The invention discloses a deep reinforcement learning method for recommending conflict-free parking spaces in real time, and the method comprises the following steps: taking vacant parking position data collected in real time as parking perception data, inputting the data into a reinforcement learning PPO model built on a VCS server, carrying out the feature extraction through a feature network, and predicting the social environment of a vehicle; obtaining corresponding strategy probability distribution of the vehicle through a strategy network, calculating a V value by using a value network, and updating the two networks until convergence; in the process, the VCS server calculates the award of the parking sensing state of the vehicle and provides a proper parking strategy for the vehicle by using the predicted next state of the vehicle; besides, the real situation is considered, an alternative scheme is provided for the vehicle by using a game rule, and the problems of difficult parking space searching and parking space conflict caused by vehicle increase and traffic jam in a real scene are solved.
本发明公开了一种实时推荐无冲突停车位的深度强化学习方法,该方法包括如下步骤:将实时收集的空置停车位置数据作为停车感知数据,输入在VCS服务器上搭建的强化学习PPO模型中,通过特征网络进行特征提取,预测车辆的社交环境;然后经过策略网络得到车辆相应的策略概率分布,并利用价值网络计算V值,进而对两个网络进行更新直至收敛;在此过程中,VCS服务器会计算车辆停车感知状态的奖励并利用预测得出的车辆下一状态为车辆提供合适的停车策略;此外,本发明考虑了真实情况,利用博弈规则为车辆提供备选方案,解决了真实场景下因车辆增加,交通拥堵导致的寻找停车位困难以及停车位冲突的问题。
Deep reinforcement learning method for recommending conflict-free parking spaces in real time
一种实时推荐无冲突停车位的深度强化学习方法
2023-03-21
Patent
Electronic Resource
Chinese
Automatic parking method based on deep reinforcement learning
European Patent Office | 2020
|Train accurate parking control method based on deep reinforcement learning
European Patent Office | 2022
|