Traffic congestion is an escalating concern., to which Traffic Signal Control (TSC) offers a promising solution. While intelligent TSC systems exhibit improved traffic efficiency on static networks, they often struggle to adapt to the dynamic nature of real-world scenarios. characterized by incidents such as acci-dents, road repairs, social events, etc. that disrupt traffic flow. To address this challenge, we propose a Drift-detection Guided Rein-forcement Learning (RL) based TSC (DDGRL-TSC) for dynamic road networks. DDGRL-TSC initiates a virtual RL process, iden-tifies environmental changes, instigates a prioritized sweeping RL process, and then launches a new RL process. This strategy allows for the effective integration of new information into the existing knowledge base, avoiding fixation on previously optimal policies, thereby reducing adaptation time to road network changes and enhancing overall traffic efficiency.
Incremental Reinforcement Learning with Prioritized Sweeping for Traffic Signal Control
2023-10-28
1314550 byte
Conference paper
Electronic Resource
English
Reinforcement learning-based traffic signal control
IEEE | 2024
|Deep Reinforcement Learning-based Traffic Signal Control
IEEE | 2023
|Air traffic control improvement using prioritized CSMA
IEEE | 2001
|