A Fast and Robust Fault Detection and Classification Scheme for AC Microgrids Using Variational Mode Decomposition and Decision Tree

Abstract

The growing integration of inverter-based renewable energy resources into AC microgrids has introduced several challenges for conventional protection systems. These challenges primarily arise from the reduced fault current contribution of power electronic converters, the presence of bidirectional power flow, continuously changing operating conditions, and uncertainties associated with communication networks. This article presents a fast, robust, and interpretable fault detection and classification framework for AC microgrids based on variational mode decomposition (VMD) and a decision tree (DT) classifier. Positive-sequence differential current magnitude and phase-angle signals measured by phasor measurement units at both ends of distribution lines are processed using VMD to extract noise-resilient intrinsic mode functions (IMFs). A kurtosis-guided adaptive IMF selection strategy is employed to identify the dominant transient mode within the 100–500 Hz frequency band. Subsequently, a spectral-energy-based fault detection index (FDI) is formulated to enable reliable detection of both low- and high-resistance faults under inverter-limited fault current conditions. Subsequently, multiple statistical and transient features extracted from the selected IMF are then utilized by an interpretable DT classifier for fault-type identification. The proposed framework is extensively validated under varying fault resistance, fault location, distributed generation penetration, topology variation, operating modes, communication delays, synchronization mismatch, and noisy measurement conditions. Simulation results demonstrate fault detection times below 8 ms and classification accuracy exceeding 99.4% while maintaining reliable performance under practical communication constraints. The practical feasibility of the proposed method is further validated through OPAL-RT real-time simulation platform.