The invention discloses a reinforcement learning traffic signal control method based on internal reward enhancement. The method comprises the following steps: acquiring traffic state data under different conditions; designing and constructing a traffic signal agent; state feature extraction is carried out based on a multi-layer perceptron, and internal reward solution is carried out based on a determinant point process; traffic signal control is carried out after traffic signal optimization based on internal reward enhancement and independent reinforcement learning. The method encourages an intelligent agent to explore a new state by introducing a diversity enhanced internal reward method, and enables a reinforcement learning agent to learn a higher reward by evaluating the diversity between adjacent states based on a deterministic point process.
本发明公开了一种基于内在奖励增强的强化学习交通信号控制方法,包括获取不同情况下的交通状态数据;设计并构建交通信号智能体;基于多层感知机进行状态特征提取,并基于行列式点过程进行内在奖励求解;基于内在奖励增强和独立强化学习优化交通信号后进行交通信号控制。本发明通过引入多样性增强内在奖励方法,鼓励智能体探索新状态,并使强化学习代理能够通过基于确定性点过程评估相邻状态之间的多样性来学习更高的奖励。
Reinforcement learning traffic signal control method based on internal reward enhancement
一种基于内在奖励增强的强化学习交通信号控制方法
2024-05-17
Patent
Electronic Resource
Chinese
Wiley | 2020
|Reward Function Design in Multi-Agent Reinforcement Learning for Traffic Signal Control
TIBKAT | 2022
|Reinforcement learning-based traffic signal control
IEEE | 2024
|Traffic signal control method based on reinforcement learning
European Patent Office | 2023
|