基于LSTM的深基坑开挖地表沉降预测研究

Study on prediction model of ground subsidence in deep foundation pit excavation based on LSTM

针对传统的BP神经网络在机械学习过程中容易出现的运行速度慢、易过度拟合和与实测值误差较大等问题,提出了一种基于循环神经网络(RNN)的长短期记忆人工(LSTM)神经网络模型。通过广州市某深基坑开挖的周围地表沉降监测数据进行LSTM神经网络训练,并对后续地表沉降值进行预测,预测结果表明:相比于传统的BP神经网络,LSTM神经网络具有较高的预测精度,可较好的实现基坑开挖过程中深基坑周围地表沉降预测。研究成果可为深基坑施工过程中周围土体变形监测和预警提供参考。

In order to solve the problems of the traditional BP neural network in the process of mechanical learning,such as slow running speed,easy to overfit and large error with the measured value,this paper proposes a kind of LSTM neural network model based on the recursive neural network(RNN).Around a deep foundation pit excavation by Guangzhou LSTM neural network training,the surface subsidence monitoring data and the subsequent deep excavation surface subsidence prediction,prediction results show that compared with the traditional BP neural network,the LSTM neural network has higher prediction accuracy,can better realize the ground settlement around deep foundation pit excavation process.The research results can provide reference for the deformation monitoring and early warning of the surrounding soil in the process of deep foundation pit construction.