Deep Reinforcement Learning for Traffic Light Timing Optimization

Existing inflexible and ineffective traffic light control at a key intersection can often lead to congestion due the complexity of dynamics, how find optimal timing strategy is significant challenge. This paper proposes optimization method based on double dueling deep Q-network, MaxPressure, Self-organizing lights (SOTL), namely EP-D3QN, which controls flows by dynamically adjusting duration in cycle, whether phase switched rules we set advance pressure lane. In each corresponds an agent, road entering divided into grids, grid stores speed position car, thus forming vehicle information matrix, as state agent. The action agent signal has four values. effective 0–60 s, phases switching depends its press set. reward difference between sum accumulated waiting time all vehicles two consecutive cycles. SUMO used simulate scenarios. We selected types evaluation indicators compared methods verify effectiveness EP-D3QN. experimental results show that EP-D3QN superior performance heavy flow scenarios, reduce travel vehicles, improve efficiency intersection.