基于随机IDA和机器学习的盾构隧道地震易损性分析  

Seismic vulnerability analysis of shield tunnel based on random IDA and machine learning algorithms

作　　者：刘鹏 丁祖德[1] 资昊 陈誉升 刘正初[2] LIU Peng;DING Zude;ZI Hao;CHEN Yusheng;LIU Zhengchu(Faculty of Civil Engineering and Mechanics,Kunming University of Science and Technology,Kunming 650500,China;Kunming Survey,Design and Research Institute Co.,Ltd.of China Railway Second Institute,Kunming 650200,China)

机构地区：[1]昆明理工大学建筑工程学院,云南昆明650500 [2]中铁二院昆明勘察设计研究院有限责任公司,云南昆明650200

出　　处：《铁道科学与工程学报》2023年第12期4848-4860,共13页Journal of Railway Science and Engineering

基　　金：国家自然科学基金资助项目(52168057,51768028)。

摘　　要：在隧道地震易损性分析中,涉及地震动、结构、岩土参数的随机性和不确定性。当考虑多种参数不确定性时,基于数值法建立结构地震易损性曲线需要进行大量工况的随机地震响应计算,面临很高的计算成本。为兼顾隧道地震易损性结果的客观性和计算的高效性,基于Python语言编写自动随机IDA脚本程序,实现隧道随机地震响应的自动分析和后处理功能。在此基础上,结合10种机器学习算法和地震易损性分析理论,提出了基于随机IDA和机器学习的盾构隧道地震易损性分析框架。以某盾构隧道为例,在建立确定性动力响应计算模型基础上,采用拉丁超立方抽样方法并结合自动随机IDA程序,完成了隧道随机增量动力分析并形成机器学习算法的数据集。通过特征选择、数据集划分、预处理及参数调优,建立了预测结构地震损伤的10种机器学习算法模型,对比分析了不同机器学习算法的预测性能,进一步讨论了土层随机参数对结构易损性的敏感性。结果表明:各机器学习算法模型在数据集上均表现出较好的预测性能,其中BPNN模型的综合预测性能最好。基于机器学习模型预测建立的隧道易损性曲线,与数值法结果总体一致,说明采用机器学习算法预测隧道地震易损性是有效的,特别是BPNN算法模型,与数值法非常接近,体现较高的可靠性。对样本数量敏感性小的机器学习算法,如BPNN算法模型在易损性预测中具有很好的适用性和应用潜力。从土体物理力学参数对盾构隧道地震易损性影响的敏感性程度而言,由大至小依次为弹性模量、泊松比、密度、阻尼、摩擦角和黏聚力。The vulnerability analysis of tunnels involves randomness and uncertainty in seismic motion,structure,and geotechnical parameters.When considering the uncertainty of multiple parameters,establishing the seismic vulnerability curve of the structure based on numerical methods is computationally expensive and requires a large number of random seismic response calculations.To balance the objectivity of the tunnel seismic vulnerability results with computational efficiency,an automatic random IDA script program,coded in Python,was developed for automatic analysis and post-processing of tunnel random seismic response.Ten machine learning algorithms were established for predicting structural earthquake damage via feature selection,dataset partitioning,preprocessing,and parameter tuning.The predictive performance of the algorithms was compared and analyzed,including the sensitivity of soil random parameters to structural vulnerability.The tunnel vulnerability curve,established based on machine learning prediction,generally agreed with the numerical method results,demonstrating the effectiveness of machine learning algorithms in predicting tunnel seismic vulnerability,particularly the BPNN algorithm,which is highly reliable,with close proximity to the numerical method.Machine learning algorithms,notably the BPNN algorithm model,which are less susceptible to sample quantity,have potential and good applicability in vulnerability prediction.With respect to the sensitivity of soil physical and mechanical parameters to shield tunnel seismic vulnerability,there is an order of decreasing importance:elastic modulus,Poisson’s ratio,density,damping,friction angle,and cohesion.

关 键 词：盾构隧道 地震响应模型 地震易损性曲线 随机IDA方法 机器学习算法 

分 类 号：U451[建筑科学—桥梁与隧道工程]