| 摘 要: 针对城市主干道公交专用道在部分时段利用不足与普通车道拥堵并存的问题，提出一种公交专用道动态共享策略，当智能网联车(CAV)具有换道动机且公交专用道满足换道安全条件时，控制CAV换入公交专用道行驶，提高道路利用效率。具体的，使用融合公交优先约束的多智能体近端策略优化(MAPPO)模型，利用MAPPO在训练阶段的多车交互信息，构造以交通流运行效率与公交专用道利用率为典型指标的奖励函数训练多车协同决策模型，引导车辆实现密度均衡、错峰借道与提前让行。仿真结果表明，相比规则模型MOBIL，MAPPO在不同公交流量(100/150/200 pcu/h)与CAV渗透率(0.1–0.9)下均能获得更高平均车速与更低平均延误，且车道密度分布更均衡，当公交流量为中等水平150 pcu/h、CAV渗透率0.9时，道路平均车速提升6%，CAV渗透率0.6时，道路平均延误降低249.11 s，同时车道密度分布更加均衡，说明此时MAPPO模型控制下的交通流已形成稳定协同，车辆实现错峰借道与提前让行。 |
| 关键词: 智能网联车，共享公交专用道，清空距离，换道行为，深度强化学习 |
|
中图分类号:
文献标识码:
|
| 基金项目: 国家自然科学基金项目（面上项目，重点项目，重大项目），海南热带海洋学院崖州湾创新研究院科研项目，海南热带海洋学院科研项目 |
|
| Research on Dynamic Sharing Strategy of Bus Lane in Intelligent Networked Environment Based on Deep Reinforcement Learning |
|
lvyuejing1, zhaoyibo1, fengyuqin2,3
|
1.Wuhan University of Science and Technology;2.Yazhou Bay Innovation Research Institute &3.International navigation College, Hainan Tropical Ocean University
|
| Abstract: To address the coexistence of underutilization of bus lanes on urban arterial roads during certain periods and congestion on general lanes, a dynamic sharing strategy for bus lanes is proposed. When connected and autonomous vehicles (CAVs) have the motivation to change lanes and the bus lane meets the safety conditions for lane changing, CAVs are controlled to switch into the bus lane to improve road utilization efficiency. Specifically, a multi-agent proximal policy optimization (MAPPO) model with integrated bus priority constraints is used. By leveraging the multi-vehicle interaction information during the training phase of MAPPO, a reward function is constructed with traffic flow operation efficiency and bus lane utilization rate as typical indicators to train the multi-vehicle collaborative decision-making model, guiding vehicles to achieve density balance, off-peak lane borrowing, and early yielding. Simulation results show that compared with the rule-based model MOBIL, MAPPO achieves higher average vehicle speeds and lower average delays under different bus flow rates (100/150/200 pcu/h) and CAV penetration rates (0.1–0.9). Moreover, the lane density distribution is more balanced. When the bus flow rate is at a medium level of 150 pcu/h and the CAV penetration rate is 0.9, the average road speed increases by 6%, and when the CAV penetration rate is 0.6, the average road delay decreases by 249.11 seconds. At the same time, the lane density distribution becomes more balanced, indicating that the traffic flow under the control of the MAPPO model has formed a stable collaboration, and vehicles achieve off-peak lane borrowing and early yielding. |
| Keywords: Key words: Intelligent network connected vehicles, sharing bus lanes, clearing distances, lane changing behavior, deep reinforcement learning |
