Gym-ANM: Reinforcement learning environments for active network management tasks in electricity distribution systems

• Software for training reinforcement learning agents to control distribution grids. Provided as customizable Gym Open AI environments. Results on a test system suggest RL algorithms are suited such tasks. Active network management (ANM) of electricity networks include many complex stochastic sequential optimization problems. These problems need be solved integrating renewable energies and distributed storage into future electrical In this work, we introduce Gym-ANM, framework designing (RL) environments that model ANM tasks in networks. provide new playgrounds research the do not require an extensive knowledge underlying dynamics systems. Along with releasing implementation introductory toy-environment, ANM6-Easy, designed emphasize common challenges ANM. We also show state-of-the-art can already achieve good performance ANM6-Easy when compared against predictive (MPC) approach. Finally, guidelines create Gym-ANM differing terms (a) topology parameters, (b) observation space, (c) modeling processes present system, (d) set hyperparameters influencing reward signal. downloaded at https://github.com/robinhenry/gym-anm .