Model:Multi-agent reinforcement learning-based energy hub model

From HUES Platform Wiki

To edit the entries in this page, select "Edit with form" in the "Page tools" menu at the top of the page.

General information

Description This is a multi-agent-based implementation of an energy hub model, loosely based on the generic energy hub model configuration. The model identifies an optimal dispatch schedule for technologies in a distributed multi-carrier system. The model was developed to test the implementation of a multi-agent reinforcement learning-based approach to energy hub modeling as a possible alternative to MILP under certain conditions. The model implements two different reinforcement learning algorithms, a Q-learning algorithm and a Continuous Actor Critic Learning Automaton (CACLA) algorithm. Both the Q-learning and CACLA implementations of the model are able to identify near-optimal dispatch solutions for the case without storage. For the case with storage, neither approach is able to identify a near-optimal solution.
Download URL
Authors Andrew Bollinger
Required software Python 3
Related publications L.A. Bollinger and R. Evins. 2016. Multi-agent reinforcement learning for optimizing technology deployment in distributed multi-energy systems. EG-ICE Workshop. Krakow, Poland.
Licence type MIT License
Tags Energy hub model, Agent-based model, Reinforcement learning
Accessible to Public


Documentation may be uploaded, linked or directly entered into the wiki.

Documentation URL(s):
Documentation file: File:EGICE Bollinger Evins.pdf
Documentation page: