集数据。采用随机搜索对LSTM循环神经网络的超参数进行优化。比较3种模型结果可知:3种模型在沉降曲线的预测效果均较好;偏最小二乘回归法能合理地解释各分量;训练数据足够时,LSTM循环神经网络的预测精度非常高;采用偏最小二乘法回归模型或LSTM模型预测J23测点变形更为妥当。The deformation prediction models of Wuqiangxi Dam are developed using stepwise regression method,partial least squares regression method and long short-term memory(LSTM)recurrent neural network.The reliable monitoring data are determined with Lahitte criterion.The stepwise regression and partial least squares regression models for deformation prediction of concrete gravity dam are constructed based on the reliable monitoring data,and the factors of water pressure,temperature and time effect are considered in the models.According to the monitoring data from 2006 to 2020 of measuring point J23 located on the crest of Wuqiangxi Dam,the stepwise regression and partial least squares regression models of the settlement curve of the measuring point are obtained.In the LSTM model,two LSTM layers are used,the rectified linear unit function is adopted as the activation function,and the input sequence length is 20.The monitoring data for measuring point J23 from 2006 to 2017 are selected as the training set,and the monitoring data from 2018 to 2020 are taken as the test set.The random search is adopted to optimize the hyper-parameters in the LSTM model.Based on the results of three models,it is found that,(a)the good prediction results of settlement curve can be obtained with the three models;(b)the partial least squares regression model can reasonably explain the factors;(c)the prediction accuracy of the LSTM model is very high when there are enough training data;and(d)the partial least squares regression model or the LSTM model is suggested to predict the deformation of measuring point J23.

关 键 词：大坝变形 逐步回归法 偏最小二乘回归法 LSTM循环神经网络 预测模型 五强溪水电站 

分 类 号：TV698.1[水利工程—水利水电工程] TV741