摘要
高速铁路列车进站停车对标的准确性是衡量列车自动驾驶系统(automatic train operation,ATO)性能的重要指标之一。实现精确对标停车不仅能确保列车运行的安全性和乘客的舒适性,还直接影响运营效率。为了精准预测高铁ATO的停车轨迹,提出一种改进的CNN-Informer的ATO停车对标模型。该模型深度挖掘列车运行过程中的多维度数据,以列车运行数据为输入,涵盖坡度、里程、运行工况等关键特征。通过卷积神经网络(convolutional neural network,CNN)的卷积层和池化层,模型能够自动学习和提取输入数据中的核心特征,从而提升预测的准确性。模型将CNN和全连接层输出的特征向量作为Informer长时间序列预测算法的输入,应用于ATO停车对标预测。为验证模型性能,论文采用某线路真实的高铁ATO控车数据进行训练、验证和测试。试验结果表明,提出的ATO停车对标模型在列车进站前能有效预测控车情况,辅助ATO系统及时调整控车策略。相比传统Informer模型,在预测ATO控车速度时,均方误差值降低了4.95%,平均绝对误差值降低了22.35%,平均绝对百分比误差值降低了4.16%;在预测站台距离时,均方误差值降低了21.83%,平均绝对误差值降低了19.69%,平均绝对百分比误差值更是显著降低了53.72%。研究不仅为高铁ATO停车对标的准确性提供了理论支持,也为未来列车停车过程的研究提供了新的思路和方法。
The accuracy of inbound stop target is one of the important indexes to measure the performance of ATO.The realization of accurate parking can not only ensure the safety of train operation and the comfort of passengers,but also directly affect the operation efficiency.In order to accurately predict ATO stopping track of high-speed railway,an improved CNN-Informer ATO stopping benchmark model was proposed in the paper.The model could deeply dig the multi-dimensional data in the train operation process.It could take the train operation data as input,covering the key features such as slope,mileage and operating conditions.Through the CNN convolutional layer and pooled layer,the model can automatically learn and extract the core features in the input data,thereby improving the accuracy of the prediction.The feature vector output of CNN and the fully connected layer was used as the input of Informer long time series prediction algorithm to predict ATO parking benchmark.In order to verify the performance of the model,thesis used the real ATO control data of a high-speed railway line for training,verification and testing.The test results show that the ATO parking benchmarking model proposed in this paper can effectively predict the train control situation before the train enters the station,and assist the ATO system to adjust the train control strategy in time.Compared with the traditional Informer model,the mean square error is reduced by 4.95%,the average absolute error is reduced by 22.35%,and the average absolute percentage error is reduced by 4.16%when predicting the ATO controlled vehicle speed.When predicting the station distance,the mean square error decreases by 21.83%,the mean absolute error decreases by 19.69%,and the mean absolute percentage error decreases by 53.72%.This study not only provides theoretical support for the accuracy of ATO stopping target of high-speed railway,but also provides a new idea and method for the study of train stopping process in the future.
作者
王心仪
程剑锋
易海旺
WANG Xinyi;CHENG Jianfeng;YI Haiwang(Signal Communication Research Institute,China Academy of Railway Sciences Co.,Ltd.,Beijing 100081,China)
出处
《铁道科学与工程学报》
北大核心
2025年第5期1936-1948,共13页
Journal of Railway Science and Engineering
基金
北京市科技计划项目(Z231100003823033)
中国铁道科学研究院集团有限公司课题(2023YJ105)
北京华铁信息技术有限公司科研项目(2023HT06)。
