Designing a composite deep learning based differential protection scheme of power transformers

Abstract

This paper proposes a novel differential protection scheme based on deep neural networks (DNN). The goal is to propose a fast, reliable, and independent protection scheme in distinguishing inrush current from internal fault in power transformers, as the most challenging issue in power transformers protection. Shallow-based techniques require spectral analysis and handcraft feature extraction to be proper methods in this major. However, they require a significant computational cost. In order to address this issue, in this paper, a novel DNN-based approach is proposed based on combining convolutional neural network (CNN) and light-gated recurrent unit (LGRU), namely CLGNN. The results show a more accurate and more reliable performance than three different shallow and three state-of-the-art DNN based techniques. Adaptability and robustness of the proposed scheme are evaluated considering CT saturation, superconducting fault current limiter (SFCL), and series compensation impacts. The obtained results prove the effectiveness and validity of the proposed DNN-based protection scheme in this paper.