v0.3.2

Important bugfix that prevents not-caught power flow calculation failures.

Full Changelog: v0.3.1...v0.3.2

v0.3.1

What's Changed

• Bugfix: Correct sampling in LoadShedding and MaxRenewable environments (only relevant for uniform and normal sampling)
• Introduce explicit state space definition (for data sampling)
• Simplify creation of multi-stage and security-constrained envs by adding separate classes for these cases to inherit from

Full Changelog: v0.3.0...v0.3.1

v0.3.0

What's Changed

• Make base class API clearer
• Simplify base class
• Add example for custom constrain definition
• Add Reward class to enable arbitrary reward functions and simplify the base class
• Enable pure constraint satisfaction problems without objective function
• Add option to use custom power flow and OPF solvers
• Add lots of type hinting
• Set-up CI/CD
• Replace setup.py with pyproject.toml
• Update to pandas 2.x (mainly reduce warnings)
• Add lightsim2grid dependency for faster power flow

Full Changelog: v0.2.0...v0.3.0

v0.2.0

What's Changed

• First version of serious documentation on readthedocs
• Add new Constraint class for easier adding of custom constraints
• Add support for piece-wise linear pandapower costs

Full Changelog: v0.1.1...v0.2.0

Version 0.1.1

Minor bugfix: Add __init__ to util.

Initial Release - Version 0.1.0

Initial Release - Version 0.1.0

This is the first official release and first stable version of the opfgym environment framework for learning the optimal power flow (OPF) with reinforcement learning (RL).

Features

• Gymnasium-compatible base class OpfEnv, which allows for easy creation of RL environments that represent OPF problems.
• Five benchmark RL-OPF environments, representing different OPF problems (Economic dispatch, voltage control, etc.)
• Various pre-implemented choosable environment design options, like different reward functions, observation spaces, etc.
• Several more advanced OPF features like multi-stage OPF, stochastic OPF, discrete actions, etc. (see examples)
• Allows for easy creation of labeled datasets for supervised learning from any OpfEnv environment.
• Fully compatible with the Gymnasium API.

Future Work

• Add more example environments to demonstrate the more advanced features.
• Add more convenience functionality to simplify tasks (e.g. action space definition or adding constraints).
• Add an advanced baseline OPF solver that can deal with discrete actions, multi-stage OPF, etc.
• Improve seeding according to Gymnasium API.