基于时空关联特征的CNN-LSTM模型在基坑工程变形预测中的应用
Application of CNN-LSTM model based on spatiotemporal correlation characteristics in deformation prediction of excavation engineering
-
摘要: 为了更准确地预测基坑工程领域的复杂时间序列,提出一种以多个监测点的监测数据构成的多维时间序列作为输入的CNN-LSTM的组合神经网络模型。首先采用卷积神经网络（CNN）对输入的监测数据进行空间特征提取,输出多个由空间特征构成的时间序列,利用长短期记忆神经网络（LSTM）对空间特征序列进行学习,预测未来的特征值,最后通过全连接层整合空间特征,输出预测的监测值。在此基础上,基于上海云岭竖井超深基坑的现场地表沉降监测数据进行工程案例验证,结果表明考虑时空关联性的组合模型精度高于仅考虑时间关联性的单一LSTM模型。Abstract: In order to predict the complex time series in excavation engineering field more accurately, a combined neural network model of CNN and LSTM is proposed, which takes the multi-dimensional time series composed of monitoring data from multiple monitoring points as the input. Firstly, the CNN is used to extract the spatial features of the input monitoring data, and the multiple time series composed of spatial features are the output. Secondly, the LSTM is used to learn the time series and predict the future state of the spatial features. Finally, the spatial features are integrated through the fully connected layer, and the predicted monitoring values are the output. This method is used to predict the ground settlement of the deep excavation of Yunling shaft in Shanghai. The results show that the accuracy of the combined model considering temporal and spatial correlation is higher than that of the single LSTM model considering temporal correlation only.
下载: