Key points are not available for this paper at this time.
Partial discharge (PD) represents a substantial hazard to the integrity and safety of high-voltage cable systems, making the detection of such phenomena both a critical and challenging scientific endeavor. Leveraging recent breakthroughs in deep learning, this study pioneers a Time Frequency Multi-View Time-Series (TFMT) model embedded within a deep learning framework to adeptly pinpoint partial discharge events in high-voltage cables. The TFMT model begins by analyzing pulse signals to extract salient features across time and frequency spectra. Following feature extraction, the model employs kernel density estimation techniques to convert pulse signal data into informative two-dimensional graphical representations. Building upon these insights, we developed a novel network architecture that synergistically fuses Convolutional Neural Networks (CNNs) with Recurrent Neural Networks (RNNs), harnessing the strengths of both to enhance diagnostic accuracy. This sophisticated integration facilitates a robust, dynamic analysis of PD activities, capturing both spatial and temporal data dependencies with finesse. The results indicate that the proposed model achieves a high accuracy of 99.49% in partial discharge recognition. The TFMT model not only delineates a new horizon for PD identification but also sets a precedent for the future of advanced diagnostic systems in the electrical power grid.
Li et al. (Mon,) studied this question.