Developing a Reinforcement Learning model for energy management of microgrids in Python

Abstract

Microgrids provide integrating platforms for distributed generating sources. Therefore, continuous controlling and monitoring of the microgrids are essential to balance power fluctuations, intermittency, etc. that are introduced by renewable generation sources. Agent-based distributed control systems have been introduced for microgrid control. As a novel approach learning ability is introduced to the agents in the system with the integration of reinforcement learning with power systems. This paper highlights how to determine the applicability of reinforcement learning for certain optimization problems together with problem mapping to the general reinforcement learning model. The goal of the application of reinforcement learning is to minimize the dependency of the microgrid on main grid while ensuring the maximum utilization of renewable energy generation. This energy management model is simulated using environment and agent class modelling using python programming. In addition to that, an artificial neural network is proposed for renewable generation forecasting to feed to the Q learning algorithm.