Midterm colloquium Nathan van der Strate

Is it possible to integrate reinforcement learning and model predictive control to yield a high performance load frequency controller under increasing levels of uncertainties, introduced by renewable energy sources?

Supervisor: Prof.dr.ir. Bart De Schutter 

Abstract:

With the increasing penetration of renewable energy sources in the power system network in recent years, maintaining grid stability has become increasingly more difficult. Intermittency of power generation and the uncertainty of renewables lead to frequency deviations which cause degradation or damage to electrical appliances and potential black-outs. Control techniques that can deal with uncertainty include data-driven and machine learning based control approaches, which often lack interpretability, safety guarantees and constraint handling capabilities. A novel approach integrating model predictive control and reinforcement learning in the multi-agent setting is applied to the power system network to regulate the frequency whilst adhering to constraints.