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): 33-39.DOI: 10.3969/j.issn.1674-0696.2026.09.04

• 智慧交通基础设施 • 上一篇    

基于PSR-GA-LSTM的桥梁健康监测缺失数据恢复方法研究

向程龙1,2，冉光明3，周朝营4，吴波4   

  1. (1. 贵州交通职业大学 路桥工程学院，贵州 贵阳 551400； 2. 贵州交通职业大学 贵州省山区桥隧工程智能建造与运维全省重点实验室，贵州 贵阳 551400； 3. 贵州省六安高速公路有限公司，贵州 安顺 561300； 4. 重庆交通大学 土木工程学院，重庆 400074)
  • 收稿日期:2026-03-27 修回日期:2026-06-12 发布日期:2026-09-23
  • 作者简介:向程龙(1988—)，男，贵州毕节人，副教授，硕士，主要从事桥梁检测、健康监测等方面的研究。E-mail：425456824@qq.com 通信作者：吴波（1991—），男，四川南充人，副教授，博士，主要从事桥梁健康监测等方面的研究。E-mail：bo.wu@cqjtu.edu.cn
  • 基金资助:
    贵州省交通运输厅科技项目（2023-122-001)；贵州省交通运输厅科研项目（2026-112-003）

Missing data recovery method for bridge health monitoring based on PSR-GA-LSTM

Xiang Chenglong1,2, Ran Guangming3, Zhou Chaoying4, Wu Bo4   

  1. (1. School of Road and Bridge Engineering, Guizhou Communications Polytechnic University, Guiyang 551400, Guizhou, China; 2. Guizhou Provincial Key Laboratory of Intelligent Construction, Operation and Maintenance of Mountain Bridges and Tunnels, Guizhou Communications Polytechnic University, Guiyang 551400, Guizhou, China; 3. Guizhou Liu’an Expressway Co., Ltd., Anshun 561300, Guizhou, China; 4. School of Civil Engineering, Chongqing Jiaotong University, Chongqing 400074, China)
  • Received:2026-03-27 Revised:2026-06-12 Published:2026-09-23

PDF

摘要： 在桥梁健康监测系统中，准确可靠的监测数据是掌握结构状态的基础。然而，由于传感器故障导致的数据缺失与失真，以及环境噪声和仪器误差的干扰，难以获取可靠的监测数据。为解决此问题，笔者提出一种基于相空间重构（PSR）、遗传算法（GA）与长短期记忆神经网络（LSTM）相结合的桥梁监测数据恢复方法。具体而言，首先运用PSR将原始信号重构到高维相空间中；随后采用遗传算法优化LSTM网络的关键参数，并训练网络学习重构数据的时序依赖关系。在此基础上，凭借门控机制捕获桥梁监测数据的长期时序依赖关系的LSTM网络生成完整数据恢复结果。基于桥梁健康监测系统中的加速度数据的实例研究验证了该方法的有效性。结果表明，文中方法在数据恢复性能方面显著超越其他基准方法，与传统LSTM模型相比，文中方法的ERMS降低约57.7%，验证了该混合方法在桥梁健康监测数据缺失恢复领域的实用性和有效性。

关键词: 桥梁工程；数据恢复；PSR-GA-LSTM；相空间重构；遗传算法；长短期记忆网络；健康监测

Abstract: In bridge health monitoring systems, accurate and reliable monitoring data is the foundation for understanding structural conditions. However, it is difficult to obtain reliable monitoring data due to data gaps and distortions caused by sensor failures, as well as interference from environmental noise and instrument errors. To address this issue, a bridge monitoring data recovery method based on the integration of phase space reconstruction (PSR), genetic algorithms (GA) and long short-term memory neural networks (LSTM) was proposed. Specifically, PSR was first employed to reconstruct the raw signal into a high-dimensional phase space. Subsequently, GA was adopted to optimize key parameters of the LSTM network, and the network was trained to learn the temporal dependencies of the reconstructed data. Based on this, the LSTM network, equipped with a gating mechanism to capture long-term temporal dependencies of bridge monitoring data, generated complete data restoration results. An empirical study based on acceleration data from a bridge health monitoring system validated the effectiveness of the proposed method. Results demonstrate that the proposed method significantly outperforms other benchmark approaches in data recovery performance. Compared to traditional LSTM models, the proposed method reduces ERMS by about 57.7%, validating the practicality and effectiveness of this hybrid approach for missing data recovery in bridge health monitoring field.

Key words: bridge engineering; data recovery; PSR-GA-LSTM; phase space reconstruction; genetic algorithm; long short-term memory network; health monitoring

中图分类号: