Loading
Phase I (Warm Up Round): Completed Phase II: Completed Phase III: Completed
32.4k
643
111
1805

📓 NeurIPS 2022: CityLearn Challenge Starter Kit

wave Welcome to NeurIPS 2022: CityLearn Challenge 

trophy Community Contribution Prize 

busts_in_silhouette Looking for teammates?

🕵️ Introduction

Buildings are responsible for 30% of greenhouse gas emissions. At the same time, buildings are taking a more active role in the power system by providing benefits to the electrical grid. As such, buildings are an unexplored opportunity to address climate change. Energy storage devices such as home batteries can reduce peak loads of the grid by shifting the energy use of buildings to different times. Solar photovoltaic generation can reduce the overall demand to the grid while also reducing emissions. However, all these resources must be carefully managed simultaneously in many buildings to unlock the full energy potential and reduce homeowners' costs.

The CityLearn Challenge 2022 focuses on the opportunity brought on by home battery storage devices and photovoltaics. It leverages CityLearn, a Gym Environment, for building distributed energy resource management and demand response. The challenge utilizes 1 year of operational electricity demand and PV generation data from 17 single-family buildings in the Sierra Crest home development in Fontana, California, that were studied for Grid integration of zero net energy communities.

Participants will develop energy management agent(s) and their reward function for battery charge and discharge control in each building to minimise the monetary cost of electricity drawn from the grid and the CO2 emissions when electricity demand is satisfied by the grid.

Electricity

📚 Further Reading on Zero Net Energy Communities

Here is a list of articles and reports on the topic of Zero Net Energy Communities. Note: you don't need to read/understand all these to participate in the challenge.

📑 Problem Statement

Challenge participants are to develop their own single-agent or multi-agent RL policy and reward function for electrical storage (battery) charge and discharge control for the buildings with the goal of reducing;

  1. district electricity cost; and
  2. district CO2 emissions.

A single-agent setup will mean one policy is used to control all building batteries whereas multi-agent setup will mean each building's battery is controlled using a unique policy. However in the multi-agent scenario, agents are allowed to share information about their observations.

To ensure that occupant comfort is guaranteed at all times, the electric load of the building will not change. However, the agents must learn when to use electricity directly from the on-site solar panels, when to charge/discharge the battery, and when to rely on the grid. If they rely on the grid, they should learn to use it when it's cheap and/or with low carbon content.

Participants can also employ MPC and RBC methods as well.

Model Predictive Control (MPC) is an advanced method of process control while satisfying a set of constraints. It solves an online optimization algorithm to find the optimal control action that drives the predicted output to the reference.

Rule Based Control (RBC) is a rule-based system that applies human-made rules to store, sort and manipulate data. RBC methods are based on a set of predefined rules to control the system. Make your first submission using this example.

💾 Dataset

The 17-building dataset is split into training, validation and test portions. During the competition, participants will be provided with the dataset of 5/17 buildings to train their agent(s) on. This training dataset is automatically downloaded when the starter kit is forked and contains the following files:

  1. schema.json: JSON model of the environment that is loaded at runtime. Participants are encouraged to manipulate this file to assess the impact of parameters on the environment and, policy and reward function design. However, only the as-provided version of schema.json will be used for evaluation.
  2. weather.csv: 8,760 observations (one year) of actual meteorological weather data for the buildings' location used to provide observation values to the environment.
  3. carbon_intensity.csv: 8,760 observations (one year) of actual CO2 emission rate from grid mix.
  4. pricing.csv: 8,760 observations (one year) of Time-of-Use (TOU) electricity cost.
  5. Building_1...5.csv: 8,760 observations (one year) of each building's time series data including, temporal (calendar) variables, end-use demand, solar generation and indoor environment variables.

Refer to the dataset and schema files.

🚀 Submission

Make your first submission using the starter kit🚀!

