Survey on AI and Machine Learning Techniques for Microgrid Energy Management Systems

In the era of an energy revolution,grid decentralization has emerged as a viable solution to meet the increasing global energy demand by incorporating renewables at the distributed level.Microgrids are considered a driving component for accelerating grid decentralization.To optimally utilize the available resources and address potential challenges,there is a need to have an intelligent and reliable energy management system(EMS)for the microgrid.The artificial intelligence field has the potential to address the problems in EMS and can provide resilient,efficient,reliable,and scalable solutions.This paper presents an overview of existing conventional and AI-based techniques for energy management systems in microgrids.We analyze EMS methods for centralized,decentralized,and distributed microgrids separately.Then,we summarize machine learning techniques such as ANNs,federated learning,LSTMs,RNNs,and reinforcement learning for EMS objectives such as economic dispatch,optimal power flow,and scheduling.With the incorporation of AI,microgrids can achieve greater performance efficiency and more reliability for managing a large number of energy resources.However,challenges such as data privacy,security,scalability,explainability,etc.,need to be addressed.To conclude,the authors state the possible future research directions to explore AI-based EMS's potential in real-world applications.

Collaborative Distributed AC Optimal Power Flow: A Dual Decomposition Based Algorithm

We propose a dual decomposition based algorithm that solves the AC optimal power flow(ACOPF) problem in the radial distribution systems and microgrids in a collaborative and distributed manner. The proposed algorithm adopts the second-order cone program(SOCP) relaxed branch flow ACOPF model. In the proposed algorithm, bus-level agents collaboratively solve the global ACOPF problem by iteratively sharing partial variables with its 1-hop neighbors as well as carrying out local scalar computations that are derived using augmented Lagrangian and primal-dual subgradient methods. We also propose two distributed computing platforms, i. e., high-performance computing(HPC) based platform and hardware-in-theloop(HIL) testbed, to validate and evaluate the proposed algorithm. The computation and communication performances of the proposed algorithm are quantified and analyzed on typical IEEE test systems. Experimental results indicate that the proposed algorithm can be executed on a fully distributed computing structure and yields accurate ACOPF solution. Besides, the proposed algorithm has a low communication overhead.