Enhancing DC microgrid performance through machine learning‐optimized droop control

Younes Saeidinia,Mohammadreza Arabshahi,Mohammad Aminirad,Miadreza Shafie‐khah

DOI: https://doi.org/10.1049/gtd2.13169

2024-04-27

IET Generation Transmission & Distribution

Abstract:Proposing a novel machine learning‐based optimized droop method for achieving simultaneous reduction in production cost (PC) and power line losses (PLL). Combining droop control techniques with machine learning methodologies, employing predictive models to estimate PC and PLL for various weighting coefficient (WC) combinations. Ensuring objectivity in WC assignment through the proposed systematic approach, eliminating subjectivity and arbitrariness in selection. A machine learning‐based optimized droop method is suggested here to simultaneously reduce the production cost (PC) and power line losses (PLL) for a class of direct current (DC) microgrids (MGs). Traditionally, a communication‐less technique known as the hybrid droop method has been employed to decrease PC and PLL in DC MGs. However, achieving the desired reduction in either PC or PLL requires arbitrary adjustments of weighting coefficients for each distributed generator in the conventional hybrid droop method. To address this challenge, this paper introduces a systematic approach that capitalizes on the benefits of artificial intelligence to accurately predict both the PC and PLL in a DC MG. Furthermore, an optimization technique relying on the gradient descendent method is employed to independently optimize both PC and PLL for each scenario. The effectiveness of the proposed method is confirmed through a comparative study with classical and hybrid droop coordination schemes under various scenarios such as rapid load changes.

engineering, electrical & electronic