Highway traffic control with ramp metering utilizing variational autoencoders

We consider the problem of highway ramp metering with Model Predictive Control (MPC). While MPC is considered one of the most robust approaches for ramp metering, the optimization problem that has to be solved is often large, nonlinear, and nonconvex, making its real-time implementation prohibitive. To deal with this issue, we develop a novel Explicit Model Predictive Control (EMPC) scheme aiming to approximate the optimal control law obtained from the MPC technique in an offline manner. Traditionally, EMPC performs well under abundant high-quality training data, while its approximation accuracy deteriorates with limited data. The novelty in the approach we propose is the deployment of Variational Autoencoders (VAEs) combined with the EMPC scheme to minimize the mismatch between the actual and the approximated control law when the conventional EMPC proves inadequate to deal with this task under limited training data. To the author's knowledge, this is the first work that combines VAEs within a conventional EMPC scheme, to solve the ramp metering problem. Simulation results verify the superiority of our approach compared to the conventional EMPC scheme.