Abstract

This paper presents a novel approach that integrates deep reinforcement learning (DRL) with the conventional virtual synchronous generator (VSG) to address dual objectives of microgrid (MG) control, frequency regulation and precise active power sharing. MGs typically consist of multiple Inverter-Based-Distributed-Generators (IBDGs) connected in parallel through different line impedances. The conventional active power loop (APL) of the VSG encounters significant steady-state frequency errors as load increases/decreases during islanded operation. To mitigate this issue, secondary-level controllers like proportional-integral (PI) control are added to the APL to regulate the frequency of IBDGs. However, PI control compromises power-sharing capabilities when the impedance values of connecting feeders for each IBDG are mismatched. To eliminate frequency errors and achieve accurate power sharing concurrently, this study adopts a DRL-based strategy. The agent collects state information from each IBDG in the microgrid as input and undergoes training using a reward function crafted to satisfy both objectives simultaneously. The performance of the trained agent is demonstrated in a two-inverter microgrid system designed in MATLAB/SIMULINK and is compared against traditional methods.

Department(s)

Electrical and Computer Engineering

Keywords and Phrases

Active Power Sharing; Deep Reinforcement Learning; Frequency Control; Inverter Based Distributed Generators; Microgrids; Twin Delayed DDPG; Virtual Synchronous Generator

Document Type

Article - Conference proceedings

Document Version

Citation

File Type

text

Language(s)

English

Rights

© 2024 Institute of Electrical and Electronics Engineers, All rights reserved.

Publication Date

01 Jan 2024

Download

Full Text Link

Share

 
COinS