摘要
轨道及桥梁结构参数随机性对车-轨-桥耦合系统的振动影响不能忽略。基于代理模型研究轨道-桥梁间3层弹簧刚度和弹簧阻尼以及桥梁刚度和阻尼的随机性对竖向车-轨-桥耦合系统动力响应的影响。首先,基于经典的车-轨-桥耦合系统力学模型(没有考虑桥墩),采用Monte-Carlo生成2 000个样本集,作为代理模型的训练集。然后,对比SSA-BP(麻雀优化BP算法)与传统BP神经网络、GA-BP神经网络(遗传优化BP算法)对车辆和桥梁响应的预测精度,同时探讨样本数量以及Levenberg-Marquardt和Bayesian Regulation训练算法对SSA-BP神经网络预测精度的影响。最后,假定各随机参数概率分布规律服从高斯型正态分布,所有随机参数变异系数均分为0.05、0.10、0.15、0.20、0.25等5个级别,采用所提出的SSA-BP神经网络研究轨道及桥梁的刚度和阻尼变化对车辆和桥梁响应极值的影响。结果表明:与经典的车-轨-桥耦合系统力学模型相比,所提出的代理模型具有更高的计算效率;SSA-BP模型对车辆和桥梁响应的预测精度高于GA-BP模型,GA-BP模型的预测精度高于传统的BP模型;SSA-BP模型采用Levenberg-Marquardt训练算法对车辆和桥梁响应的预测精度优于Bayesian Regulation训练算法的预测精度;道砟和桥梁之间弹簧刚度的随机变化对桥梁随机振动响应尤为明显;钢轨和轨枕之间弹簧刚度的随机性对车体响应的影响不可忽视,而桥梁刚度和阻尼随机性对车体的影响可不考虑。研究成果可为车轨桥系统随机振动响应预测进一步研究提供依据和参考。
The vibration of the vehicle-track-bridge coupling system is significantly influenced by the parameters of both the track and the bridge.In view of this,the paper introduced a proxy model to analyze the effect of randomness in spring stiffness and damping within the track-bridge layers and the bridge itself on the dynamic responses of the whole system.Initially,by using a traditional numerical model of the vehicle-track-bridge system without considering the piers,2000 sample sets were generated via Monte Carlo,which formed the training set for the proxy model.The predictive performance of SSA-BP(Sparrow Search Algorithm)with BP neural network and GA-BP(Genetic Algorithm)neural network for vehicle and bridge responses was compared.Furthermore,the effects of sample size and training algorithms(Levenberg-Marquardt and Bayesian Regulation)on the prediction accuracy of SSA-BP neural networks were analyzed.By assuming a Gaussian normal distribution for the probability distribution of each random parameter,the parameters’coefficients of variation were categorized into five levels:0.05,0.10,0.15,0.20,and 0.25,respectively.The effect of each parameter's variation on the system responses was compared and analyzed based on the SSA-BP neural network.The results show that the proposed surrogate model improves computational efficiency as compared to traditional numerical computation models.The SSA-BP model outperforms the GA-BP model in predicting vehicle and bridge responses,with the GA-BP model being superior to the traditional BP model.The prediction performance of the SSA-BP model for vehicle and bridge responses using the Levenberg-Marquardt training algorithm outperforms that of the Bayesian Regulation training algorithm.In particular,the random variation of spring stiffness between the ballast and the bridge has a significant effect on the random vibration response of the bridge.The randomness of the spring stiffness between the steel rail and the ballast and the randomness of the spring stiffness between the track an
作者
何旭辉
赵永帅
蔡陈之
HE Xuhui;ZHAO Yongshuai;CAI Chenzhi(School of Civil Engineering,Central South University,Changsha 410075,China;Hunan Key Laboratory of Disaster Prevention and Mitigation of Rail Transit Engineering Structures,Changsha 410075,China)
出处
《铁道科学与工程学报》
EI
CAS
CSCD
北大核心
2024年第8期3225-3236,共12页
Journal of Railway Science and Engineering
基金
国家自然科学基金资助项目(51925808,52327810,52378546)
湖南省自然科学基金资助项目(2023JJ30665)。
