基于机器学习的地震动强度指标敏感性分析与破坏势评估  

Sensitivity analysis of ground motion intensity measures and evaluation of potential damage based on machine learning

作　　者：吴梓楠 韩小雷[1,2] 马建峰 季静[1,2] WU Zinan;HAN Xiaolei;MA Jianfeng;JI Jing(State Key Laboratory of Subtropical Building Science,South China University of Technology,Guangzhou 510640,China;School of Civil Engineering&Transportation,South China University of Technology,Guangzhou 510640,China)

机构地区：[1]华南理工大学亚热带建筑科学国家重点实验室,广东广州510640 [2]华南理工大学土木与交通学院,广东广州510640

出　　处：《建筑结构学报》2023年第11期216-225,235,共11页Journal of Building Structures

基　　金：国家自然科学基金项目(52178483);广州市重点研发计划(202103000038);广东省自然科学基金项目(2020A1515010739)。

摘　　要：为评估地震动的潜在破坏势,采用基于机器学习的敏感性分析方法,揭示影响破坏势的关键地震动强度指标。以弹塑性多自由度体系为研究对象,选取1 100条真实地震动记录开展时程分析;以最大层间位移角为损伤指标,通过相关性分析从25个地震动强度指标中筛选出7个信息重叠度低,且与损伤指标有较强对数相关性的强度指标作为敏感性分析对象;选用极端随机森林算法开展敏感性分析,量化强度指标对损伤指标的影响程度;基于分析结果将周期划分为加速度敏感区、速度敏感区和位移敏感区,并确定各区域的关键地震动强度指标。地震动破坏势评估结果表明:关键地震动强度指标可有效表征破坏势;与多元对数线性回归相比,极端随机森林可更好地评估破坏势,且其在评估精度上的优势随结构基本周期的增大而扩大。To evaluate the potential damage caused by ground motion,machine learning-based sensitivity analysis was carried out to reveal important intensity measures.The nonlinear responses of multiple degree of freedom systems subjected to 1100 ground motion records were utilized as the study cases,and the maximum inter-story drift was taken as the damage measure.Based on the correlation analysis,seven intensity measures which have low information overlap and present high logarithmic correlation with damage measure were derived as the object of sensitivity analysis from 25 intensity measures,and the sensitivity analysis was then carried out based on the extremely randomized trees algorithm to quantify the impact of the derived intensity measures.According to the analysis results,the period was divided into acceleration-sensitive,velocity-sensitive and displacement-sensitive regions,and the corresponding important intensity measures were also proposed.The evaluation results illustrate that the potential damage is represented effectively by the proposed important intensity measures.Moreover,it is validated that the extremely randomized trees algorithm is superior to the multivariate logarithmic linear regression,and the advantage of the former in evaluation accuracy expands with the increase of the basic period of the structure.

关 键 词：多自由度体系 敏感性分析 相关性分析 机器学习 地震动强度指标 

分 类 号：TU311.3[建筑科学—结构工程]