Intelligent fault detection and classification in DC microgrids using MEMD and Deep Neural Decision Forests

Fault detection and classification in the Distribution Generation (DG) integrated DC microgrid demands a very robust and efficient approach to handle many inherent issues and challenges related to zero crossing, low inertia factor, high variation in short-circuit current and different modes of operation. This paper presents a hybrid novel approach comprised of Multivariate Empirical Mode Decomposition (MEMD) for feature extraction from the non-stationary power signals and Deep Neural Decision Forests (DNDF) for classification for the correct decision. The reason for exploiting MEMD for fault detection and classification is that it enables the simultaneous decomposition of multiple related faulted power signals, ensuring mode alignment, preserving joint information, and facilitating direct comparison of their shared non-stationary and non-linear characteristics. On the other hand, the DNDF approach is exploited for microgrid fault classification because they combine deep learning’s powerful feature extraction with decision forests’ robust, interpretable, and noise-tolerant classification. The proposed approach is tested through extensive simulation under various normal and faulted operating conditions, even under varying fault resistances and operational uncertainties. MEMD decomposes multivariate microgrid signals, enabling the extraction of meaningful temporal features. Later, these features are used as input to a DNDF to categorize fault types. Extensive simulations on test DC microgrid configurations demonstrate that the integrated MEMD and Deep Learning (DL) strategy provides superior Accuracy, robustness, and resilience in comparison to conventional and other prominent techniques.