基于分布式强化学习的高速公路控制模型

A Freeway Traffic Flow Control Model Based on Distributed Reinforcement Learning

针对公路交通流非线性、不确定性和模糊性特点,提出了面向控制的交通网络宏观动态离散模型,并且引入分布式强化学习来解决交通网络的控制与诱导问题。以传统网络交通流模型Metanet为基础,对其作了改进,引入起讫点的因素到模型中,提出基于OD的网络交通流动态模型Metanet-OD。根据交通网络的特点,将分布式强化学习DRL引入到交通网络中,进行匝道控制和可变显示牌的诱导控制,设定了强化学习的动作空间,并给出了DRL算法。在仿真试验中对控制效果进行了验证。

A control-oriented macroscopic dynamic traffic flow discrete model applicable to the nonlinear,uncertain,fuzzy system of freeway traffic was proposed and discussed.Distributed Reinforcement Learning（DRL） was introduced to control and guide the traffic system.The traditional freeway traffic model Metanet was upgraded to an improved Metanet-OD within which the origins and destinations of freeway traffic was taken into account.The DRL was used in ramp metering and VMS guidance for freeway network.The actions space of DRL was designed and the DRL algorithm was presented.The control efficiency of the proposed model was also verified with simulation.