Image-Based Partial Discharge Identification in High Voltage Cables Using Hybrid Deep Network
Obaid Aldosari, Mohammed A. Aldowsari, Salem Batiyah, N. Kanagaraj
Abstract
Deep learning and digital image technologies have combined to create a potentially effective tool for identifying partial discharge (PD) patterns precisely. However, it is necessary to investigate which algorithm guarantees the best performance. The more common tools are restricted by a lack of training data and an advanced model in itself. Therefore, the main goal of this paper is to develop an efficient hybrid network comprising two deep networks, long short-term memory (LSTM), and convolutional neural network (CNN), for identifying the form of PD. A total of <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$8186\times 25$ </tex-math></inline-formula> (non-PD <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$\times $ </tex-math></inline-formula> PD) images were applied to assess the proposed methods. The size of the PD type was increased to 3675 images using data augmentation techniques. The results indicated that the integration of CNN and LSTM networks can provide a more robust implementation for PD detection. The integrated CNN-LSTM deep network based on data augmentation outperformed features derived from a single deep network. The recall, F-measure, and classification precision have 99.9% as a validation accuracy with a 99.8% intersection over union and a loss of 0.004.