Fault Identification with High SPV Penetrated Distribution Grid Using Modified Ceemdan Method

Abstract

Microgrids are not a prospective future solution but an existing paradigm, ensuring the energy sufficiency through decentralized and resilient power generating systems. Considering the techno-economical benefits, SPV is widely used in the present day as decentralised power sources. Along with the energy security, SPV contributes the system complexity in the active distribution system, which adds further complexities in the protection systems to identify the faults. To address this complex issue, this paper proposes a new technique for fault detection in a solar-integrated distribution grid system that combines Modified Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (MCEEMDAN) for feature extraction and sophisticated machine learning classifiers for classification. The Modified MCEEMDAN approach successfully decomposes fault signals into intrinsic mode functions (IMFs), allowing for the extraction of entropy-based features that represent the nonlinear properties of these signals. Several entropy measures, including Approximate Entropy and Sample Entropy, were investigated to improve fault classification performance. The findings show that the XGBoost classifier beat other models, obtaining an accuracy of 97.5% using approximation entropy features, demonstrating its ability to detect flaws. This study emphasizes the importance of feature selection in optimizing computing efficiency and model performance. The results provide important insights into the integration of signal processing methods and machine learning for improved problem identification in electrical systems, opening the way for future advances in the reliability and efficiency of power distribution networks. The entire analysis is done on a IEEE-33 bus distribution system modelled in Matlab Simulink.