The invention discloses a traffic light control method based on deep reinforcement learning and inverse reinforcement learning, and the method comprises the steps: firstly building a Markov decision model of a traffic light control system, and building a traffic light control frame based on deep reinforcement learning according to an existing deep network model; the innovation point of the method is that a relative entropy inverse reinforcement learning algorithm is introduced to optimize the reward function design. According to the system state transition trajectory generated by expert decision, decision logic, namely a hidden reward function, contained in the expert is extracted through an inverse reinforcement learning algorithm, effective utilization of expert experience is achieved, and the algorithm has good robustness for noise in the expert trajectory. According to the method, the effect better than that of a traditional control scheme can be achieved in balanced traffic flow and non-balanced traffic flow scenes of a single intersection, and the control performance of a deep reinforcement learning algorithm is further improved.
本发明公开了一种基于深度强化学习和逆强化学习的交通灯控制方法,首先建立交通灯控制系统的马尔科夫决策模型,并依据现有深度网络模型,搭建基于深度强化学习的交通灯控制框架。本发明的创新点在于引入了相对熵逆强化学习算法以优化奖励函数设计。根据专家决策生成的系统状态转移轨迹,通过逆强化学习算法提取专家内含的决策逻辑,即隐藏奖励函数,实现了对专家经验的有效利用,算法对专家轨迹中的噪声具有较好的鲁棒性。本发明能够在单个交叉路口的均衡车流和非均衡车流场景下,取得优于传统控制方案的效果,并进一步提升深度强化学习算法的控制性能。
Traffic light control method based on deep reinforcement learning and inverse reinforcement learning
一种基于深度强化学习和逆强化学习的交通灯控制方法
2023-03-07
Patent
Electronic Resource
Chinese
Deep reinforcement learning traffic light control method
European Patent Office | 2024
|Traffic Light Control Using Reinforcement Learning
IEEE | 2024
|Deep Reinforcement Learning-based Traffic Signal Control
IEEE | 2023
|Intelligent Traffic Light via Policy-based Deep Reinforcement Learning
Springer Verlag | 2022
|