摘要
基于卷积神经网络,提出了一种快速预测高速铁路无砟轨道-简支梁桥系统震致损伤的方法。为得到更多地震动信息,通过连续小波变换,将一维地震动数据输入变换成三维图像输入。通过对比损伤样本库中的结果,验证了提出方法的可靠性,分析了卷积神经网络的不同超参数对预测结果和训练时长的影响,得到了贝叶斯优化后的卷积神经网络超参数组合,对比了不同抗震分析方法得到高速铁路无砟轨道-简支梁桥系统震致损伤所需时间。利用优化后的卷积神经网络预测了高速铁路无砟轨道-简支梁桥系统中不同关键构件的震致损伤。研究表明:初始学习率是影响网络预测准确度的最主要因素,学习率下降系数、最小批次及训练轮数会对网络预测结果造成一定影响。而训练卷积神经网络所需时长主要由训练轮数及最小批次决定。提出的方法对高速铁路无砟轨道-简支梁桥系统中不同构件的震致损伤均具有较高预测准确度,网络结构具有较高的适用性,优化后的卷积神经网络训练耗时更短且对高速铁路无砟轨道-简支梁桥系统震致损伤预测更准确。研究成果可为震后高速铁路系统震致损伤的快速修复提供参考。
A rapid prediction method of seismic-induced damage in high-speed railway ballastless track simply-supported bridge system is proposed based on convolutional neural networks.To obtain more information of seismic motion,one-dimensional seismic motion data is transformed into three-dimensional image through continuous wavelet transform as the input of convolutional neural network.The reliability of the proposed method is validated by comparing with results in damage samples database.The influence of different hyperparameters of convolutional neural networks on prediction results and training duration are analyzed,and a combination of hyperparameters of convolutional neural networks optimized by Bayesian optimization is obtained.The time required for seismic analysis of high-speed railway ballastless track simply-supported bridge system using different seismic analysis methods is compared.The optimized convolutional neural network is utilized to predict seismic-induced damage of different key components in high-speed railway ballastless track simply-supported bridge system.The research indicates that the initial learning rate is the most significant factor affecting the accuracy of network prediction,while the learning rate decay factor,batch size,and number of training epochs have certain effects on the network prediction results.The training duration of convolutional neural network is mainly determined by the number of training epochs and batch size.The proposed method demonstrates high prediction accuracy for seismic-induced damage in various components of high-speed railway ballastless track simply-supported bridge system,and the network structure exhibits high applicability.The optimized convolutional neural network has shorter training time and more accurate prediction for seismic-induced damage in high-speed railway ballastless track simply-supported bridge system.The research findings can provide reference for rapid repair of seismic-induced damage in high-speed railway systems after earthquakes.
作者
吴凌旭
蒋丽忠
钟天璇
易江
冯玉林
赵坚
周旺保
WU Lingxu;JIANG Lizhong;ZHONG Tianxuan;YI Jiang;FENG Yulin;ZHAO Jian;ZHOU Wangbao(School of Civil Engineering,Central South University,Changsha 410075,China;Central South University Engineering Structure Seismic Research Center,Central South University,Changsha 410075,China;Research Institute of Earthquake Resistance of High-speed Railway Engineering Structure,National Engineering Research Center of High-speed Railway Construction Technology,Changsha 410075,China;Grid Planning&Research Center,Guizhou Power Grid Co.,Ltd.,Guiyang 550000,China;School of Civil Engineering and Architecture,East China Jiaotong University,Nanchang 330013,China;Hunan Yunjing Construction Company Limited,Changsha 410075,China)
出处
《地震工程与工程振动》
CSCD
北大核心
2024年第5期26-36,共11页
Earthquake Engineering and Engineering Dynamics
基金
国家自然科学基金项目(52078487,52178180,52478517)
国家重点研发计划课题(2022YFC3004304)
湖南省科技人才托举工程(2022TJ-Y10)
中南大学前沿交叉研究项目(2023QYJC006)
澳门特别行政区科学技术发展基金项目(SKL-IOTSC(UM)-2024-2026)
智慧城市物联网国家重点实验室(澳门大学)开放课题(SKL-IoTSC(UM)-2024-2026/ORP/GA08/2023)
中南大学中央高校基本科研业务费专项资金项目(2022ZZTS0155)
中国博士后科学基金面上项目(2022M713544)
高速铁路建造技术国家工程研究中心开放基金项目(HSR202202)。
关键词
高速铁路无砟轨道-简支梁桥系统
震致损伤
快速预测
卷积神经网络
贝叶斯优化
high-speed railway ballastless track simply-supported bridge system
seismic-induced damage
rapid prediction
convolutional neural network
Bayesian optimization
