中文核心期刊
CSCD来源期刊
中国科技核心期刊
RCCSE中国核心学术期刊
重庆交通大学学报（自然科学版）

重庆交通大学学报（自然科学版） ›› 2026, Vol. 45 ›› Issue (9): 24-32.DOI: 10.3969/j.issn.1674-0696.2026.09.03

• 智慧交通基础设施 • 上一篇    

基于贝叶斯流的斜拉桥概率有限元模型修正方法

刘迅1,2，林学春3，卓卫东2，郑豪峰4，谷音2   

  1. (1. 荆门市交通运输综合执法支队，湖北 荆门 448000； 2. 福州大学 土木工程学院，福建 福州 350108； 3. 福建省交通建设质量安全中心，福建 福州 350001； 4.福州机场复线高速公路有限公司，福建 福州 350007)
  • 收稿日期:2026-03-04 修回日期:2026-07-06 发布日期:2026-09-23
  • 作者简介:刘迅(1991—)，男，重庆垫江人，工程师，博士，主要从事桥梁工程方面的研究。E-mail：634455345@qq.com 通信作者：卓卫东(1966—)，男，福建莆田人，教授，博士，主要从事桥梁与结构工程方面的研究。E-mail：zhuowd@fzu.edu.cn
  • 基金资助:
    国家自然科学基金面上项目（52278159）；福建省交通运输科技项目（ZH202305）

Probabilistic finite element model updating method of cable-stayed bridges based on Bayesian flow

Liu Xun1,2, Lin Xuechun3, Zhuo Weidong2, Zheng Haofeng4, Gu Yin2   

  1. (1. Jingmen Transportation Comprehensive Law Enforcement Detachment, Jingmen 448000, Hubei, China; 2. College of Civil Engineering, Fuzhou University, Fuzhou 350108, Fujian, China; 3. Fujian Provincial Transportation Construction Quality and Safety Center, Fuzhou 350001, Fujian, China; 4. Fuzhou Airport Double Line Expressway Co., Ltd., Fuzhou 350007, Fujian, China)
  • Received:2026-03-04 Revised:2026-07-06 Published:2026-09-23

PDF

摘要： 为实现高效且准确的斜拉桥概率有限元模型修正，提出一种基于贝叶斯流的斜拉桥概率有限元模型修正方法：该方法先采用贝叶斯流，建立有限元模型的模拟测量数据与模型修正参数之间的可逆映射；再基于桥梁的实际测量数据，通过贝叶斯流推断模型修正参数的后验分布，从而实现桥梁概率有限元模型修正。文中方法提供了一个桥梁概率有限元模型修正的“离线训练、在线推断”框架；一旦训练完成，贝叶斯流模型可基于新的测量数据，近乎实时地快速给出修正参数的后验分布。文中方法应用于一座在役大跨斜拉桥的有限元模型修正，并与桥梁竣工验收荷载试验结果进行对比。结果表明：模型修正后计算得到的恒载索力和模态频率的平均相对误差分别仅为1.38%和1.90%；与未修正的初始有限元模型相比，由修正模型计算得到的试验车载作用下主梁跨中挠度增量的相对误差从10.37%降至1.08%，最大索力增量的相对误差从18.98%降至5.95%。工程实例应用验证了文中方法的有效性。

关键词: 桥梁工程；斜拉桥；贝叶斯流；离线训练；在线模型修正；结构健康监测

Abstract: To achieve efficient and accurate probabilistic finite element model updating (PFEMU) of cable-stayed bridges, a PFEMU method of cable-stayed bridges based on Bayesian flow was proposed. Firstly, in the proposed method, Bayesian flow was adopted to establish a reversible mapping between the simulated measurement data of the FEM and the model updating parameters. Secondly, based on the actual measurement data of the bridge, the posterior distribution of the model updating parameters was inferred through Bayesian flow, thereby realizing the PFEM of the bridge. The proposed method provided an “offline training and online inference” framework for the PFEMU of a bridge. Once the training was completed, the Bayesian flow model could quickly infer the posterior distribution of the updating parameters in near real time based on new measurement data. The proposed method was applied to the FEMU of an in-service long-span cable-stayed bridge, and the results were compared with those obtained from the bridge completion load test. The results show that after model updating, the average relative errors of the calculated cable forces under dead load and modal frequencies are only 1.38% and 1.90% respectively. Compared with the results of the initial FEM, the relative error of the mid-span deflection increment of the main girder under test vehicle load calculated by the updated model is reduced from 10.37% to 1.08%, and the relative error of the maximum cable force increment is reduced from 18.98% to 5.95%. The effectiveness of the proposed method has been verified through engineering examples.

Key words: bridge engineering; cable-stayed bridges; Bayesian flow; offline training; online model correction; structural health monitoring

中图分类号: