A Noise-Aware Framework for Reliable Detection of High-Impedance Faults in DC Microgrids

Abstract

This article proposes an adaptive variational mode decomposition (AVMD) framework for robust high-impedance fault (HiF) detection under non-Gaussian noise conditions. The method formulates parameter selection as a statistically guided optimization problem, where the decomposition penalty factor is optimized using reconstruction fidelity and higher-order statistical features. The proposed approach demonstrates stable reconstruction performance across varying noise conditions and repeated evaluations, with low variability in key metrics such as root mean square error (RMSE), signal-to-noise ratio (SNR), and kurtosis. Results demonstrate tightly bounded RMSE, SNR, and kurtosis across repeated trials, with a coefficient of variation below 0.5 %, confirming strong statistical robustness. Furthermore, an adaptive $\boldsymbol {\alpha }$ -reuse strategy is introduced to reduce computational overhead, achieving ~(30 %–40 %) latency. The proposed approach is validated under Gaussian, impulsive, and nonlinear noise scenarios at multiple SNR levels, and real-time feasibility is demonstrated through hardware-in-the-loop implementation on an embedded platform. Further, across all nine OPAL-RT hardware case studies, AVMD correctly discriminates HiF, low-impedance fault (LiF), and load-change (LC) events at latencies of 240–515 ms with optimal maximum-A posteriori (MAP)-selected $\boldsymbol {\alpha }$ values establishing a computationally efficient, noise-aware, and unified Bayesian framework for dependable HiF detection in embedded hardware and cloud supervisory systems.