Decentralized energy-efficient microgrid control Using Graph neural networks and LSTM-based Event-Triggered control

Abstract

As microgrid systems become more complex and interconnected, traditional control strategies face significant challenges in terms of scalability, efficiency, and responsiveness. Existing models, often relying on time-triggered approaches, result in excessive communication, energy waste, and slower system responses. The main purpose of this work is to formulate a decentralized control architecture that communicates better, regulates voltage and frequency, and stabilizes the microgrids. To address these limitations, this research introduces an innovative decentralized control framework that combines Graph Neural Networks (GNNs) and Long Short-Term Memory (LSTM) networks, integrated with Event-Triggered Control to optimize microgrid operations. This methodology applies GNNs to capture the spatial dependencies among microgrid components like generators, storage, and loads. Meanwhile, the LSTMs identify the temporal dynamics associated with variations in load and generation. System control actions are then triggered only when necessary, hence reducing communication overhead considerably. The results demonstrates 55 % less communication load was reported, voltage regulation accuracy increased by 45 %, and other efficiency measures for frequency regulation improved by 35 %. Along with these, other performance metrics indicate a 30 % improvement of the Voltage Stability Index (VSI) going from 0.47 to 0.33 and lowering the Frequency Regulation Error (FRE) by 20 % from 4.5 % to 3.6 %. All of which consolidated the evidence of the efficiency of the approach suggested to control microgrid operations in a real-time adaptive energy-efficient manner. These findings highlight the powerful combination of GNNs and LSTMs for achieving adaptive, energy-efficient, and real-time control in decentralized microgrid systems.