基于机器学习方法的浅埋暗挖隧道地表位移预测研究  

Study on Surface Displacement Prediction of Shallow Buried Tunnel Based on Machine Learning Methods

作　　者：何海琦 史磊磊 孔恒 He Haiqi;Shi Leilei;Kong Heng(Beijing Construction Engineering Group Co.,Ltd.,Beijing 100055,China;Key Laboratory of Urban Security and Disaster Engineering of Ministry of Education,Beijing University of Technology,Beijing 100124,China)

机构地区：[1]北京建工集团有限责任公司,北京100055 [2]北京工业大学城市与工程安全减灾教育部重点实验室,北京100124

出　　处：《市政技术》2025年第1期128-136,166,共10页Journal of Municipal Technology

基　　金：国家自然科学基金项目(52478383,52438005)。

摘　　要：浅埋暗挖隧道的地表位移受众多因素影响,如隧道几何参数、地质力学参数和施工参数等。因此,建立一种能够应对复杂条件的地表位移智能预测方法,是浅埋暗挖隧道关注的核心问题,也是实现隧道智慧化施工的重要基础。采用4种机器学习方法,分别为支持向量回归、随机森林、反向传播神经网络和极限学习机,并结合粒子群优化算法确定4种机器学习模型的最优超参数或随机数,从而建立一种针对浅埋暗挖隧道地表位移的混合智能预测方法。以北京地铁12号线安华桥站—安贞桥站区间段为工程背景,构建了包含73组浅埋暗挖隧道地表位移数据的数据库,并基于3种性能评价指标(决定系数R2、平均绝对百分比误差MAPE和均方根误差RMSE)对4种混合智能预测模型的预测性能进行了对比分析。此外,采用SHAP模型研究输入特征对输出值的贡献,讨论了各影响因素对地表位移的相对重要性。结果表明,支持向量回归混合智能预测模型在测试集中性能最佳,R2为0.86,MAPE为0.0079,RMSE为1.94 mm;土的压缩模量对地表沉降影响最大,其次为隧道的中心埋深和隧道横截面的宽度。The surface displacement of shallow buried tunnel is influenced by multiple factors,such as tunnel geometrical parameters,geomechanical parameters and construction parameters.Therefore,an intelligent prediction method for surface displacement which can cope with complex conditions is the core issue of shallow buried tunnels and important foundation for intelligent tunnel construction.Combined with particle swarm optimization algorithm,four machine learning methods,namely support vector regression,random forest,backpropagation neural network,and extreme learning machine are applied to determine the optimal hyperparameters or random numbers of the four machine learning models so that a hybrid intelligent prediction method for surface displacement of shallow buried tunnels is established.Taking the section of Beijing Metro Line 12 from Anhuaqiao Station to Anzhenqiao Station as an example,a database containing 73 sets of surface displacement data of shallow buried tunnels is built.Based on three performance evaluation indicators(determination coefficient of R2,average absolute percentage error MAPE,and root mean square error RMSE),the predictive performance of four hybrid intelligent prediction models are compared and analyzed.In addition,the SHAP model is used to study the contribution of input features to output values,and the relative importance of various influencing factors on surface displacement is discussed.The results show that the support vector regression hybrid intelligent prediction model has the best performance in the test set,with R2 of 0.86,MAPE of 0.0079,and RMSE of 1.94 mm;The compressive modulus of soil has the greatest impact on surface settlement,followed by the central burial depth of the tunnel and the width of the tunnel cross-section.

关 键 词：机器学习 粒子群优化算法 浅埋暗挖隧道 智能预测 地表位移 

分 类 号：U455[建筑科学—桥梁与隧道工程]