[Bldg-sim] CityLearn Challenge Sign-up Period is Now Open

Nagy, Gyorgy Zoltan nagy at utexas.edu
Tue Jan 7 10:00:34 PST 2020

Dear all,

We are now opening the period to sign up<https://docs.google.com/forms/d/e/1FAIpQLSf8PeqKqw9lzI7xSmjXqdTzzqbYdl3GrgOb7hpPtXETjQVlSg/viewform> for the CityLearn Challenge. Which will begin with the release of the first dataset on January 15th and is endorsed by International Building Performance Simulation Association (IBPSA). For more information, please visit the website <https://sites.google.com/view/citylearnchallenge/home> of the challenge or the Github<https://github.com/intelligent-environments-lab/CityLearn> repository.

Reinforcement learning (RL) has gained popularity in the research community as a model-free and adaptive control algorithm for the built-environment. Despite its potential, there are still open questions regarding its plug-and-play capabilities, performance, safety of operation, and learning speed. Yet, a lack of standardization in previous research has made it difficult to compare different RL algorithms with each other, as different publications aimed at solving different problems.

In an attempt to tackle these problems, we have organized this challenge using CityLearn, an OpenAI Gym Environment for the implementation of RL agents for demand response at the urban level. The environment allows the implementation of single-agent and multi-agent RL controllers. The RL agents can control domestic hot water (DHW) and chilled water storage devices in different sets of 9 simulated buildings in 4 different climate zones within the US. The RL agents must minimize a multi-objective cost function of 5 equally weighted metrics related to demand response and load shaping in an entire district of buildings.

The RL agents will be designed, tuned and pre-trained using a design dataset. There will be an evaluation and a challenge dataset with different buildings in different locations (but within the same climate zones as the design dataset). The agents will be provided with basic information about the buildings, their climate zone, and the similarities between their demand profiles. Participants will submit their agents to us for their evaluation on the evaluation dataset, and we will update a leaderboard with their scores. At the end of the challenge, on March 31st, participants will submit their agents for the final run on the challenge dataset, which will determine the winners. Although the design dataset will allow the teams to pre-train their controllers, some action exploration and not just exploitation is encouraged, as the evaluation and challenge datasets will be different than the design dataset. Only the design dataset will be public during the challenge.

Teams can sign up any time (max. 3 people per team). On January 15th, we will upload the design dataset to the Github repository. Teams can submit their agent(s) and get feedback from the evaluation dataset until March 15th, and they must submit their final agent(s) to be run on the challenge dataset by March 31st. The teams with the top performing agent(s) will be encouraged to submit their findings to a special issue in a reputed journal (in preparation), more details will be disclosed as soon as possible.

We appreciate your interest, and for any questions about the challenge please don’t hesitate to contact us on citylearn at utexas.edu<mailto:citylearn at utexas.edu> We are looking forward to seeing you sign up and obtaining state-of-the-art results!

Here are all the links again:

SignUp: https://docs.google.com/forms/d/e/1FAIpQLSf8PeqKqw9lzI7xSmjXqdTzzqbYdl3GrgOb7hpPtXETjQVlSg/viewform
Website: http://www.citylearn.net/
GitHub: https://github.com/intelligent-environments-lab/CityLearn
eMail: citylearn at utexas.edu<mailto:citylearn at utexas.edu>


Prof. Dr. Zoltan Nagy
Intelligent Environments Laboratory: http://nagy.caee.utexas.edu
Department of Civil, Architectural & Environmental Engineering
The University of Texas at Austin
Twitter: @Z0ltanNagy<https://twitter.com/Z0ltanNagy>

The CityLearn Challenge — Multi-Agent Reinforcement Learning for energy management in cities — http://www.citylearn.net
Organizing Member — UT Grand Challenge  — Whole Communities Whole Health:https://bridgingbarriers.utexas.edu/whole-communities-whole-health/
Organizing Member — BuildSys 2019: http://buildsys.acm.org/2019/
Guest Editor ­— Sustainability SI: Sustainable Technologies for Net Zero Energy Buildings (NZEB) https://www.mdpi.com/journal/sustainability/special_issues/Net_Zero_Energy_Buildings

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.onebuilding.org/pipermail/bldg-sim-onebuilding.org/attachments/20200107/568ea079/attachment.html>

More information about the Bldg-sim mailing list