基于LSTM神经网络的地铁车站改造沉降时间序列预测  

Time Series Prediction of Subway Station Reconstruction Settlement Based on LSTM Neural Network

作　　者：刘俭 Liu Jian(The Fourth Engineering Co.,Ltd.of China Railway 18th Bureau Group,Tianjin 300000,China)

机构地区：[1]中铁十八局集团第四工程有限公司,天津300000

出　　处：《市政技术》2023年第5期143-148,共6页Journal of Municipal Technology

摘　　要：为了预测地铁车站改造施工过程中出入口站厅的沉降,提出了采用LSTM神经网络预测沉降的方法。依托北京市西土城地铁车站改造工程,对该车站改造过程中的建筑物沉降值进行了预测,并结合实测值利用LSTM神经网络对沉降趋势进行了回归分析。研究结果表明:采用LSTM神经网络进行沉降预测有着一定的合理性,其可以较好地捕捉沉降变化的趋势,该模型的预测值与实测值误差在10%以内,具有很好的应用价值。In order to predict the settlement of the entrance and exit halls during the renovation construction of the subway station,a method of settlement prediction by long short term mermory(LSTM)neural network was proposed.Based on the renovation project of Xitucheng subway station in Beijing,the building settlement values during the renovation of the station were predicted.Combination with the measured values,the settlement trends were regressed by LSTM neural network.The results show that the settlement prediction by LSTM neural network is reasonable which can capture the settlement trend well.The error between the predicted and measured values is within 10%,which is of good application value.

关 键 词：LSTM 神经网络 非线性时间序列 沉降预测 

分 类 号：U457[建筑科学—桥梁与隧道工程] TU196.2[交通运输工程—道路与铁道工程]