(312h) Accelerating the Solution of Model Predictive Control Problems Via Machine Learning

Model predictive control (MPC) is widely used to control complex process systems. The online implementation of MPC requires the efficient solution of an optimization problem. Despite the significant advances in optimization theory and algorithms for a wide class of optimization problems, the solution of MPC problems remains challenging due to the nonlinear nature of most process systems and the limited computational budget.

In this talk, we focus on using machine learning (ML) to accelerate the solution of optimization problems that arise in MPC applications. Specifically, we show how ML can guide the selection of the best solution strategy and initialize cutting plane methods for mixed integer MPC. Finally, we discuss open problems/challenges that arise when applying ML techniques to accelerate MPC.