Reinforcement Learning with Knowledge Transfer for Residential Demand Response

With the electrification of society and the rising concern about the energy system's impact on the environment, there is an increasing need to efficiently operate the energy networks. Data-driven control techniques seem a promising alternative to conventional control mechanisms for this partial observable control problem. Combining building physical models with data-driven control mechanisms allows the latter to be trained in a realistic environment before deployment. The goal of this work is to investigate how multi-carrier energy systems (electric power, heat, ...) and their distributed controllers can benefit from this combination, and how machine learning can improve controller performance.

Publication year:2022

Accessibility:Open