基于CNN-LSTM的混凝土坝渗流预测

Seepage Prediction of Concrete Dams Based on CNN-LSTM

建立高性能的混凝土坝渗流预测模型是渗流安全监控的重要手段,也是渗流安全性态评价的基础,结合卷积神经网络(CNN)和长短期记忆神经网络(LSTM)两种深度学习算法,构建混凝土坝渗流预测模型(CNN-LSTM),该模型先利用CNN提取渗流监测时间序列的特征,然后利用LSTM生成特征描述,建立输入与输出间的映射关系,实现对混凝土坝的渗流预测。工程实例应用表明,CNN-LSTM模型在混凝土坝渗流预测应用中的数据拟合能力和预测精度较好,且不易陷入局部最优解,可为混凝土坝的渗流预测和安全监控提供科学依据。

High performance seepage prediction model of concrete dam is a vital procedure for dam seepage monitoring.It is also the foundation of dam seepage evaluation.Combining two deep learning algorithms of convolutional neutal networks(CNN)and long short-term memory(LSTM),a concrete dam seepage prediction model(CNN-LSTM)was constructed.The features of the time series of seepage monitoring were extracted with the CNN.And then the LSTM was used to generate the feature descriptions.The relationship between the inputs and outputs was established to realize the seepage prediction of concrete dams.The applicability of this model was illustrated using an engineering case.The results show that the CNN-LSTM model has good data fitting ability and prediction accuracy in the application of concrete dam seepage prediction,and it can avoid falling into the local optimal solution,which provides a scientific support on the dam seepage prediction and safety monitoring.