A Model Predictive Control scheme for freeway traffic systems based on the Classification And Regression Trees methodology

Traffic control algorithms based on optimization may not always be applicable on-line, since they are computationally demanding and may not always comply with the typical sampling times adopted for traffic systems. In this paper, we propose a new approach to freeway traffic control based on the Classification And Regression Trees (CART) methodology. In our approach, a standard centralized receding horizon model predictive controller is replaced with a controller based on a set of regression trees, trained in order to reproduce the behaviour of the original controller. The result is a controller which does not need to solve an optimization problem at each time step. This makes it adequate for the on-line usage. The effectiveness of the proposed control approach, designed relying on a macroscopic model, is evaluated in simulation, on a microscopic model of the Grenoble South Ring developed on the basis of real data.