The invention relates to a train accurate parking control method based on deep reinforcement learning, and the method comprises the steps: collecting ATO stop data, and carrying out the preprocessing, and obtaining an expert data set X; performing simulation learning of behavior cloning based on the expert data set X, and initializing a strategy network of the train; training the strategy network on line through a deep determination strategy gradient method to obtain a deep optimization network; and outputting and storing the deep optimization network, and applying the deep optimization network to accurate parking control of the train. The method can achieve the following technical effects: ATO historical data of existing lines are fully utilized, and imitation learning is carried out offline; decoupling the coupling relationship between the PI control parameters and the vehicle characteristics by means of the generalization ability of the neural network; according to the technology, the real vehicle online debugging time of ATO software can be shortened, and the stop precision is improved; the strategy network and the value network of reinforcement learning can be adjusted along with the change of train characteristics, and the lifelong learning effect is achieved.
本申请涉及一种基于深度强化学习的列车精确停车控制方法,通过收集ATO停站数据,并进行预处理,得到专家数据集X;基于所述专家数据集X进行行为克隆的模仿学习,初始化列车的策略网络;通过深度确定策略梯度方法,在线训练所述策略网络,得到深度优化网络;输出并保存所述深度优化网络,并将所述深度优化网络用于列车精确停车控制。能够达到如下技术效果:充分利用既有线路的ATO历史数据,离线进行模仿学习;依靠神经网络的泛化能力,解耦PI控制参数与车辆特性的耦合关系;本技术能减少ATO软件真车在线调试时间,并提高停站精度;强化学习的策略网络和价值网络会随着列车特性变化而做出调整,实现了终生学习的效果。
Train accurate parking control method based on deep reinforcement learning
一种基于深度强化学习的列车精确停车控制方法
2022-12-20
Patent
Electronic Resource
Chinese
IPC: | G06N COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS , Rechnersysteme, basierend auf spezifischen Rechenmodellen / B61L Leiten des Eisenbahnverkehrs , GUIDING RAILWAY TRAFFIC / G06Q Datenverarbeitungssysteme oder -verfahren, besonders angepasst an verwaltungstechnische, geschäftliche, finanzielle oder betriebswirtschaftliche Zwecke, sowie an geschäftsbezogene Überwachungs- oder Voraussagezwecke , DATA PROCESSING SYSTEMS OR METHODS, SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL, SUPERVISORY OR FORECASTING PURPOSES |
Train temporary parking strategy optimization method based on deep reinforcement learning
European Patent Office | 2023
|Deep reinforcement learning method suitable for automatic parking of high-speed train
European Patent Office | 2022
|Accurate parking control method for newly opened line train
European Patent Office | 2023
|Automatic parking method based on deep reinforcement learning
European Patent Office | 2020
|