马铭骏, 潘泓, 骆冠勇,等.基于DNN的盾构施工地层横向水平变形预测[J].河南理工大学学报（自然科学版）,2023,42(1):201-208.

MA M J, PAN H, LUO G Y,et al.Prediction of lateral horizontal deformation of surrounding strata caused by shield construction based on deep neural network[J].Journal of Henan Polytechnic University(Natural Science) ,2023,42(1):201-208.

基于DNN的盾构施工地层横向水平变形预测

马铭骏1,2, 潘泓1,2, 骆冠勇1,2, 曹洪1,2

1.华南理工大学 土木与交通学院，广东 广州  510640；2.华南理工大学 亚热带建筑科学国家重点实验室，广东 广州  510640

摘要:为了得到盾构施工地层横向水平变形的有效预测模型，提出一种考虑主要盾构施工参数的深度神经网络（deep neural networks，DNN）模型。由广州某盾构工程得到盾构施工引起的周边地层水平变形数据，利用现场实测数据和主要盾构施工参数对模型进行训练并结合遗传算法优化DNN模型网络拓扑结构。结合遗传算法计算80个模型后，最终确定DNN模型在训练过程中均方误差MSE经50轮迭代即收敛并趋于零值，模型预测值和实测值基本吻合（拟合优度R2 >0.9），且残差随机分布于零值线附近，说明该模型预测效果良好。研究结果可为类似工程构建水平变形预测模型提供参考。

关键词:盾构施工;现场实测法;深度神经网络;横向水平变形;变形预测模型

doi:10.16186/j.cnki.1673-9787.2021010100

基金项目:国家自然科学基金资助项目（51978282）；广东省自然科学基金资助项目（2018A030313383，2020A1515010583）

收稿日期:2021/01/21

修回日期:2021/04/23

出版日期:2023/01/25

Prediction of lateral horizontal deformation of surrounding strata caused by shield construction based on deep neural network

MA Mingjun1,2, PAN Hong1,2, LUO Guanyong1,2, CAO Hong1,2

1.School of Civil Engineering and Transportation，South China University of Technology，Guangzhou  510640，Guangdong，China；2.State Key Laboratory of Subtropical Building Science，South China University of Technology，Guangzhou  510640，Guangdong，China

Abstract:In order to obtain an effective prediction model for the horizontal deformation of the strata under shieldconstruction，a deep neural network（DNN） model considering the main construction parameters was proposed.The lateral horizontal deformation data of surrounding strata caused by shield construction was obtained from a certain project in Nansha District，Guangzhou.A deep neural network model optimized with genetic algorithm，trained with the field measured data and the main construction parameters，was established.After training 80 models with genetic algorithm，the mean square error（MSE） of DNN model converged to zero in 50 rounds of iteration in the training process and the residuals randomly distributed near the zero line.Moreover，the predicted value was basically consistent with the measured value and determination coefficients R2 were above 0.9.The study results could provide a reference for the construction of lateral horizontal deformation prediction model of similar projects.

Key words:shield construction;field measurement;deep neural network;lateral horizontal deformation;deformation prediction model

 024_2021010100_马铭骏_H.pdf