基于卷积神经网络-长短期记忆的施工期盾构管片上浮过程预测模型

Prediction Model of Shield Segment Floating Process During Construction Based on Convolutional Neural Networks and Long Short-Term Memory

摘要:

为了实现施工期盾构管片上浮过程的智能预测，采用动力水准仪对施工期盾构管片上浮过程进行自动化监测并建立了基于卷积神经网络?长短期记忆（CNN-LSTM）深度学习算法的管片上浮过程智能预测模型。结果表明：管片上浮阶段呈现出“阶梯状”，即管片上浮主要发生在盾构掘进期间，且掘进状态的上浮量最大，占峰值的75.24 %~98.29 %；CNN-LSTM模型对施工期盾构管片上浮过程具有较好的预测效果，在训练集上的均方误差MSE、平均绝对误差MAE和决定系数R2分别为0.038 7、0.148 2和0.999 3，在测试集上为0.030 7、0.138 9和0.801 9；相较于反向传播（BP）模型，CNN-LSTM模型在训练集与测试集上的性能均有所提升，且测试集的提升更明显，最高可达89.71 %。研究结果可为盾构管片上浮的现场实测及预防处治提供新思路。

关键词:盾构隧道;管片上浮;上浮过程;智能预测;自动监测

Abstract:

To realize the intelligent prediction of the floating process of shield segment during construction， the dynamic level was used to automatically monitor the floating process of shield segment， and an intelligent prediction model of segment floating process based on convolutional neural networks and long short-term memory （CNN-LSTM） deep learning algorithm was established. The results show that the floating stage of segment is in a shape of “ladder”， i.e.， the floating of segment mainly occurs during shield tunneling. In addition， the floating amount in the heading state is the largest， accounting for 75.24 % to 98.29 % of the peak value. The mean square error （MSE）， average absolute error （MAE） and determination coefficient R2on the training set are 0.038 7， 0.148 2， and 0.999 3 respectively， and 0.030 7， 0.138 9 and 0.801 9 on the test set. Compared with the back propagation （BP） model， the performance of the CNN-LSTM model in the training set and test set has been improved， and the improvement of the test set is more obvious， up to 89.71 %. The research results can provide a new idea for field measurement， prevention， and treatment of shield segment floating.

Key words:shield tunnel;segment uplift;floating process;intelligent prediction;automatic monitoring