基于MSGWO-LSTM的车桥非线性系统地震响应预测研究  

Research on earthquake response prediction of nonlinear vehicle bridge systems based on MSGWO-LSTM

作　　者：刘汉云 王子逸 韩艳 王力东 胡朋 国巍[2] 余志武[2] LIU Hanyun;WANG Ziyi;HAN Yan;WANG Lidong;HU Peng;GUO Wei;YU Zhiwu(School of Civil Engineering,Changsha University of Science&Technology,Changsha 410114,China;School of Civil Engineering,Central South University,Changsha 410075,China)

机构地区：[1]长沙理工大学土木工程学院,湖南长沙410114 [2]中南大学土木工程学院,湖南长沙410075

出　　处：《铁道科学与工程学报》2025年第2期734-747,共14页Journal of Railway Science and Engineering

基　　金：国家自然科学基金资助项目(52108433,52178452,52278546,52022113);湖南省科技创新计划资助项目(2021RC4031,2024RC3170);湖南省自然科学基金资助项目(2024JJ5018,2024JJ2002);高速铁路建造技术国家工程研究中心开放基金资助项目(HSR202004)。

摘　　要：强震下高铁桥梁结构易进入非线性阶段,导致系统响应增大难预测,可能威胁行车安全。为此,提出融合多策略灰狼优化-长短期记忆网络(MSGWO-LSTM)代理模型,提升高速铁路车-桥耦合非线性系统地震响应预测精度。建立车桥耦合系统OpenSees非线性模型,基于大量地震动动力分析,构建桥梁位移、车体加速度和轮轨力响应数据库。在传统LSTM代理模型基础上,引入Dropout层以防止模型训练过拟合,引入灰狼优化算法(GWO)进行超参数自动选优,从而构建并训练了GWO-LSTM代理模型。采用多类评判指标,考虑结构线性/非线性、不同车速等工况,对比传统LSTM和GWOLSTM这2个代理模型的预测效果,发现GWO-LSTM在部分工况不满足需求,故融合多策略提出MSGWO-LSTM代理模型,进一步提升模型的预测精度。研究结果表明:GWO-LSTM代理模型预测的R^(2)稳定在0.95~0.99之间,MAE、MSE和RMSE等评价指标均接近0,且MAPE指标大多数控制在1%左右,明显优于传统LSTM代理模型,说明GWO-LSTM显著提升了车-桥耦合系统地震响应预测精度。相比单输入单输出模式,考虑多变量相互影响的多输入多输出模式所构建的代理模型的非线性适应性与泛化性更好。在多输入多输出和非线性工况下,GWO-LSTM模型预测响应有小部分预测指标超过10%限值,而MSGWO-LSTM所有预测指标均小于限值,进一步提升了模型的预测精度和泛化能力。During strong earthquakes,high-speed railway(HSR)bridges are prone to enter the nonlinear phase,leading to significant difficulties in predicting system responses and potentially in threatening train running safety.Therefore,this study proposed a Multi-Strategy Gray Wolf Optimization-Long Short-Term Memory(MSGWO-LSTM)surrogate model to improve the accuracy of earthquake response prediction for the HSR nonlinear system.An OpenSees nonlinear model of the vehicle-bridge coupling system was established.Based on a large amount of seismic dynamic analysis,a database of bridge displacement,vehicle acceleration,and wheel-rail force response was constructed.Based on the traditional LSTM surrogate model,a Dropout layer was introduced to prevent overfitting during model training,and the Grey Wolf Optimization Algorithm(GWO)was introduced for automatic hyperparameter selection,thus constructing and training the GWO-LSTM surrogate model.By using multiple evaluation indicators and considering structural linearity or nonlinearity,different vehicle speeds,and other working conditions,the predictive performance of traditional LSTM and GWO-LSTM surrogate models was compared.It was found that GWO-LSTM did not satisfy the requirements in certain working conditions;therefore,a MSGWO-LSTM surrogate model was proposed by integrating multiple strategies,further improving the prediction accuracy of the model.The research results show that the R^(2) predicted by the GWO-LSTM surrogate model is stable between 0.95 and 0.99,and all other predictive indicators are close to 0.Moreover,the MAPE index is primarily controlled at around 1%,which is significantly better than the traditional LSTM surrogate model.This indicates that the GWO-LSTM model can significantly improve the accuracy of earthquake response prediction.In MIMO and nonlinear conditions,the prediction response of the GWO-LSTM model has a small number of predictors exceeding the 10%limit,while all prediction indicators of MSGWO-LSTM are below the limit,further improving the model�

关 键 词：桥梁工程 地震响应预测 LSTM代理模型 高速铁路 OPENSEES 

分 类 号：U443.5[建筑科学—桥梁与隧道工程]