A Review on a Data-Driven Microgrid Management System Integrating an Active Distribution Network: Challenges, Issues, and New Trends

Abstract

The advent of renewable energy sources (RESs) in the power industry has revolutionized the management of these systems due to the necessity of controlling their stochastic nature. Deploying RESs in the microgrid (MG) as a subset of the utility grid is a beneficial way to achieve their countless merits in addition to controlling their random nature. Since a MG contains elements with different characteristics, its management requires multiple applications, such as demand response (DR), outage management, energy management, etc. The MG management can be optimized using machine learning (ML) techniques applied to the applications. This objective first calls for the microgrid management system (MGMS)’s required application recognition and then the optimization of interactions among the applications. Hence, this paper highlights significant research on applying ML techniques in the MGMS according to optimization function requirements. The relevant studies have been classified based on their objectives, methods, and implementation tools to find the best optimization and accurate methodologies. We mainly focus on the deep reinforcement learning (DRL) methods of ML since they satisfy the high-dimensional characteristics of MGs. Therefore, we investigated challenges and new trends in the utilization of DRL in a MGMS, especially as part of the active power distribution network (ADN).

Keywords

• active power distribution network
• energy management system
• microgrid management system
• machine learning
• deep reinforcement learning
• sparse reward