引入谈判博弈的Q-学习下的城市交通信号协调配时决策

由于城市交通路网中交叉口间交通信号决策是相互影响的,并且车联网技术使得交叉口交通信号配时agent间能进行直接交互,此决策问题可用博弈框架来描述。建立了城市路网中相邻交叉口间交通流关联模型,通过嵌入谈判博弈模型来设计Q-学习方法,此方法中利用谈判参考点来进行配时行为的选择。仿真实验分析表明,相对于无协调的Q-学习算法,谈判博弈Q-学习取得更好的控制效果和稳定性能。谈判博弈Q-学习在处理交通拥挤及干扰交通流时,能根据交通条件灵活地改变交通信号配时决策,具有较强的适应能力。

Urban Traffic Signal Timing Decision based on Q-learning with Negotiation Game Mechanism

Because the traffic signal decision between intersection in urban traffic network is interactional, and Sinternet of vehicles can make the intersection traffic signal agent interact directly, this decision problem can be described by the game framework. A traffic flow correlation model between adjacent intersections in urban traffic network was established, and Q- learning method was designed by embedding negotiation game model where negotiation reference point was used to choose timing behavior. The simulation experiment shows that the negotiation game Q-learning achieves better control effect and convergence performance compared with the uncoordinated Q-learning algorithm.When dealing with disturbing and congested traffic flow, negotiation game Q- learning has the flexibility to change the traffic signals according to the traffic conditions and necessity.