基于机器学习算法的盾构掘进地表沉降预测方法

陈仁朋 戴田 张品 吴怀娜

摘要：针对有限元、地层损失率等方法难以考虑多参数耦合作用情况下的地表沉降预测的问題，基于BP神经网络（BPNN）和随机森林算法（RF）两种机器学习算法的多参数、非线性拟合能力，提出了预测盾构掘进过程中地表最大沉降以及纵向沉降曲线的预测方法.通过粒子群算法（PSO）确定机器学习算法的最优超参数，通过k折交叉验证方法提高预测方法的鲁棒性.结果表明BP神经网络的预测结果误差较大，难以预测到较大的地表沉降，随机森林算法能够准确预测地表最大沉降和纵向沉降曲线.

关键词：盾构隧道；地表沉降；机器学习；优化

中图分类号：U455.43；P642.26文献标志码：A

基金项目：国家自然科学基金重点项目（51938005），National Natural Science Foundation of China（51938005）；创新平台与人才计划-湖湘高层次人才聚集工程-创新团队（2019RS1030）

Prediction Method of Tunneling-induced Ground Settlement Using Machine Learning Algorithms

CHEN Renpeng1，2，3，DAI Tian1，2，3，ZHANG Pin4，WU Huaina1，2，3

（1. Key Laboratory of Building Safety and Energy Efficiency of Ministry of Education，Hunan University，Changsha 410082，China；2. National Center for International Research Collaboration in Building Safety and Environment，Hunan University，Changsha 410082，China；3. College of Civil Engineering，Hunan University，Changsha 410082，China；4. Department of Civil and Environmental Engineering，Hong Kong Polytechnic University，Kowloon，Hong Kong，China）

Abstract：It is difficult to consider the prediction of ground settlement under the coupling effect of multiple fac-tors for the finite element method and formation loss rate. Based on the multi-factor and nonlinear fitting ability of back-propagation neural network（BPNN）and random forest（RF），these two machine learning algorithms are adopted to predict the tunneling-induced ground settlement. The optimum hyper-parameters of the two machine learning al-gorithms are determined by particle swarm optimization（PSO），and k-fold cross validation method is used to improve the robustness of the prediction method. The prediction results indicate that the prediction error of BP neural network is larger and it’s hard for BP neural network to predict the large settlement. The random forest algorithm can accu-rately predict the maximum settlement and longitudinal ground settlement curve.

Key words：shield tunnel；ground settlement；machine learning；optimization

土压平衡盾构由于适应地层广，对地面交通影响较小，是城市地铁隧道建设主要工法.盾构掘进过程中扰动地层，改变地层应力场，使地面产生沉降，威胁周边建筑物和构筑物的安全.准确预测地层沉降能够降低盾构掘进过程中引发的安全风险[1-2].现有的地表沉降预测方法可以分为：1）经验和半经验公式[3-4]；2）解析法[5]；3）数值模拟[6-7]；4）模型试验[8]；5）机器学习算法预测[9-12].

经验和半经验公式法可以描绘地表沉降的一般形态[3-4]，但由于不同地层经验参数的取值差异性较大，因此预测的沉降分布在较大范围内，且预测结果的准确性不能保证.解析法的计算模型较为简单[5]，仅考虑地……