Research on flexible antenna and distributed deep learning pattern recognition for partial discharge monitoring of transformer

Abstract Power transformer is an important part of the power system, and continuous monitoring of partial discharges can provide a more reasonable program for fault diagnosis and operational maintenance of the transformer. However, the rigid partial discharge UHF antenna can not be installed in a conformal fit with the monitored equipment, and the partial discharge UHF signal attenuation is serious, resulting in low detection energy efficiency and gain performance can not meet the demand. The centralized deep learning local discharge pattern recognition method has low training efficiency, and distributed deep learning can improve the training efficiency, but the heterogeneous data from multiple sources will reduce the model accuracy. Due to this, this paper designs a UHF flexible composite helical antenna with miniaturization, wide bandwidth, high gain and high bending deformation stability, and investigates a federated learning pattern recognition method based on residual contraction network, which substantially improves the training efficiency while ensuring the accuracy.