摘要
在城市交通环境下,通过分析控制方法灵活性与稳定性的关系,提出一种考虑网络稳定性的多智能体强化学习控制方法。该方法将稳定状态引入信号控制决策模块,建立稳定规则库,在基本多智能体强化学习控制系统上,设置了一套独立运行的稳定监督装置,对违反稳定规则的控制策略进行校正,以约束控制方法灵活性的方式提升其稳定性,以监督控制的形式实现了多智能体强化学习控制。在时变交通流场景下,以典型路网进行VISSIM仿真试验。结果表明:基于稳定监督控制的多智能体强化学习控制方法提高了算法的运行效率,同时保证了控制效果,适用于复杂交通网络。
In the urban traffic environment,a multi-agent reinforcement learning control method considering network stability is presented by analysis of the relationship between the flexibility and stability.This method introduces stable state into the decision module and sets up a set of stability rules.An independent operating mechanism is proposed based the basic multiagent reinforcement learning control system.The function of this mechanism is to calibrate the strategy,which violates the stability rules.Then,the stability of the constraint control method is improved,and the multi-agent reinforcement control is realized in the form of supervisory control.Under traffic demand of time-varying scene,VISSIM simulation is conducted on the typical road network.The result shows that supervision control based multi-agent reinforcement learning method improves the efficiency of the algorithm and ensures the control effect.It is applicable for the complex traffic network.
作者
张轮
张希雨
夏凡
赵文文
ZHANG Lun;ZHANG Xiyu;XIA Fan;ZHAO Wenwen(Key Laboratory of Road and Traffic Engineering,Ministry of Education,Tongji University,Shanghai 201804,China)
出处
《交通与运输》
2020年第4期86-91,共6页
Traffic & Transportation
关键词
交通工程
交通信号
多智能体强化学习
Q学习
网络稳定性
监督机制
Transportation engineering
Traffic signal
Multi-agent reinforcement learning
Q-learning
Network stability
Supervision mechanism
