Hybrid feature-based and optimized classification algorithm for arc fault detection in hybrid AC/DC microgrid

Abstract The increasing penetration of distributed renewable energy sources (DRES) in hybrid AC/DC microgrids (HAC/DC μ G) introduces significant safety concerns. Among these, arc faults (AFs) on the DC side pose a particular risk due to the absence of natural current zero-crossing, making detection and protection more challenging compared to AC systems. AFs are unavoidable during normal operating conditions due to dielectric breakdown. They can cause fire hazards, system malfunctions, and even fatalities. To address this issue, this study proposes an AF detection framework that captures AF V-I characteristics using both linear ( <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" overflow="scroll"> <mml:msub> <mml:mrow> <mml:mi class="MJX-tex-calligraphic" mathvariant="script">L</mml:mi> <mml:msub> <mml:mrow/> <mml:mrow> <mml:mi>A</mml:mi> </mml:mrow> </mml:msub> </mml:mrow> <mml:mrow> <mml:mi>F</mml:mi> </mml:mrow> </mml:msub> </mml:math> ) and non-linear ( <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" overflow="scroll"> <mml:msub> <mml:mrow> <mml:mi class="MJX-tex-calligraphic" mathvariant="script">N</mml:mi> <mml:msub> <mml:mrow/> <mml:mrow> <mml:mi>A</mml:mi> </mml:mrow> </mml:msub> </mml:mrow> <mml:mrow> <mml:mi>F</mml:mi> </mml:mrow> </mml:msub> </mml:math> ) AF modeling. A feature extraction approach is applied to arc voltage and current signals, where statistical time-domain features are combined with frequency-domain features obtained using the Fast Fourier Transform (FFT). These features are then used to train a Deep Fully Connected Neural Network (DFCNN) with three hidden layers, enhancing the detection capability. To further enhance detection accuracy and develop a highly precise AF detection model, the DFCNN is optimized using the Pelican Optimization Algorithm, which fine-tunes its parameters for superior performance. The mathematical models of <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" overflow="scroll"> <mml:msub> <mml:mrow> <mml:mi class="MJX-tex-calligraphic" mathvariant="script">L</mml:mi> <mml:msub> <mml:mrow/> <mml:mrow> <mml:mi>A</mml:mi> </mml:mrow> </mml:msub> </mml:mrow> <mml:mrow> <mml:mi>F</mml:mi> </mml:mrow> </mml:msub> </mml:math> and <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" overflow="scroll"> <mml:msub> <mml:mrow> <mml:mi class="MJX-tex-calligraphic" mathvariant="script">N</mml:mi> <mml:msub> <mml:mrow/> <mml:mrow> <mml:mi>A</mml:mi> </mml:mrow> </mml:msub> </mml:mrow> <mml:mrow> <mml:mi>F</mml:mi> </mml:mrow> </mml:msub> </mml:math> models are developed and implemented in MATLAB Simulink. Additionally, an AF generator setup is designed and operated on the Electron A0240 HAC/DC platform to emulate real-time AF tests. The findings indicate that the proposed approach improves classification performance compared to other classifiers. The framework integrates the modeling of both <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" overflow="scroll"> <mml:msub> <mml:mrow> <mml:mi class="MJX-tex-calligraphic" mathvariant="script">L</mml:mi> <mml:msub> <mml:mrow/> <mml:mrow> <mml:mi>A</mml:mi> </mml:mrow> </mml:msub> </mml:mrow> <mml:mrow> <mml:mi>F</mml:mi> </mml:mrow> </mml:msub> </mml:math> and <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" overflow="scroll"> <mml:msub> <mml:mrow> <mml:mi class="MJX-tex-calligraphic" mathvariant="script">N</mml:mi> <mml:msub> <mml:mrow/> <mml:mrow> <mml:mi>A</mml:mi> </mml:mrow> </mml:msub> </mml:mrow> <mml:mrow> <mml:mi>F</mml:mi> </mml:mrow> </mml:msub> </mml:math> models, time-frequency domain features, and optimized DFCNN classification to evaluate AF detection in HAC/DC μ G, achieving a detection accuracy of 98.97% with an average detection time of 45 ms.