摘要
面向特异风环境桥梁风振实时推演,开展了雷暴风作用下大跨度桥梁抖振响应智能预测研究。以苏通大桥实测数据为基础,分析了风场参数与主梁抖振响应之间的相关性,确定了桥梁雷暴风效应的主要关联参数。基于前馈神经网络(FNN)、卷积神经网络(CNN)、长短期记忆网络(LSTM)、门控循环单元(GRU)等典型神经网络模型,以主要风场关联参数及历史抖振响应作为输入,开展了桥梁抖振响应预测网络架构与模型训练,并对比分析了4种模型的预测效果。研究结果表明:雷暴风作用下大跨度桥梁的抖振响应主要与平均风速、平均风向、脉动风速均方差、紊流积分尺度等风场参数密切相关;待预测的桥梁抖振响应与历史风场及桥梁状态参数有关,需考虑二者的记忆效应;FNN与CNN未能较好地表征该记忆效应,故预测结果与实测值仅趋势相近,预测误差相对较大;GRU与LSTM的预测效果总体较好,GRU在雷暴风风速较大时的预测效果最优;LSTM在高风速下的预测效果略低于GRU,但在风速较低时的抖振预测精度最高,即具有更强的泛化能力。研究结果可为雷暴风易发区大跨度桥梁的安全运维提供借鉴与参考。
An intelligent prediction of the buffeting responses of a long-span bridge under thunderstorm winds was performed for real-time extrapolation of bridge wind-induced vibrations in an exceptional wind environment.Based on the actual measured data of the Sutong Bridge,the correlation between the wind field parameters and the buffeting responses of the main girder was analyzed,and the main parameters related to the thunderstorm wind effects were determined.Adopting the main wind parameters and historical buffeting responses as input parameters,a prediction network was constructed and trained based on typical artificial neural network models involving feedforward neural networks(FNNs),convolutional neural networks(CNNs),long short-term memory(LSTM)networks,and gated recurrent units(GRUs).Finally,the predictive effects of the four models were compared and analyzed.The analytical results indicate that the thunderstorm wind-induced buffeting responses of the long-span bridge are primarily related to the mean wind speed,mean wind direction,root mean square of the fluctuating wind speeds,and turbulence integral scale.The responses to be predicted depend on the historical wind field parameters and bridge motions,which necessitate the consideration of the memory effect of these two factors.The FNN and CNN fail to capture the memory effect well;therefore,the predicted results are only similar to the measured values in the trend,and the prediction errors are relatively large.The prediction effects of the GRU and LSTM are generally better,and the GRU provides the best prediction when the wind speed is high.At high wind speeds,the prediction effect of the LSTM is slightly weaker than that of the GRU.However,a higher prediction accuracy of the buffeting responses is obtained by the LSTM at a low wind speed,demonstrating better generalization of the approach.These results can provide a reference for the safe operation and maintenance of long-span bridges in thunderstorm and wind-prone areas.
作者
陶天友
邓鹏
王浩
石棚
TAO Tian-you;DENG Peng;WANG Hao;SHI Peng(Key Laboratory of C&PC Structures of Ministry of Education,Southeast University,Nanjing 211189,Jiangsu,China;School of Civil Engineering,Southeast University,Nanjing 211189,Jiangsu,China)
出处
《中国公路学报》
EI
CAS
CSCD
北大核心
2023年第8期87-95,共9页
China Journal of Highway and Transport
基金
国家自然科学基金项目(52278486,51978155)。
关键词
桥梁工程
抖振响应
深度学习
雷暴风
大跨度桥梁
响应预测
bridge engineering
buffeting response
deep learning
thunderstorm wind
long-span bridge
response prediction
