基于CNN-LSTM-AM的大坝变形预测  

Dam Deformation Prediction Based on CNN-LSTM-AM

作　　者：赖国梁 刘小生[1] LAI Guo-liang;LIU Xiao-sheng(School of Civil and Surveying Engineering,Jiangxi University of Science and Technology,Ganzhou 341000,China)

机构地区：[1]江西理工大学土木与测绘工程学院,江西赣州341000

出　　处：《水电能源科学》2024年第10期158-161,157,共5页Water Resources and Power

基　　金：国家自然科学基金项目(42171437)。

摘　　要：为提高大坝变形预测模型的预测精度,以长短期记忆(LSTM)作为基础模型预测大坝变形,在LSTM网络层前加入卷积神经网络(CNN)卷积层,以卷积层中卷积核刻画数据的局部模式实现数据特征的深度挖掘,提取大坝变形多因素序列时空特征;LSTM网络层后加入注意力机制层用于区分特征信息的重要程度并给予不同的关注度,进一步优化网络模型,构建了基于CNN-LSTM-AM的大坝预测模型。应用该大坝预测模型在工程实例中与LSTM、CNN-LSTM、LSTM-AM模型的预测结果和残差进行对比分析,CNN-LSTM-AM模型的预测结果和拟合度均更优;并以均方误差、均方根误差、平均绝对误差及决定系数R2作为精度评定指标对比各模型间预测性能,结果表明引入注意力机制能够提升模型预测性能,证实了基于CNN-LSTM-AM构建的大坝预测模型具有工程应用价值。In order to improve the prediction accuracy of the dam deformation prediction model,LSTM was used as the base model to predict the dam deformation,convolutional layer of CNN network was added before the LSTM network layer,and the convolutional kernel in the convolutional layer portrayed the local patterns of the data to realize the deep mining of the data features so as to extract the multifactorial sequential spatial and temporal features of the dam deformation.An attention mechanism layer was added after the LSTM network layer for distinguishing the importance of feature information and giving it different levels of attention to further optimize the network model.A dam prediction model based on CNN-LSTM-AM was constructed.Compared with the prediction results and residuals of the LSTM,CNN-LSTM,and LSTM-AM models in engineering examples,the CNN-LSTM-AM model has better prediction results and goodness-of-fit.EMSE,ERMSE,EMAE,and R2 were used as accuracy evaluation indexes to compare the prediction performance among models,indicating that the introduction of the attention mechanism can improve the model prediction performance,and confirming that the dam prediction model based on CNN-LSTM-AM has engineering application value.

关 键 词：卷积神经网络 长短期记忆网络 注意力机制 大坝变形预测 预测精度 

分 类 号：TV698.1[水利工程—水利水电工程]