Power Transformer Discharge Fault Identification Based on Deep Learning

Abstract

Power transformer is one of the most important power transmission and transformation equipment in the power system. Transformer fault diagnosis helps to ensure the safety and stability of system operation. The hidden dangers of transformers due to insulation in power systems are usually manifested in the form of partial discharges. In order to detect the deterioration of the insulation in time and avoid safety accidents, the fault identification of power transformers is very important. Aiming at the problem that signal samples are difficult to obtain, a lightweight deep learning model driven by the phase resolved partial discharge (PRPD) and the phase resolved pulse sequence (PRPS) is proposed to complete fault type identification. Finally, the experimental results show that compared with single-modal data, multi-modal data driven based model can better exert the generalization ability of deep learning. In addition, the deep learning model is robust to the adjustment of the number of convolution kernels, which helps to adjust the model scale according to actual needs.