TBMEnv - An Environment to Assess the Accuracy of Thermal Building Models under Realistic Conditions
TBMEnv is an program that can be used to evaluate and benchmark models of model-based thermal building optimization algorithms, which are most likely model-based reinforcement learning approaches. The main goal of the environment is to support the evaluation of thermal (machine learning) building models under realistic conditions as this cannot be achieved using available environments like BOPTEST or Sinergym. For more details and background please see the accompanying paper. In case you have no access to ACM DL a copy can be found here.
Begin by checking out the source code:
git clone https://github.com/fzi-forschungszentrum-informatik/tbmenv.git
cd tbmenvThis is code is written in Python. Given that Python and pip are installed, you can install the environment by typing:
pip3 install -e ./source/As an alternative this repository contains a docker-compose.yml specifying a docker runtime. Assuming that docker and docker compose are available it is possible to start the container with the following command:
docker compose up --buildThe second option may be more suitable, especially for those just wanting to play around with the demo notebooks provided.
Please see the usage example notebook.
We would like to thank the Institute for Automation and Applied Informatics of the Karlsruhe Institute of Technology for providing the data and information about the building that has been used as blueprint for Scenario 1. The dataset has been recorded in the research project "Flexkälte – Flexibilisierung vorhandener Kälteanlagen und deren optimierter Einsatz in einer Realweltanwendung" funded by the German Federal Ministry for Economic Affairs and Climate Action.
The measurement data utilized in Scenario 2 has been taken from here.
Please consider citing us if this environment and/or the accompanying paper was useful for your scientific work. You can use the following BibTex entry:
@inproceedings{10.1145/3600100.3625683,
author = {W\"{o}lfle, David and Vossberg, Samed Rouven and Schmeck, Hartmut},
title = {TBMEnv - An Environment to Assess the Accuracy of Thermal Building Models under Realistic Conditions},
year = {2023},
isbn = {9798400702303},
publisher = {Association for Computing Machinery},
address = {New York, NY, USA},
url = {https://doi.org/10.1145/3600100.3625683},
doi = {10.1145/3600100.3625683},
booktitle = {Proceedings of the 10th ACM International Conference on Systems for Energy-Efficient Buildings, Cities, and Transportation},
pages = {357–360},
numpages = {4},
keywords = {reinforcement learning, building control, building energy management, evaluation, benchmark, environment, smart building},
location = {Istanbul, Turkey},
series = {BuildSys '23}
}Other reference formats are provided here (export citation button).
Please open an issue here on GitHub for any question or remark regarding the implementation. Please feel free to contact David Wölfle for all other inquiries.
Code is copyright to the FZI Research Center for Information Technology and released under the MIT license.