基于VMD-GRU-DTW组合模型的在役桥梁健康状态预测预警技术  

Prediction and Early Warning Technology for Structural Health Conditions of In-service Bridges Based on a Hybrid VMD-GRU-DTW Model

作　　者：张江广 桂成中 段志 曹星宇 ZHANG Jiangguang;GUI Chengzhong;DUAN Zhi;CAO Xingyu(School of Civil Engineering,Institute of Disaster Prevention,Sanhe 065201,China;Key Laboratory of Building Damage Mechanism and Defense of China Seismological Bureau,Sanhe 065201,China;ZhongziHighway Maintenance and Testing Technology Co.,Ltd.,Beijing 100097,China)

机构地区：[1]防灾科技学院土木工程学院,河北三河065201 [2]中国地震局建筑物破坏机理与防御重点实验室,河北三河065201 [3]中咨公路养护检测技术有限公司,北京100097

出　　处：《防灾科技学院学报》2025年第1期31-40,共10页Journal of Institute of Disaster Prevention

基　　金：中央高校基本科研业务费专项(ZY20230202);廊坊市科技计划项目(2022011066);中国电建市政建设集团有限公司科研课题项目(DJ-ZDXM-2022-35)。

摘　　要：为了提高桥梁健康监测中数据预测的精度,解决不同位置传感器因受力差异导致的数据相关性下降问题以及部分监测项目中仅获取单一变量数据的局限性,提出了一种新型组合模型。该模型结合多分量分析与滑动多步预测技术,首先利用变分模态分解将传感器数据分解为本征模态函数,然后通过自监督学习和超参数优化,引入门控循环单元对每个模态函数进行多步预测。预测完成后,采用动态时间规整评估预测数据与历史数据的相似性,识别可能处于最不利状态的传感器。最终,模型在实际桥梁监测数据中得到了验证,证明了其在解决单一传感器数据复杂性和多特征记录缺失问题方面的有效性,并为桥梁健康状况预警提供了可靠的解决方案。In this paper,a new VMD-GRU-DTW combined model is proposed to enhance the accuracy of data prediction in bridge health monitoring,address the decrease of data correlation due to force variations across different sensor locations,and mitigate the limitations of acquiring only single-variable data in certain monitoring projects.It integrates multi-component analysis with sliding multi-step prediction techniques.First,Variational Mode Decomposition(VMD)is used to decompose the sensor data into Intrinsic Mode Functions(IMfs).Then,Gated Recurrent Units(GRus),optimized through self-supervised learning and hyperparameter tuning,are applied in multi-step prediction for each IMF.After prediction,Dynamic Time Warping(DTW)is employed to assess the similarity between predicted data and historical data,identifying sensors that may be in the most unfavorable state.Finally,the proposed model is applied in the monitoring data of the real bridge,demonstrating its effectiveness in addressing the complex single-sensor data and the missing multi-feature records,and thus providing a reliable solution for predictive maintenance and early warning in bridge health monitoring.

关 键 词：数据预测 变分模态分解 门控循环单元 动态时间规整 桥梁健康监测 粒子群优化 

分 类 号：U446[建筑科学—桥梁与隧道工程]