After the success of the first iteration of ML4Physics challenge, come and engage with a new important real-world challenge to help tackle climate change by enabling massive renewable integration into the power grid! Your mission: develop new ML surrogate models to speed-up power-flow simulations by several order of magnitude with realistic criteria, in order to run advanced near real-time congestion risk assessment!

hero

# Competition overview

This challenge aims at promoting the use of ML based surrogate models to solve physical problems, and more specifically contributes to operate the power grid in the context of increasing level of uncertainty due to the integration of renewable energy sources. The power grid operators should be able to manage the security of large power grids by taking some appropriate actions allowing to avoid power lines overload. For this purpose, the operators require a huge number of simulations in near real-time to analyze the consequence of their actions and to assess the risk on the grid. The standard physical simulations are represented as nonlinear and nonconvex problem which are estimated using Newton Raphson solvers and are relatively costly.

The overall aim in this challenge is to build innovative ML surrogate models to compute the grid power flows, while finding the best trade-off between the inference precision and computation costs.

To evaluate the solutions, the competition rely on our recently proposed benchmarking framework called LIPS (Learning Industrial Physical Systems). This framework will be used to evaluate candidate solutions provided by the participants regarding significant criteria organized into 4 categories namely: ML related criteria, Physical Compliance criteria, industrial readiness and OOD generalization criteria. For each submitted solution, a global score will be computed based on the aforementioned criteria to rank it.

The help the development of new physics-informed ML models, the participants could use the NVIDIA MODULUS Framework, and benefit from NVIDIA support through a dedicated webinar at the beginning of the competition.

The participants should train and fine-tune their models on their own computers based on the provided dataset. A cluster of 8 NVIDIA RTX A6000 GPUs will be made available by Exaion for participants who may not have their own GPU resources. For the training phase, Baseline solutions will be made available to help participants.

More details about NVIDIA SDKs & RTX GPUs (opens new window) resources


# LIPS is a benchmarking framework developped to facilitate AI-based physical simulation evaluation

Physical simulations are at the core of many critical industrial systems. However, today's physical simulators have some limitations such as computation time, dealing with missing or uncertain data, or even non-convergence for some feasible cases. Recently, the use of data-driven approaches to learn complex physical simulations has been considered as a promising approach to address those issues. However, this comes often at the cost of some accuracy which may hinder the industrial use.

To drive the above mentioned new research topic towards a better real-world applicability, we propose a new benchmark suite "Learning Industrial Physical Simulations" (LIPS) to meet the need of developing efficient, industrial application-oriented, augmented simulators. The proposed benchmark suite is a modular and configurable framework that can deal with different physical problems.

More details on LIPS :

# Timeline

  • Competition kick-off April 16th 2024
  • Competition end date July 15th 2024

# Competition phases

# This competition will run over 3 phases:

  • Warm-up phase : participants can get familiar with provided material and the competition platform, make their first submissions and provide feedback to organizers. Based on this feedback, organizers can adjust and improve the competition for the next phase.
  • Development phase : participants will develop their solutions and will be able to test their already trained models against a provided validation dataset. They can also have access continuously to the global score corresponding the submitted solution.
  • Final phase : the organizers prepare the final ranking and official results. A dedicated event will be organized to announce the winners.


# Protocol

The competition will be hosted by the Codabench platform (opens new window). Participants will have to:

  1. create an account;

  2. download a starting kit to prepare their submission;

  3. upload on the Codabench platform their trained ML models. Then, the platform will use the LIPS framework to compute scores for the submission. The score will be published on the Codabench competition page and the participant will also have access to an additional page with the detailed metrics.



# Who can take part

Anyone interested in solving physical problems using ML technics is encouraged to participate in this competition. It could be a great opportunity to gather people from

ML and the Scientific computing communities to leverage the synergies between these two domains.



# Prizes

  • 🏆 1st Prize : 3000 €
  • 🥈 2nd Prize : 2000 €
  • 🥉 3rd Prize : 1000 €
# Special prizes:
  • Most accurate ML model (without speedup consideration) : 1000 €
  • Best student solution : 1000 €


# Organizer

A logo with the organizers will be posted here.

# Organization Team:

  • Mouadh Yagoubi (IRT SystemX)
  • Milad Leyli-Abadi (IRT SystemX)
  • Jean-Patrick Brunet (IRT SystemX)
  • Maroua Gmati (IRT SystemX)
  • Antoine Marot (RTE)
  • Jérôme Picault (RTE)
  • Asma Farjallah (NVIDIA)
  • Marc Schoenauer (Inria)

# Contact us

Email address: ml4physim-challenge@irt-systemx.fr
Competition Discord channel: https://discord.gg/RZaQkHfBtg
Competition codabench page: https://www.codabench.org/competitions/2378/
Competition website: https://ml-for-physical-simulation-challenge.irt-systemx.fr/powergrid-challenge/

# The competition is hosted on CodaLab and sponsored by IRT SystemX, RTE, NVIDIA and Exaion.