چكيده به لاتين
The rapid growth of inverter-based distributed generation (IBDGs) in microgrids has introduced significant challenges in network protection, particularly for fault detection and isolation. The dynamic behavior and limited fault current contribution of IBDGs render conventional protection schemes ineffective, compromising system reliability and leading to economic and social impacts. This work proposes an advanced protection system using an optimized Random Forest model to achieve high-speed, high-accuracy fault detection, classification, and localization, even for high-impedance faults. The methodology extracts key features (such as RMS, mean, maximum, and standard deviation) from three-phase voltage and current signals sampled during the first post-fault cycle. Feature selection identifies the 20 most significant features, enabling the model to detect faults from normal events such as load switching, capacitor switching, and n-1 contingencies, classify fault types (e.g., single-phase, phase-to-phase-to-ground), identify faulty phases, locate fault positions, and estimate fault resistance. The system also distinguishes internal faults from external faults, preventing maloperation of the protection system and ensuring accurate fault detection without unnecessary tripping. Simulations were conducted on the IEC microgrid test system using MATLAB, and the validation and machine learning model training were implemented in Python. The results demonstrate an average response time of 1.89 microseconds. The proposed model achieves perfect accuracy (1.00) for fault detection, faulty zone identification, and fault location determination. Fault type classification achieves 99.81% accuracy, while faulty phase identification attains 98.83% accuracy. Fault resistance estimation achieves 83.36% accuracy with the Random Forest model and improves to 95.07% using an ensemble model. These results highlight the system’s robustness, precision, and suitability for real-time microgrid applications.
