Home // ENERGY 2022, The Twelfth International Conference on Smart Grids, Green Communications and IT Energy-aware Technologies // View article

Training an Energy Management Simulation with Multi-Agent Reinforcement Learning

Authors:
Alexander Haemmerle
Kapil Deshpande
Philipp Moehl
Georg Weichhart

Keywords: Energy Management; Multi-Agent Reinforcement Learning; Photo-Voltaic; Battery Storage; Microgrid;

Abstract:
In this paper, we report on the application of multi- agent reinforcement learning to the development of a microgrid energy management simulation. The simulation is made up of energy producers and consumers as well as storage devices. We regard these components as agents that are trained in a shared environment with reinforcement learning. A significant share of energy production in the microgrid is provided by renewable energy sources with stochastic characteristics, e.g., photo-voltaic installations. The stochastic nature of such producers, as well as of consumers, is captured in energy consumption/production profiles that are used for training the respective agents. For our results, the agents have been trained with an actor-critic algorithm, using real-world energy profile data for photo-voltaic installations and industrial consumers in Austria. A centralised critic addresses the multi-agent nature of the energy management problem. Running what-if analyses is an application scenario for the trained simulation. In such analyses the effects of different microgrid configurations on energy management performance can be investigated. The presented work has been conducted in the context of the projects RESINET and Zer0p.

Pages: 22 to 29

Copyright: Copyright (c) IARIA, 2022

Publication date: May 22, 2022

Published in: conference

ISSN: 2308-412X

ISBN: 978-1-61208-967-6

Location: Venice, Italy

Dates: from May 22, 2022 to May 26, 2022