基于深度Q-Learning的信号灯配时优化研究

Research on Signal Timing Optimization Based on Deep Q-Learning

交通信号配时具有不确定性,合理确定城市交通信号配时对提升交通系统的运行效率至关重要。随着社会的不断进步和车辆的不断增长,传统交通流管理系统亟待改进,先进的城市交通控制系统是提高城市交通运行效率的重要途径之一,同时也是城市现代化的一个重要标志。一般的静态控制方法对交叉口的红绿灯配时策略能够起到一定的作用,但是效果不明显,因此有必要结合时下热门的深度强化学习来解决交通信号灯配时问题。通过基于深度Q-Learning的配时方案来优化交叉口的通行问题,选择正确的红绿灯配时从而最大化交通效率。基于经验回放机制,运用深度神经网络进行网络训练和预测输出。实验结果表明,改进的深度强化学习方法相较于传统静态控制方法的控制效果更佳。

The timing of traffic signal is uncertain.Reasonable timing of urban traffic signal is quite important to improve the operation efficiency of the traffic system.With the continuous progress of the society and the growth of vehicles,it is necessary to improve the traditional traffic flow management system.Advanced urban traffic control system is one of the important ways to improve the efficiency of urban traffic operation,and it is also an important symbol of urban modernization.The general static control method can play a certain role in the traffic light timing strategy,but the effect is not obvious.Therefore,it is necessary to combine the current popular deep reinforcement learning to solve the problem of traffic signal timing.The timing scheme based on deep Q-learning is adopted to optimize the traffic problems at intersections,and the correct timing of traffic lights is selected to maximize traffic efficiency.Based on experience replay,the deep neural network is used to train the network and predict the output.The experiment shows that the improved deep reinforcement learning method is more effective than the traditional static control method.