The invention discloses a deep reinforcement learning traffic signal control method based on multiple targets and multiple agents. The method provided by the invention comprises the following steps: sensing pedestrian crossing demands at an intersection based on video data, and extracting pedestrian crossing tracks; designing a state space and a reward function considering carbon emission and pedestrian crossing; and estimating the dynamic stock of the motor vehicles on the road. Aiming at the problems of traffic jam, environmental pollution and the like, the method reduces the parking frequency and waiting time of vehicles at the intersection while meeting the crossing requirements of the pedestrians at the intersection, reduces the vehicle exhaust emission at the intersection, and shortens the waiting time of the pedestrians.
本发明公开了一种基于多目标多智能体的深度强化学习交通信号控制方法。本发明提出的方法包括:基于视频数据的交叉口行人过街需求感知,提取行人过街轨迹;设计考虑碳排放和行人过街的状态空间和奖励函数;路段机动车动态存量估计。针对交通拥堵、环境污染等问题,该方法在满足交叉口行人的过街需求的同时减少车辆在交叉口处的停车次数和等待时间,降低交叉口处的车辆尾气排放,缩短行人的等待时间。
Deep reinforcement learning traffic signal control method based on multiple targets and multiple agents
基于多目标多智能体的深度强化学习交通信号控制方法
2024-09-10
Patent
Electronic Resource
Chinese
Deep Reinforcement Learning-Based Traffic Signal Control
ASCE | 2024
|Deep Reinforcement Learning-based Traffic Signal Control
IEEE | 2023
|Traffic signal control method based on deep reinforcement learning
European Patent Office | 2021
|