基于PSO-VMD和稀疏注意力机制改进BiLSTM的桥梁健康诊断  

Improved BiLSTM bridge health diagnosis based on PSO-VMD and sparse attention mechanism

作　　者：华谦 HUA Qian(China Railway Southwest Scientific Research Institute Co.,Ltd.,Chengdu 611731,China)

机构地区：[1]中铁西南科学研究院有限公司,成都611731

出　　处：《国外电子测量技术》2025年第1期110-118,共9页Foreign Electronic Measurement Technology

基　　金：成都市科技局重点研发支撑计划(2022-YF05-00474-SN)。

摘　　要：为提高对桥梁健康监测的准确性,提出基于粒子群优化算法(PSO)-变分模态分解(VMD)-改进双向长短期记忆网络(BiLSTM)的桥梁健康监测方法。首先,采用PSO算法求解VMD算法的最佳分解个数K和惩罚因子α,并将优化后的VMD算法对采集到的桥梁健康监测挠度数据进行降噪处理,提高对桥梁挠度信号分解效果;然后利用稀疏注意力机制改进的BiLSTM网络,对桥梁挠度的预测。结果表明:采用PSO-VMD算法可有效剔除桥梁挠度数据中的噪声,且通过重构后的挠度信号在信噪比、均方根误差、平均绝对误差的评价指标分别为30.59 dB、3.89 mm、3.10 mm;采用PSO-VMD结合改进BiLSTM网络能准确预测桥梁挠度,平均绝对百分比误差、均方根误差均小于1.5 mm,绝对系数达到0.98以上。由此得出,基于PSO-VMD-改进BiLSTM的桥梁健康监测具有一定的工程应用价值。To improve the accuracy of bridge health monitoring,a bridge health monitoring method based on PSOVMD improved BiLSTM is proposed.Firstly,Particle Swarm Optimization(PSO)algorithm is used to solve the optimal decomposition number K and penalty factorαof Variational Mode Decomposition(VMD)algorithm,and then the optimized VMD algorithm is used to denoise the collected bridge health monitoring deflection data to improve the decomposition effect of bridge deflection signals.Then,using sparse attention mechanism to improve the BiLSTM network,and applying the improved BiLSTM network for predicting bridge deflection.The results show that the PSOVMD algorithm can effectively remove noise from bridge deflection data,and the evaluation indicators of signal-to-noise ratio,root mean square error,and average absolute error of the reconstructed deflection signal are 30.59 dB,3.89 mm,and 3.10 mm,respectively;The use of PSO-VMD combined with improved BiLSTM network can accurately predict bridge deflection,with an average absolute percentage error and root mean square error of less than 1.5mm,and an absolute coefficient of over 0.98.From this,it can be concluded that the bridge health monitoring based on PSO-VMD improved BiLSTM has certain engineering application value.

关 键 词：桥梁健康监测 PSO算法 VMD算法 BiLSTM网络 挠度预测 

分 类 号：TU446[建筑科学—岩土工程] TP399[建筑科学—土工工程]