首页|期刊导航|市政技术|基于自注意力机制的实桥车辆荷载识别方法研究

基于自注意力机制的实桥车辆荷载识别方法研究OA

Research on a Self-attention-based Vehicle Load Identification Method for Solid Bridges

中文摘要英文摘要

针对实际运营桥梁中车辆荷载识别易受温度效应、测量噪声和复杂工况干扰的问题,提出了一种融合数字图像相关(DIC)机器视觉监测、变分模态分解(VMD)与Transformer自注意力机制的桥梁车辆荷载识别方法.首先,利用DIC系统获取了桥梁跨中多测点竖向动挠度响应,并采用VMD分离原始挠度信号中的低频趋势项与温度效应,提取了车致挠度响应;其次,通过峰值筛选、定长切片与时间匹配,构建了桥梁车致挠度-车辆荷载数据集;最后,建立了基于Transformer的车辆荷载识别模型,并与CNN、BiLSTM、BiLSTM-Transformer模型进行了对比分析.依托1座30 m预应力混凝土简支空心板梁桥的实桥监测数据,共构建了2 931个有效样本.结果表明,所提出的车辆荷载识别方法能够有效提取车致挠度并建立可靠的数据集,深度学习方法整体识别精度较高.其中,Transformer模型表现最优,其测试集RMSE为0.609、R2为0.997、5%相对误差范围内的总体识别准确率达99.66%,验证了自注意力机制能够更有效地挖掘多通道车致挠度序列中的全局关联特征.该研究成果可为实桥车辆荷载精准识别与桥梁健康监测工程应用提供参考.

To address the problem of vehicle load identification being affected by temperature effects,measurement noise and complex operational conditions for in-service bridges,a bridge vehicle load identification method integrating digital image correlation(DIC)-based vision monitoring,variational mode decomposition(VMD)and a Transformer self-attention mechnism is proposed.First,vertical dynamic deflection responses of multiple mid-span measurement points are obtained by the DIC system.VMD is employed to separate the low-frequency trend terms from temperature effects of the raw deflection signal to extract the vehicle-induced deflection response;Then,a bridge vehicle-in-duced deflection-vehicle load dataset are constructed through peak screening,fixed-length segmentation and time matching.Finally,a Transformer-based vehicle load identification model is established and compared with CNN,BiLSTM,and BiLSTM-Transformer models.Based on monitoring data from a 30 m prestressed concrete simply sup-ported hollow slab bridge,2 931 valid samples are constructed.The results show that the proposed method can effec-tively extract vehicle-induced deflection and build a reliable dataset.And that deep learning methods overall achieve high identification accuracy.The overall recognition accuracy is relatively higher by deep learning method.Among them,the Transformer model performs best,with an RMSE of 0.609,an R2 of 0.997,and an overall accuracy of 99.66%within 5%relative error range on the test set.This indicates that the self-attention mechanism is more effec-tive in capturing global correlation features in multi-channel vehicle-induced deflection sequences.This research result can provide reference for accurate identification of vehicle loads on solid bridges and engineering applications of bridge health monitoring.

叶以挺;应柳祺;张文达;胡洁亮;张一名;应国刚

浙江未来交通产业发展有限公司,浙江 杭州 310000||宁波朗达科技有限公司,浙江宁波 315100宁波朗达科技有限公司,浙江宁波 315100宁波朗达科技有限公司,浙江宁波 315100宁波朗达科技有限公司,浙江宁波 315100宁波朗达科技有限公司,浙江宁波 315100宁波朗达科技有限公司,浙江宁波 315100||浙大宁波理工学院,浙江宁波 315100

交通工程

桥梁车辆荷载识别深度学习自注意力机制

bridge vehicle load identificationdeep learningself-attention mechanism

《市政技术》 2026 (9)

45-57,65,14

"科技甬江2035"重大应用示范项目(宁波市重大科技任务攻关项目)(2024Z016)

10.19922/j.1009-7767.2026.09.045

评论