Operating multiple trains as a group under Train-to-Train communication is an efficient way in sparse railway networks. However, it is a barrier to accurate tracking for follower trains with the uncertain status of the preceding trains under communication delay. Therefore, we proposed a train trajectory prediction approach, aiming to obtain an accurate state of preceeding trains in a long time horizon. Firstly, we established a hybrid prediction model combined Long Short-Term Memory (LSTM) with the Kalman Filter (KF) and using the sliding window mechanism to enhance future prediction accuracy. Secondly, to fusion LSTM prediction data and the Kalman calculation model, we proposed the Hybrid Train Trajectory Prediction algorithm by adopting real-time calculation of the covariance matrix using real data and predicted data. Finally, we apply an experiment considering network characteristics in Western China. The result shows that the Mean Absolute Error (MAE) of series prediction in the longer time domain is 44% of the Recurrent Neural Network (RNN) model and 88% of the traditional LSTM model.
Trajectory Prediction of High-Speed Train Group Tracking Based on LSTM-KF Hybrid Model
2024-09-24
1115826 byte
Conference paper
Electronic Resource
English
Train wheel size prediction based on EEMD-LSTM model
British Library Conference Proceedings | 2022
|Pedstrian Trajectory Prediction Based on LSTM
IEEE | 2023
|Speed trajectory optimization for a high-speed train
TIBKAT | 2020
|LSTM-Based Trajectory Tracking Control for Autonomous Vehicles
SAE Technical Papers | 2022
|