基于机器学习的地铁盾构掘进参数智能预测与控制方法研究  

Research on Intelligent Predicting and Controlling Method of Shield Tunneling Parameters Based on Machine Learning

作　　者：刘颖彬 廖少明[1] 杜华 白立峰 钟壬旭 LIU Yingbin;LIAO Shaoming;DU Hua;BAI Lifeng;ZHONG Renxu(Department of Geotechnical Engineering,Tongji University,Shanghai 200092,China;Jinhua Jinyi East Rail Transit Co.,Ltd.,Jinhua 321017,China)

机构地区：[1]同济大学地下建筑与工程系,上海200092 [2]金华市金义东轨道交通有限公司,浙江金华321017

出　　处：《铁道标准设计》2023年第7期137-145,154,共10页Railway Standard Design

基　　金：国家自然科学基金项目(52090082);金华市科技计划项目(20193027)。

摘　　要：盾构掘进的精细化和智能化控制是现代隧道施工技术的发展趋势,为更好地预测和控制盾构掘进状态,提出一种预测和控制盾构掘进参数的智能模型。该模型考虑了多个影响盾构掘进参数的非线性因素,建立了6种基于机器学习与海鸥算法相结合的混合算法(SOA-ML)盾构掘进参数智能预测模型,并提出基于层次分析法的最佳预测模型判别方法;进一步利用最佳预测模型提出了基于PSO算法的掘进参数控制方法。以金华至义乌至东阳市域轨道交通工程为例,验证了模型的有效性。运用结果表明:SOA算法可有效地对机器学习算法的超参数调优,且SOA种群数量越大,搜索的范围越广,最佳适应度收敛性越快;6种算法模型均具有较好的预测性能,根据层次分析法对预测模型进行性能排序为BP>ELM>CNN>RF>SVM>LSTM;基于BP-PSO的盾构掘进参数预测和控制过程具有消耗时间小、预测与优化性好的特点。The fine and intelligent control of shield tunneling addresses the development trend of modern tunneling construction technology.In order to better predict and control shield tunneling state,an intelligent framework for predicting and controlling shield tunneling parameters is proposed in the paper.The model considers several nonlinear factors affecting shield tunneling.Six intelligent models based on the combination of machine learning and seagull algorithm(SOA-ML)are used to predict the shield tunneling state.Due to the differential applicability of different algorithms,an algorithmic evaluating method based on analytic hierarchy process is proposed.Moreover,taking the best algorithm as prediction model,the controlling method of driving parameters based on PSO algorithm is further established.Besides,the model is applied to the rail transit project from Jinhua to Yiwu to Dongyang,and its validity is proven.The results show that The SOA algorithm can effectively optimize the hyperparameters of the machine learning algorithm,and the larger the number of SOA species,the wider the search scope,and the faster the convergence of the best fitness.The six algorithm models all have good predictive performance.According to the analytic hierarchy Process(AHP),the performance ranking of prediction models is as follows:BP>ELM>CNN>RF>SVM>LSTM.The prediction and control process of shield tunneling parameters based on BP-PSO is characterized by low time consumption and good prediction and optimization.

关 键 词：地铁 盾构掘进 盾构法 智能模型 机器学习 盾构参数 优化算法 层次分析法 

分 类 号：U455.43[建筑科学—桥梁与隧道工程] U456.3[交通运输工程—道路与铁道工程]