A Novel Neural Network-Based Droop Control Strategy for Single-Phase Power Converters

Managing parallel−connected single−phase distributed generators in low−voltage microgrids is challenging due to the volatility of renewable energy sources and fluctuating load demands. Traditional droop control struggles to maintain precise power sharing under dynamic conditions and varying line impedances, leading to inefficiency. This paper presents a novel adaptive droop control strategy integrating artificial neural networks and particle swarm optimization to enhance microgrid performance. Unlike prior methods that optimize artificial neural network parameters, the proposed approach uses particle swarm optimization offline to generate optimal dq−axis voltage references that compensate for line effects and load variations. These serve as training data for the artificial neural network, which adjusts voltage in real time based on line impedance and load variations without online optimization. This decoupling ensures computational efficiency and responsiveness, maintaining voltage and frequency stability during rapid load changes. Addressing dynamic load fluctuations and line impedance mismatches without inter−generator communication enhances reliability and reduces complexity. Simulations demonstrate that the proposed strategy maintains stability, achieves accurate power sharing with errors below 0.5%, and reduces total harmonic distortion, outperforming conventional droop control methods. These findings advance adaptive control in microgrids, supporting seamless renewable energy integration and enhancing the reliability and stability of distributed generation systems.