Multi-intersection traffic signal control: A decentralized MPC-based approach

Abstract

Traffic signal control is considered as one of the most important urban traffic management tools, due to its effectiveness in reducing traffic congestion, resulting in smoother and more secure vehicle flows. This work proposes a decentralized Model Predictive Control (MPC) strategy for the minimization of the queue length in a multi-intersection road network. Specifically, we show that our efficient linear formulation enables real-time control of the intersections’ signals, while taking into account safety constraints and pedestrian requests. A novel hyper-parameter tuning algorithm for decentralized MPC (based on Bayesian Optimization) is also proposed. The method is finally tested on a microscopic traffic simulator faithfully reproducing the layout of a real multi-intersection framework in Monza, Italy, fed with real traffic profiles. Simulation results illustrate the effectiveness of the proposed control approach, which can be easily scaled up to larger networks by keeping comparable performance with the state-of-the-art centralized methods.