摘要
高速列车在长期服役条件下,其车辆悬挂系统等参数与设计值差异较大。多体动力学仿真模型难以模拟真实运营环境,且计算效率较低。为更加准确、快速地评价各种轨道结构以及不平顺激励下车体的垂向振动响应,根据实测轨道不平顺与车体垂向加速度的时空数据传递特征,建立一种卷积长短期记忆组合模型,该模型将轨道不平顺与列车运行速度作为输入,实现对车体垂向加速度的预测。结果表明,卷积长短期记忆模型预测的平均绝对百分比误差值为5.64%,相比动力学仿真模型减少3.57%。在预测一段3 km长线路的垂向车体加速度时,动力学仿真模型需要花费约53 s,而卷积长短期记忆网络只需要花费约1.6 s,预测效率提升33倍。
The parameters of the vehicle suspension system of high-speed trains under long-term service conditions are quite different from the design values.Multi-body dynamics simulation model can hardly simulate the real operation environment,which has low calculation efficiency.In order to more accurately and quickly evaluate the vertical vibration response of the vehicle body under various track structures and excitation of irregularities,a convolution neural network-long short term memory(CNN-LSTM)combined model was established based on the spatiotemporal data transmission characteristics between the measured track irregularity and vertical car-body acceleration.The model took track irregularities and running speed as input data to realize the prediction of the vertical car-body acceleration.The results show that the value of the mean absolute percentage error of the CNN-LSTM model is 5.64[WTB4]%[WTBZ],which is reduced by 3.57[WTB4]%[WTBZ]compared with the dynamic simulation model.In the case of predicting the vertical car-body acceleration on a 3 km long line,the dynamic simulation model takes about 53 s,while the CNN-LSTM network only takes about 1.6 s,increasing the prediction efficiency by 33 times.
作者
何庆
利璐
李晨钟
汪健辉
王平
HE Qing;LI Lu;LI Chenzhong;WANG Jianhui;WANG Ping(Key Laboratory of High-speed Railway Engineering,Ministry of Education,Southwest Jiaotong University,Chengdu 610031,China;School of Civil Engineering,Southwest Jiaotong University,Chengdu 610031,China)
出处
《铁道学报》
EI
CAS
CSCD
北大核心
2023年第6期106-113,共8页
Journal of the China Railway Society
基金
国家重点研发计划(2022YFB2602905)
国家自然科学基金(U1934214)
四川省科技计划(2023NSFSC1975)。
