Reinforcement Learning for Energy Sharing in a Smart Microgrid under Partial Information

Electric energy has always been an important staple for human life and evolution during the last decades. During the last century we have come to conclude that fossil combustibles are not infinitely available and as electric energy demand is getting higher each yearquestions arise regarding energy saving, efficiency and environment pollution. Renewable energy sources have been a great achievement in the development of better alternatives. As alternatives became available, the main focus of recent studies has moved towards the management of electric energy for a more efficient and rational use of the resources. The posed challenges are the increasing global electricity demand, high reliability requirements and environmental protection, to name a few. The evolution of the power system has led to smaller and more distributed generations and the concepts of Smart Grid and Microgrid have emerged alongside Government Policies trying to make the transition easier. With this work we try to help towards the transition into this newer spectrum, while addressing social concerns and user privacy, by considering a scenario where neighboring users share a Storage System to improve the community welfare while limiting information sharing with the grid. The goal is to provide the community with fewer costs while maintaining user comfort. The novel scenario considers a decentralized control with each user taking decisions given some general grid’s information, but not sharing its own information with any other entity. We apply Reinforcement Learning to optimize the system and compare the results with Centralized control.