Adaptive fault diagnosis in renewable integrated microgrids using hybrid machine learning approach

Abstract

Reliable fault detection, classification, and localization are crucial for maintaining the reliability and self-healing capabilities of smart grids, particularly in microgrids integrated with renewable energy sources. This paper presents an adaptive fault diagnosis framework using K-Nearest Neighbors (KNN) and Artificial Neural Networks (ANN), that models a practical microgrid architecture incorporating conventional and renewable energy sources, dynamic loads, and distributed transmission lines. The framework evaluates the performance of K-Nearest Neighbors (KNN) and Artificial Neural Networks (ANN) for fault detection, classification, and localization directly from the raw voltage and current measurements in per unit, thereby avoiding image-based transformations or complex feature extraction. By leveraging per-unit normalization of measurements, the proposed scheme achieves compatibility across diverse microgrid configurations. This standardized approach enhances scalability and flexibility, ensuring reliable performance under varying operating conditions and facilitating practical implementation in real-world systems. Simulation results demonstrate that the hybrid KNN–ANN framework achieves 99.9% accuracy in fault detection and classification, and 98.45% accuracy in fault localization demonstrating strong reliability for intelligent protection applications. To validate robustness and scalability, the system is modelled in per-unit representation and tested on two additional microgrid configurations. The findings confirm the adaptability and effectiveness of the proposed methodology, highlighting its potential as a reliable solution for intelligent protection in modern smart grid infrastructures.