基于多智能体团队强化学习的交通信号控制

城市的区域交通信号协调系统是一个十分复杂的系统,难以建立准确的数学模型,通过引入主-从式团队强化学习方法于区域交通信号协调控制,就可以根据实时的交通状态信息动态来进行决策,自动地适应环境以便取得更好的控制效果.由于问题状态空间太大且难以直接存储和表示,采用径向基函数神经网络进行值函数近似.通过训练自适应非线性处理单元,达到较好的近似表示效果,解决了多个交叉路口的交通信号协调控制问题.通过仿真实验,结果表明该方法的控制效果明显优于单点控制策略.

Urban Traffic Signal Control Based on Multi-agent Team Reinforcement Learning Methods

Urban traffic signal coordination control system is very complicate,so it is very difficult to build a precise mathematical model for it.In this paper,we use multi-agent team reinforcement leaning algorithm to control the traffic signal,thus the decision can be made dynamically according to real-time traffic status information,and the change of environment can be adapted automatically.As the state space is too big to be stored and expressed directly,we apply radial basis function neural network to approximate the value function.By training adapted non-linear processing unit,the approximation is improved and thus the coordination control of traffic signal at multi junctions is solved.The simulation results show that the effectiveness of the new control algorithm is obviously better than single agent method.