Partial discharge pattern recognition method based on Transfer Learning and DenseNet model

With the development of sensing technology, a large number of partial discharge (PD) time domain data are generated in the field of gas-insulated integrated electrical appliances (GIS). Traditional pattern recognition methods cannot identify the defect types of such data. At the same time, deep learning methods generally face the problem of small samples. In order to solve the above problems, this paper proposes a PD pattern recognition method based on transfer learning and DenseNet model. Firstly, the finite-difference-time-domain(FDTD) method is used to simulate PD in GIS, and the time domain waveform image dataset of four PD defects was established. Secondly, Image processing techniques such as image enhancement, normalization and image compression are used to process the time-domain waveform spectrum. Finally, Using CNN network and transfer learning, the recognition accuracy of the model is improved to 96.38%, and the convergence is obviously improved. The identification performance of different CNN network structures is studied, and the results show that the accuracy of DenseNet model is significantly higher than other structures. This study can be used to diagnose PD defects in GIS equipment,