This repo contains scenario data corresponding to the different grid2op environments used in the project.
Corresponds to the basic environment (36 substations) from open sourced RTE interactiveAI use case will be reused (see predefined "l2rpn_icaps_2021" environment).
More details about this environment can be found in Grid2Op documentation.
This environment will use the same data as for the ai4realnet_small environment but will alter observations as seen by the agent: default approach is to use NoisyObservation class from Grid2op.
Other approaches will be tested during the project through dedicated agent developed by Task 4.2: Random perturbation agent, Gradient estimation perturbation agent, RL-based perturbation agent. It is also possible to consider that different backends can be used:
- a “light” backend, such as LightSim2Grid, can be used for training,
- a more “precise” backend (thus more representative of reality), such as PyPowsybl, can be used for evaluation.
This more advanced environment (118 substations) corresponds to L2RPN competition organized by RTE in 2023 ("l2rpn_idf_2023").
More details about this environment can be found in Grid2Op documentation.
This environment will use the same data as for the ai4realnet_large environment but will alter observations as seen by the agent. Default approach is to use NoisyObservation class.