As a key indicator reflecting the structure and performance variations of bridge, deflection can intuitively represent the stress and deformation characteristics of bridge and provide critical information for health assessment.This paper aims to develop an accurate and stable deflection prediction method. A hybrid model combining signal decomposition method with deep learning is proposed.First, deflection big data collected from structural health monitoring system are decomposed into multiple intrinsic mode functions (IMFs) using the complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN).Then the frequency-domain characteristics of each component are analyzed and used as the model inputs to enhance the feature representation.To further improve the predictive performance, the whale optimization algorithm (WOA) is applied to globally optimize the hyperparameters of the extended long short-term memory (xLSTM) network.Finally, the CEEMDAN-WOA-xLSTM model is constructed to perform deflection prediction intelligently.Simulation experiments are employed to demonstrate the performance of the model in long-term time series prediction.Actually, the model completes the prediction with R2 value of 0.934 4 and RMSE value of 0.239 9.In comparison to the conventional LSTM and xLSTM models, R2 values increased by 4.92% and 1.80%, while RMSE values decreased by 29.14% and 13.13%, respectively, thus indicating a significant reduction of prediction error.Meanwhile, the maximum values of the 95% confidence intervals of the model’s absolute and squared prediction errors are 0.192 9 and 0.063 6, respectively, thus demonstrating an excellent error control and generalization capability.It is expected that the proposed model can pave a way to the robust technical solutions for bridge health monitoring and informed maintenance decision-making.
近年来,基于深度学习方法的挠度预测方法因其在非线性时序建模中的优越性逐渐成为研究的热点。顾思思等[17]基于毫米波雷达采集的桥梁挠度数据,结合卷积神经网络(Convolutional Neural Networks,CNN)和门控制循环单元(Gate Recurrent Unit,GRU)对高速公路桥梁挠度进行预测,所得结果优于传统的长短期记忆网络(Long Short Term Memory,LSTM)。聂华伟等[18]基于挠度、温度、应力数据搭建了双向长短期记忆网络(Bi-directional Long Short Term Memory, Bi-LSTM)模型,实现了对某特大桥的挠度预测。郭永刚等[19]结合模态分解、粒子群优化和LSTM来进行桥梁挠度预测,实现了桥梁智慧安全监测。Sun等[20]提出了一种基于CNN和部分最小二乘回归方法的桥梁损伤检测与定位方法,通过倾斜与挠度数据估计节点荷载变化来实现损伤定位。Xiao等[21]融合CNN与LSTM特征提出概率深度学习模型,显著提高了预测精度与模型鲁棒性。Hao等[22]利用基于时变滤波的经验模态分解与CNN-GRU相结合的方法成功实现了超长跨度桥梁监测数据的重构与预测。
扩展长短期记忆网络(Extended Long Short Term Memory, xLSTM)[24]模型是在传统LSTM模型基础上发展而来的一种新型时间序列预测模型。通过引入多模块协同机制,xLSTM模型有效提升了对于复杂时序数据的建模能力。具体而言,xLSTM模型通过短期记忆单元(sLSTM)与矩阵记忆单元(mLSTM)协同建模来实现对多层次特征的高效学习与并行更新,模型的整体结构如图1所示。
XuZ F, SongJ, ZhangX, et al.Application of digital twin system considering measured traffic flow in structural state evaluation of cable-stayed bridge [J].J Highw Transp Res Dev, 2023, 40(6): 93-102.
YinH Y, LiuD X, TangL.Study on real mode selection of bridge structure based on K-means clustering algorithm [J].J Highw Transp Res Dev, 2020, 37(5): 73-82.
WangL B, WangQ L, ZhuZ, et al.Current status and prospects of research on bridge health monitoring technology [J].China J Highw Transp, 2021, 34(12): 25-45.
XiaoF, MaoY, TianG, et al.Partial‐model‐based damage identification of long‐span steel truss bridge based on stiffness separation method [J].Struct Control Health Monit, 2024, 2024(1): 5530300.
[15]
PanP, XingC, BaiJ, et al.A remote deflection detection method for long-span bridges using adaptive mask and high-resolution camera [J].Measurement, 2022, 201: 111774.
[16]
NikedlA, ShariatmadarH.Unsupervised damage assessment under varying ambient temperature based on an adjusted artificial neural network and new multivariate covariance-based distances [J].Struct Health Monit, 2024, 23(6): 3814-3831.
[17]
PrakashG.A deflection-based practicable method for health monitoring of in-service bridges [J].Meas Sci Technol, 2021, 32(7): 075108.
[18]
NeumannL, WittenbergP, MendlerA, et al.Confounder-adjusted covariances of system outputs and applications to structural health monitoring [J].Mech Syst Signal Process, 2025, 224: 111983.
JiangL, YangH, LiuW, et al.Early warning for continuous rigid frame bridges based on nonlinear modeling for temperature-induced deflection [J].Sensors, 2024, 24(11): 3587.
[24]
WangY, ZhouY, ZhaoY, et al.Comparison of dynamic amplification factor of deflection and bending moment of highway continuous box-girder bridges by mode superposition [J].Materials, 2024, 17(5): 1041.
[25]
GuS S, ChangS X, HanM M, et al.Bridge deflection prediction driven by millimeter wave radar data [J].Sci Technol Eng, 2023, 23(11): 4874-4880.
SunH, SongL, YuZ.A deep learning-based bridge damage detection and localization method [J].Mech Syst Signal Process, 2023, 193: 110277.
[32]
XiaoX, WangZ, ZhangH, et al.A novel method of bridge deflection prediction using probabilistic deep learning and measured data [J].Sensors, 2024, 24(21): 6863.
[33]
HaoC, LiuB, LiY, et al.A data recovery method for extra-long-span railway bridge health monitoring based on TVFEMD and CNN-GRU [J].Meas Sci Technol, 2024, 35(8): 086139.