The invention provides a control method based on meta reinforcement learning for a train in multiple scenes. The control method is characterized in that an old task data collection module, an old task model training module, a new task curve generation module and a train control module are included; the control method comprises the steps that an old task data collection module collects a plurality of automatic driving curves generated by an old task, then the old task data collection module divides the old task data into three training sample sets according to three scene types, then the three training sample sets are used for training to obtain three available models respectively, a comparative learning method and a meta-reinforcement learning method are combined in available model training, and then in a new task, the new task data collection module collects the new task data. And selecting a corresponding available model according to the scene type corresponding to the new task, and quickly generating a new driving speed curve to control the train to run by finely adjusting model parameters. By adopting the control method provided by the invention, the driving speed curve of the train can be quickly and efficiently generated when the train encounters a new task, the operation efficiency is improved, and the economic loss is reduced.
本发明提供了一种列车在多场景下基于元强化学习的控制方法,其特征在于:包括旧任务数据收集模块、旧任务模型训练模块、新任务曲线生成模块和列车控制模块;所述控制方法包括:旧任务数据收集模块收集旧任务产生的多个自动驾驶曲线,然后按3种场景类型分为3个训练样本集,然后利用3个训练样本集分别训练得到3个可用模型,可用模型训练中结合对比学习和元强化学习的方法,然后在新任务中,根据新任务对应的场景类型选择相应的可用模型,通过对模型参数进行微调,快速生成新的驾驶速度曲线控制列车运行。采用本发明的控制方法,能在列车遇到新任务时快速、高效地生成列车的驾驶速度曲线,提高运营效率、减小经济损失。
Train control method based on meta reinforcement learning in multiple scenes
一种列车在多场景下基于元强化学习的控制方法
2024-11-12
Patent
Electronic Resource
Chinese
IPC: | B61L Leiten des Eisenbahnverkehrs , GUIDING RAILWAY TRAFFIC |
Control method of man-machine mixed driving intelligent train in multiple scenes
European Patent Office | 2023
|Train attitude control method based on DQN reinforcement learning
European Patent Office | 2020
|SYSTEM AND METHOD FOR REINFORCEMENT LEARNING-BASED TRAIN CONTROL
European Patent Office | 2024
|Train accurate parking control method based on deep reinforcement learning
European Patent Office | 2022
|