Hybrid Optimal Adaptive Maximum Entropy Reinforcement Learning for Balanced Islanding Detection in Distributed Generation

ABSTRACT Grid‐connected solar systems confront the persistent challenge of islanding, a phenomenon wherein the system persists in generating power even after disconnection from the main grid. Detecting islanding poses multifaceted challenges, encompassing model complexity, non‐detection zones, power quality ramifications, nuisance tripping occurrences, and synchronization and timing issues. In response, this research proposes an innovative solution that integrates passive and active techniques with a Reinforcement Learning algorithm. The passive component harnesses DWT for signal processing, while the active component utilizes a pioneer evolutionary algorithm (PEA) for adaptive optimization. This fusion results in the development of the Hybrid Optimal Adaptive Maximum Entropy Reinforcement Learning (HOAMERL) algorithm, engineered to enhance islanding detection efficiency and accuracy. To evaluate the method, it is simulated in IEEE‐14 bus environment. Additionally, the power system incorporated with a low‐pass filter to optimize nuisance tripping and a Static VAR Compensator (SVC) to improve power quality. The research findings unveil compelling outcomes, with a tripping time of 45 ms, a detection time of 1.29 ms, a notable reduction in total harmonic distortion (THD) by 0.90%, and an impressive detection accuracy of 99.92%. These findings highlight the HOAMERL algorithm's remarkable capability to address the intricate challenges of islanding detection, signifying its potential to significantly elevate the reliability and performance of grid‐connected solar systems in real‐world applications.