Reinforcement Learning and Adaptive Optimal Control of Congestion Pricing

Abstract

The increasing road traffic congestion has urged researchers to look for solutions to tackle the problem. Many different interventions reduce traffic jams including, optimizing traffic-lights, using video surveillance to monitor road conditions, strategic road network resilience, and congestion pricing. This paper uses a nonlinear model for dynamic congestion pricing, considering manual-toll and automatic toll lanes using wireless communication technologies. The model can adjust the traveling demand and improve traffic flow performance by charging more for entering express lanes. We linearize the model about the equilibrium states and propose a reinforcement learning-based adaptive optimal control approach to learn the optimal control gain of the linearized model. Further, we rigorously show that the developed optimal controller can ensure the stability of the original nonlinear closed-loop system by making its output asymptotically converge to zero. Finally, the proposed approach is validated by numerical simulations.