OPF-Gym

OPF-Gym is a Python library that provides reinforcement learning (RL) environments for learning the optimal power flow (OPF) problem.

It has three main contributions:

• Five benchmark environments for the RL-OPF problem to enable comparability of research. (see Benchmarks)

• Various pre-implemented environment design options for representing the OPF problem as an RL environment. (see Environment Design Options)

• Convenient creation of custom RL-OPF environments with advanced features like multi-stage OPF, security-constrained OPF, etc. (see Create Custom Environments)

The library uses the gymnasium RL environment API, pandapower for grid modelling and power flow calculations, and integrates the SimBench benchmarks power grids and time-series data by default.

All pandapower OPF variants can be represented as an RL environment by OPF-Gym. Additionally, advanced OPF problems like multi-stage OPF, security-constrained OPF, mixed continuous and discrete actions, stochastic OPF, etc. are possible as well.

Contact thomas.wolgast@uol.de for questions, feedback, and collaboration.

The repository can be found on GitHub.

---

Contents:

• Installation
• Getting Started
  • Loading and running a benchmark environment
  • Adjusting the environment design options
  • Define a custom RL-OPF environment
• Benchmarks
  • Economic Dispatch
  • Voltage Control
  • Load Shedding
  • Reactive Power Market
  • Maximize Renewable Feed-In
• The OpfEnv Base Class
  • Methods:
• Environment Design Options
• Create Custom Environments
  • Class Initialization
  • Pandapower network definition
  • Observation space definition
  • Action space definition
  • Advanced concepts
• Advanced Features by Examples
  • Multi-Stage OPF
  • Security-Constrained OPF
  • Mixed Continuous and Discrete Actions
  • Stochastic OPF
  • Partial Observability
• Support for Supervised Learning