🖊 Evaluation Criteria

Participants' submissions will be evaluated upon an equally weighted sum of two metrics at the aggregated district level where district refers to the collection of buildings in the environment. The metrics include 1) district electricity cost, $C_\textrm{entry}$ and 2) district CO2 emissions, $G_\textrm{entry}$ with the goal of minimizing the sum of each metric over the simulation period, $t=0$ to $t=n$ and $e$ episodes. The simulation period is 8,760 time steps i.e. one year, and participants can train on as many episodes of the simulation period, $e$, as needed. $C_\textrm{entry}$ is bore by the individual buildings (customers) and $G_\textrm{entry}$ is an environmental cost. Each metric is normalized against those of the baseline where there is no electrical energy storage in batteries ($C_\textrm{no battery}$, $G_\textrm{no battery}$) such that values lower than that of the baseline are preferred. Participants are ranked in ascending order of $\textrm{score}$.

 

In Phase I, the leaderboard will reflect the ranking of participants' submissions based on the 5/17 buildings training dataset.

By Phase II, the leaderboard will reflect the ranking of participants' submissions based on an unseen 5/17 buildings validation dataset as well as the seen 5/17 buildings dataset. The train and validation dataset scores will carry 40% and 60% weights, respectively in the Phase 2 score.

Finally in Phase III, participants' submissions will be evaluated on the 5/17 buildings training, 5/17 validation and remaining 7/17 test datasets. The train, validation and test dataset scores will carry 20%, 30% and 50% weights, respectively in the Phase 3 score.

The winner(s) of the competition will be decided using the leaderboard ranking in Phase III.

Pikachu

📅 Timeline

Phase I (Warm Up Round, July 18 - Aug 15, 2022)

This phase allows participants to familiarize themselves with the competition environment and raise issues bordering source code, training data quality and documentation to be addressed by the organizers. We will also create a solution example so participant can see their submissions show up on the leaderboard.

Phase II (Validation Round, Aug 15 - Sep 30, 2022)

This phase marks the beginning of the entry evaluation and ranking of submitted entries against the training + validation dataset. Participants will be able to see how each of their submission ranks against each other and how their latest submission ranks against other participants' submissions.

Phase III (Test \& Winner Round, Oct 1st - Oct 31st, 2022)

This phase marks the training + validation + test dataset's evaluation of submitted entries. At the end of this phase, collection of new submissions will be halted and the final leader board and selected winners will be published.

🏆 Prizes

This challenge has Leaderboard Prizes, Community Prizes, and Co-authorship Prizes

Leaderboard Prizes

Top three teams or participants on leaderbaord in Phase 3 will receive the following prizes.

  • 🥇 USD 8000 1st on leaderboard
  • 🥈 USD 5000 2nd on leaderboard
  • 🥉 USD 2000 3rd on leaderboard

Co-Authorship

In addition to the cash prizes, we will invite the top three teams to co-author a summary manuscript at the end of the competition. At our discretion, we may also include honourable mentions for academically interesting approaches, such as those using exceptionally little computing or minimal domain knowledge. Honourable mentions will be invited to contribute a shorter section to the paper and have their names included inline.

Community Contribution Prize: ACM SIGEnergy travel grants to ACM BuildSys or ACM e-Energy

The ACM Special Interest Group on Energy Systems and Informatics (SIGEnergy) sponsors three travel grants for up to USD 1000 each to attend ACM BuildSys or ACM e-Energy in 2023.

The Community Contribution Prizes will be awarded based on the discretion of the organizers, and the popularity of the posts (or activity) in the community (based on the number of likes ❤️) - so share your post widely to spread the word!

The prizes typically go to individuals or teams who are extremely active in the community, share resources - or even answer questions - that benefit the whole community greatly!

You can make multiple submissions, but you are only eligible for the Community Contribution Prize once. In case of resources that are created, your work needs to be published under a license of your choice, and on a platform that allows other participants to access and use it.

Notebooks, Blog Posts, Tutorials, Screencasts, Youtube Videos, or even your active responses on the challenge forums - everything is eligible for the Community Contribution Prizes. We are looking forward to see everything you create!

📱 Contact

🤝 Sponsors

sponsor_logo

Participants

Notebooks

See all
Going below 1.0 score with stablebaseline3
By
adrien_forbu
5 months ago
2
CommNet and DDPG
By
mark_haoxiang
6 months ago
1
9
0
Baby Steps
By
Chemago
6 months ago
0