深埋长大隧道地温预测的机器学习算法对比研究  

Comparative study of machine learning algorithms for predicting the geothermal temperature in deep-buried and long tunnels

作　　者：周权[1] 罗锋 柴波[1] 周爱国[1] ZHOU Quan;LUO Feng;CHAI Bo;ZHOU Aiguo(Institute of Geological Survey,China University of Geosciences(Wuhan),Wuhan 430074,China;China Railway First Survey and Design Institute Group Co.,Ltd.,Xi’an 710043,China)

机构地区：[1]中国地质大学(武汉)地质调查研究院,湖北武汉430074 [2]中铁第一勘察设计院集团有限公司,陕西西安710043

出　　处：《安全与环境工程》2025年第1期137-147,共11页Safety and Environmental Engineering

基　　金：国家自然科学基金项目(42377186)。

摘　　要：地热对隧道施工、工程结构及运营安全等均有较大的危害,随着我国基础设施建设布局西移,隧道建设的地质条件愈发复杂,隧道埋深和长度不断增加,隧道施工期高温热害问题频发。针对传统地温预测方法中预测精度不高、数据运用不充分,单一机器学习模型解译性差等问题,以A隧道为研究对象,将决策树(decision tree,DT)、支持向量机(support vector machine,SVM)、随机森林(random forest,RF)进行耦合,提出了基于DT-SVM-RF模型的深埋长大隧道地温预测方法。在分析隧道综合测井、地应力及岩石热物理试验、航空物探数据后,选取深度、声波波速等10个影响因子作为模型的输入,采用随机交叉验证和空间交叉验证对模型的鲁棒性、泛化能力进行检验,构建LASSO回归、随机森林、互信息3种回归模型,分析10个影响因子的特征重要性排序。结果表明:在测试集上多元线性回归、支持向量机、人工神经网络和决策树-支持向量机-随机森林(decision tree-support vector machinerandom forest,DT-SVM-RF)模型决定系数(R^(2))分别为0.76、0.91、0.88、0.93,均方误差MSE分别为17.64、6.25、8.46、5.20,DT-SVM-RF模型具有相对更优的预测性能,深度、岩石导温系数、岩石导热系数、最大水平主应力特征较为重要,说明DT-SVM-RF模型能有效地提高地温预测的准确率。研究结果可为类似隧道地温预测提供一种精度更高的可行新思路。Geothermal heat presents significant hazards to tunnel construction,engineering structures,and operational safety.With the infrastructure development in China shifting westward,the geological conditions for tunnel construction are becoming increasingly complex.As a result,the depth and length of tunnels are increasing,and high-temperature thermal damage during tunnel construction is becoming a frequent issue.To address the limitations of traditional geothermal temperature prediction methods which suffer from low accuracy,insufficient data utilization,and poor interpretability of individual machine learning models,this study focuses on Tunnel A and integrates the decision tree(DT),support vector machine(SVM),and random forest(RF)techniques.A deep-buried and long tunnel geothermal temperature prediction method based on the DT-SVM-RF model is proposed.After analyzing comprehensive logging data,rock thermophysical experimental data,and aerial geophysical exploration data,10 influencing factors,including depth and acoustic wave velocity,were selected as inputs for the model.Robustness and generalization capabilities of the model were tested using random cross-validation and spatial cross-validation.Three regression models,namely LASSO regression,random forest(RF),and mutual information,were constructed to analyze the ranking of feature importance for the 10 influencing factors.The results indicate that the multiple linear regression,support vector machine,artificial neural network,and decision tree-support vector machine-random forest(DT-SVM-RF)models have a coefficient of determination(R^(2))of 0.76,0.91,0.88,and 0.93 respectively on the test set.The mean squared error(MSE)values are 17.64,6.25,8.46,and 5.20 respectively.The DT-SVM-RF model demonstrates relatively superior predictive performance,with depth,rock thermal conductivity coefficient,rock thermal conductivity coefficient,and maximum horizontal principal stress being the most significant features.This suggests that the DT-SVM-RF model can effectively enhance

关 键 词：隧道热害 隧道安全 多元线性回归 支持向量机(SVM) 随机森林(RF) 人工神经网络(ANN) 特征选择 

分 类 号：X948[环境科学与工程—安全科学]