The invention discloses a bottleneck area lane change control method and device based on deep reinforcement learning, and a medium, and the method comprises the steps: selecting a certain proportion of vehicles to enable the vehicles to change lanes in advance, thereby avoiding the traffic capacity reduction caused by the centralized lane change of vehicles close to a bottleneck area, achieving the averaging of the vehicle density of each lane through the design of a reward function, and improving the traffic capacity. Concentrated lane changing and frequent lane changing behaviors for pursuing speed are avoided to the maximum extent, and the stability of traffic flow in a bottleneck area is guaranteed. Besides, an optimal lane changing strategy is obtained through interactive updating training of the DQN and the environment, the optimal lane changing strategy can coordinate and control lane changing behaviors of vehicles in the bottleneck area, traffic jam caused by centralized lane changing of vehicles close to the bottleneck area is avoided by changing lanes in advance at the upstream of the bottleneck area, and the problem that the traffic capacity of the bottleneck area is reduced is effectively solved. Compared with a traditional traffic flow-based control scheme, the bottleneck area switching control method has the advantages that the average travel time of the vehicles is shortened, and the passing efficiency is improved.
本发明公开了基于深度强化学习的瓶颈区换道控制方法、设备及介质,通过选取一定比例的车辆使其提前换道来避免临近瓶颈区车辆集中换道而导致通行能力下降的发生,通过奖励函数的设计实现了每条车道车辆密度的平均,以最大限度避免了集中换道和为了追求速度的频繁换道行为,保证了瓶颈区交通流的稳定。此外,通过DQN网络与环境的交互更新训练,得到最优换道策略,该最优换道策略可协调控制瓶颈区车辆的换道行为,通过在瓶颈区上游进行提前换道,避免临近瓶颈区车辆集中换道导致交通拥堵的发生,有效地解决瓶颈区通行能力下降的问题。本发明的瓶颈区换到控制方法与传统的基于交通流控制方案相比,减少了车辆的平均行程时间,提高了通行效率。
Bottleneck area lane changing control method and device based on deep reinforcement learning and medium
基于深度强化学习的瓶颈区换道控制方法、设备及介质
2023-11-17
Patent
Electronic Resource
Chinese
IPC: | G08G Anlagen zur Steuerung, Regelung oder Überwachung des Verkehrs , TRAFFIC CONTROL SYSTEMS |
Harmonious Lane Changing via Deep Reinforcement Learning
IEEE | 2022
|European Patent Office | 2024
|Deep Reinforcement Learning Approach for Automated Vehicle Mandatory Lane Changing
Transportation Research Record | 2022
|