Integration of Accelerated Deep Neural Network Into Power Transformer Differential Protection

Differential protection scheme is the main protection scheme of power transformers, which still holds the risk of sending false trips subject to inrush currents. This article aims to develop a differential protection scheme to discriminate power transformer magnetizing current from internal faults to decrease the risk of false trips. In this article, an accelerated convolutional neural network (CNN) based approach is designed for the discrimination between internal faults and inrush current. The main competitive advantage of the proposed algorithm is its capability in fusing the feature extraction and fault detection blocks into a single deep neural network block by enabling the network to discover important features automatically. The result of this point is that the algorithm is more efficient in terms of speed, hardware usage, and accuracy. The proposed method is applied to a simulated 230-kV network and an experimental prototype. Different cases with various external factors are simulated to calculate reliability indexes. The comparison between the accelerated CNN, conventional CNN, and nine widely used methods demonstrates the faster and more reliable performance of the proposed algorithm.