迁移学习在结构工程中的应用与实例分析  

Application and case study of transfer learning in structural engineering

作　　者：杨渊[1,2] 练继建 周观根[2] 陈志华 刘红波 JANKOWSKI Łukasz[4] 杜颜胜[1] YANG Yuan;LIAN Ji-jian;ZHOU Guan-gen;CHEN Zhi-hua;LIU Hong-bo;JANKOWSKI Lukasz;DU Yan-sheng(School of Civil Engineerng,Tianjin University,Tianjin 300072,China;Zhejiang Southeast Space Frame Co.,Ltd.,Hangzhou 311209,China;School of Hydraulic and Hydro-power Hebei University of Engineering,Handan 056021,China;Polish Academy of Sciences,Warsaw 02-106,Poland)

机构地区：[1]天津大学建筑工程学院,天津300072 [2]浙江东南网架股份有限公司,浙江杭州311209 [3]河北工程大学水利水电学院,河北邯郸056021 [4]波兰科学院,华沙02-106,波兰

出　　处：《空间结构》2023年第1期31-38,74,共9页Spatial Structures

基　　金：国家自然科学基金项目(51878443);波兰国家科学中心项目(2018/31/B/ST8/03152).

摘　　要：大数据和人工智能技术正在对各行业产生深刻影响,带来新的发展机遇.工程结构个体特征强,其数据资源难以有效利用,建立的机器学习模型也难以有效推广到其他工程结构中.针对此问题,提出了基于Benchmark模型建立工程结构通用机器学习模型,通过迁移学习完成实际工程结构任务的基本思路,并探讨了工程结构中的迁移学习分类、流程、研究现状和展望.最后,通过结构损伤识别的实例分析探讨了工程结构中的自迁移学习,证明了在工程结构中使用迁移学习的有效性.Big data and artificial intelligence technology have a profound impact on all industries and bring new development opportunities.The engineering structures have strong individual characteristics,and the data resources are difficult to be effectively utilized.Furthermore,the machine learning models established by the data resources are also difficult to be effectively extended to other engineering structures.This paper puts forward the basic idea of creating the general machine learning models of engineering structure based on the benchmark model.Furthermore,the classification,process,research status,and the prospect of transfer learning in engineering structures are discussed.Finally,self-transfer learning in engineering structures is discussed with a case study of the structural damage identification.Moreover,the effectiveness of using transfer learning in engineering structures is proved.

关 键 词：迁移学习 工程结构 损伤识别 大数据 机器学习 深度学习 

分 类 号：TU311.41[建筑科学—结构工程]