Skip to content

Latest commit

 

History

History
51 lines (42 loc) · 2.37 KB

README.md

File metadata and controls

51 lines (42 loc) · 2.37 KB
Raw

l2rpn_2023_alert

Train an alert module for a grid2op agent, which sends alerts if the grid is likely to collaps due to agent actions.

Any agent compatible with the grid2op framework can be specified, and is ran to collect data. Contingencies are simulated, and the policy is ran for a specified number of steps. Data on wether the agent managed to save the grid or not, along with failure timestep, is registered.

The collected data is then used to create a learning target Y for a supervised training model. The input data X is extracted from a list of grid observations.

At evaluation, the trained model is used to raise alerts, and should be attached to the agent used to collect the data.

Usage

  1. Gather data for supervised learning:
  • Open data_collector.py and configure 'main()' to load your desired agent (grid controller)
  • Run the file
  • 💡 data_collector.py can also be run in parallel, check out parallel_data_collection.py if interested
  1. Pre-process data and create train-test set:
  • Open pre_process_data.ipynb and run the relevant steps
  • ⚠️ You must run the code which creates a folder containing:
    • X_train.csv
    • Y_train.csv
    • X_test.csv
    • Y_test.csv
  1. Train model to predict grid survival:
  • Open either models_keras.py (MLP models) or models_xgboost.py (boosted tree models)
  • In 'main()', update path to the folder containing your training and test data
  • Optionally configure model parameters, training settings etc
  • Run file
  1. Evaluate the performance of your alert module:
  • Open eval.ipynb
  • Run and save some scenarios
  • Load all results or perform case study
  • 💡 Running scenarios can take some time. One can run EvalRunner directly (see eval_runner.py)

Installation

  1. Clone respository
  2. Create a virtual environment
  • python3 -m venv venv_alert
  1. Activate venv
  • source venv_alert/bin/activate
  1. Cd to repository
  • cd l2rpn_2023_alert
  1. Install requirements
  • pip install -r requirements.txt

Done!

License

Copyright (c) [2023-2024], RTE (https://www.rte-france.com)

See LICENSE file

This agent code is part of L2RPN (Learning To run a Power Network) Open-Science initiative, which aims at accelerating the development of AI solutions for power grid operations management.