‼️Phase 1 Live + Updated Dataset Released

Flextrack Challenge 2025
#1

Hi everyone,

The Competition Phase of the FlexTrack Challenge is now live!

In the Warm-Up Phase, participants focused on detecting when demand response (DR) events occurred. In this phase, the goal shifts to quantifying how much energy was shifted or shed during those events. This step is key to improving coordination of assets like solar inverters, batteries, and HVAC systems for a cleaner, more affordable energy future.

Phase 1 Task

  • Regression task to predict a building site’s demand response capacity.
  • Uses the same training data as Warm-Up Round, now with added demand response capacity data.
  • Final rankings will be based on the private test set, visible only to organisers.

New Dataset Release (v0.2)

We’ve released v0.2 of the dataset, which applies to Phase 1 of the competition.

What’s new:

  • The updated test for this Phase is: flextrack-2025-public-test-data-v0.2.csv.

  • The test set now includes an additional column: Demand_Response_Flag. Some rows contain this value, while others are left empty (NaN).

  • Submission format update:

  • You must include Demand_Response_Flag in your submission. For rows where it’s empty, predict the flag.

  • Add a new column: Demand_Response_Capacity_kW for your capacity predictions.

  • Refer to the sample prediction file or random_submission.py in the starter kit for correct formatting.

Scoring:

Live Townhall Q&A

Join the live Townhall Q&A with organisers and domain experts to have your queries answered and get a head start for the new round.

Date: 22nd September Details: Townhall Info

Panellists:

  • Dr Emily Yap, University of Wollongong – Digital tools, IoT, and immersive tech for sustainable buildings.
  • Matt Amos, CSIRO – Digital infrastructure to improve energy flexibility.

Mark your calendars and bring your questions!

All The Best,
Team AICrowd

1 Like
#2

Hi , training data for warm up phase and phase 1 are same, is there something i am missing or is it a mistake. I’d appreciate any answer

#3

@yusuf_dogu : That is intended, and the training data are indeed the same. We released a new v0.2 (along with the rest of the items like the public test set, random predictions etc), to avoid confusion among participants.

1 Like