基坑施工引起周边管线沉降的一个深度学习模型

国网上海市电力公司经济技术研究院,上海 200233

基坑施工;地下管线沉降;预测;卷积神经网络;长短期记忆神经网络

A Deep Learning Model of Surrounding Pipeline Settlement Caused by Foundation Pit Construction
HU Shengxiang,WANG Shijun

Institute of Economics and Technology,Shanghai Electric Power Company of State Grid, Shanghai 200233, China

Foundation pit construction; Underground pipeline settlement; Prediction; Convolutional neural network;Long-short-term memory based neural network

DOI: 10.13512/j.hndz.2023.02.19

备注

基坑施工会引起周围地下管线沉降是一个动态过程,传统地下管线沉降预测方法不能反映这一动态过程、从而难以真正实现信息化施工。根据某基坑施工过程中周边地下管线沉降的监测数据,使用卷积神经网络(CNN)对监测数据进行特征提取,使用基于长短期记忆(LSTM)的神经网络对沉降进行分析,建立了基于深度学习网络的地下管线沉降预测模型,探讨了超参数和建筑物类型对预测准确度的影响。结果表明,利用所建CNN–LSTM联合人工神经网络,预测值与实际监测值比较接近;网络设置时,初始学习率、隐藏单元数、最大回代轮次可分别取为0.006、240、60;网络预测结果好坏与管线是否是刚性无关,预测值与实测值吻合的高低顺序是电力管线(柔性)、上水管线(刚性)、信息管线(柔性)、电信管线(柔性)。研究成果对基坑工程安全施工具有一定的参考价值。
The settlement of surrounding underground pipelines caused by foundation pit construction is a dynamic process. The traditional prediction method of underground pipeline settlement cannot reflect this dynamic process, so it is difficult to truly realize information construction. Based on the monitoring data of the settlement of the surrounding underground pipeline during the foundation pit construction,the convolutional neural network(CNN) is used to extract the feature of the monitoring data , and the neural network based on long-short-term memory (LSTM)is used to analyze the settlement. The prediction model of underground pipeline settlement based on deep learning network is established, and the influence of hyperparameters and building types on the prediction accuracy is discussed. The results show that the predicted value and the monitoring value are relatively coincident by the using the CNN-LSTM combined with artificial neural network. The initial learning rate,number of hidden units, and the maximum number of iterations can be set as 0.006, 240 and 60, respectively in the network design. The prediction results of the network are independent of whether the pipeline is rigid or not,and the order of the agreement between the predicted value and the measured value is power pipeline(flexible), water supply pipeline(rigid), information pipeline(flexible), telecommunication pipeline(flexible) . The results have a certain reference value for the safe construction of foundation pit engineering
·