Data efficient reinforcement learning and adaptive optimal perimeter control of network traffic dynamics

The main objective is to devise a data-efficient adaptive perimeter controller for the MFD framework without relying on any knowledge on the traffic dynamics, i.e., model-free. Traffic networks are subject to model errors and demand uncertainties. Thus the MFD parameters are uncertain and even unknown. We aim to circumvent the requirement of perfect system information when designing the optimal perimeter controller. Traditional data-driven methods such as RL lack data efficiency and do not consider the heterogeneity in real-time data resolution. We try to enhance the data efficiency of the RL approach and make it robust to the time-varying data resolution.

Recommended citation: Chen, C., Huang, Y. P., Lam, W. H., Pan, T. L., Hsu, S. C., Sumalee, A., Zhong, R. X. (2022). "Data efficient reinforcement learning and adaptive optimal perimeter control of network traffic dynamics." Transportation Research Part C: Emerging Technologies. 142, 103759. https://doi.org/10.1016/j.trc.2022.103759