A Novel Approach for Diagnosing Transformer Internal Defects and Inrush Current Based on 1DCNN and LSTM Deep Learning

Abstract

In power systems, power transformer (Pt) protection plays a vital role in ensuring that customers have a reliable power supply. Correctly recognizing inrush currents from internal defects and preventing differential relay malfunctions are two of biggest challenges facing the differential protection of power transformers. Although previous approaches suggested to overcome these issues have promising outcomes, increasing the accuracy and reducing the execution time and complexity of transformer differential relays are still interesting topics for researchers. Accordingly, a new fault diagnostic method based on wavelet transform (WT) and deep learning is introduced in paper. In the proposed approach, Discrete WT is used to extract the features of differential currents, and combined one-dimensional convolutional neural networks and long short-term memory (1DCNN-LSTM)) is applied for classify internal faults from other abnormal events. High accuracy, no need for any thresholds or transformer parameters and fast fault detection are the main advantages of the proposed approach. The simulation results for a 132/11 kV, 63 MVA power transformer approved the proposed method for its ability to accurately differentiate between inrush currents and internal defects after 5 ms, as well as its accuracy for abnormal event classification of about 99.4%.