Study on Identification of Partial Discharge Type of Oil-Paper Insulation Based on Deep Separable Convolutional Neural Network

The identification of partial discharge types is of great significance for the diagnosis and evaluation of transformer insulation. With the rapid advancement of deep learning in fields like machine vision and image processing, applying deep learning algorithms to partial discharge type recognition has become a new trend in power transformer fault diagnosis. Firstly, this article established a simulation test platform for partial discharge of oil paper insulation in the laboratory, conducted long-term partial discharge tests using the constant voltage method, and obtained ultrasonic signals under typical discharge defect models. Then a two-dimensional time-frequency spectrum was constructed, and a depthwise separable convolutional neural network model was used to classify the spectrum, thereby achieving identification of partial discharge types. The results indicate that there are differences in the time-frequency spectra of ultrasound signals among different defect models; The partial discharge identification model used in this article can ensure high accuracy while effectively reducing the size of the model and improving training efficiency; Compared with traditional machine learning models, the model used in this paper has a higher recognition accuracy, reflecting the superiority of convolutional neural networks in image recognition.