ENBIS Webinar by Thijs Peirelinck: "Reinforcement learning with knowledge transfer for residential demand response"

4 May 2021; 12:30 – 13:30

Reinforcement learning with knowledge transfer for residential demand response

Thijs Peirelinck (KU Leuven, Belgium)


This webinar is part of a webinar series about new challenges for statistics facing the ecological transition and mainly focused on energy.

Abstract:

The increasing share of renewable energy sources in our power system comes with challenges, in part due to their variability and intermittency. The research community has considered Demand Response (DR) as a method to mitigate these challenges. Besides battery energy storage, Thermostatically Controlled Loads (TCLs) have been proposed to offer demand flexibility. A common challenge with DR is developing scalable control algorithms. State-of-the-art control methods incorporate Machine Learning (ML). The presentation will highlight our work using reinforcement learning for these residential DR applications with TCLs. While reinforcement learning has shown promising results, it also has proven to be sample inefficient. The ML community has addressed this with transfer learning. This has the potential to considerably reduce training requirements, increase learning rates and improve asymptotic performance once large amounts of data have been gathered. Recently, this trend has become more widespread and also found its way to DR applications. We will present our TCL models, the DR Markov decision process, and the reinforcement learning algorithms used to solve them, together with the results.

Short Bio:

Thijs Peirelinck is currently a Ph.D. student at the Electrical Energy and Applications (ELECTA) research group of KU Leuven, Belgium. His main research interest is residential Demand Response. Thijs obtained his BSc from KU Leuven, Belgium in 2014. He then moved on to do his master’s degree in energy engineering at KTH Stockholm, Sweden, and INP Grenoble, France. After this, he pursued an advanced master’s degree in Artificial Intelligence at KU Leuven, which he obtained in 2017.