The invention provides a multi-scene parking control method for an unmanned train. The multi-scene parking control method is characterized in that related hardware comprises a train control dispatching center, a central cloud computing module and a plurality of edge cloud computing modules; the control method comprises the steps that the edge cloud computing module collects training data, the center cloud computing module adopts a knowledge distillation method to train and lighten a normal parking deep learning model, and the edge cloud computing module and the center cloud computing module are combined to adopt a model aggregation method to train a temporary parking deep learning model. And each edge cloud computing module respectively controls normal parking and temporary parking of the train running on the corresponding sub-line by using the normal parking deep learning model and the temporary parking deep learning model. By the adoption of the control method, multi-scene parking control over the unmanned train can be achieved, the parking precision can be improved, and the communication pressure can be reduced.
本发明提供了一种无人驾驶列车的多场景停车控制方法,其特征在于:涉及的硬件包括列车控制调度中心、中心云计算模块和多个边缘云计算模块;所述控制方法包括:边缘云计算模块采集训练数据,中心云计算模块采用知识蒸馏的方法训练并轻量化正常停车深度学习模型,边缘云计算模块与中心云计算模块相结合采用模型聚合的方法训练临时停车深度学习模型,各个边缘云计算模块利用正常停车深度学习模型和临时停车深度学习模型分别控制对应子线路上运行列车的正常停车和临时停车。采用本发明所述的控制方法,能实现对无人驾驶列车的多场景停车控制,还能提高停车精度、减小通信压力。
Multi-scene parking control method for unmanned train
一种无人驾驶列车的多场景停车控制方法
2023-08-11
Patent
Elektronische Ressource
Chinesisch
Train turn-back rail unmanned accurate parking method
Europäisches Patentamt | 2024
|Precise unmanned vehicle parking method for port crane scene
Europäisches Patentamt | 2023
|