Identification and classification of cross-country faults in transformers using K-NN and tree-based classifiers

Abstract

Today, despite the advances made, differential relays still fail to identify cross-country faults and some events. Therefore, in this paper, signal processing techniques and artificial intelligence were used to identify and classify cross-country faults with other events in transformers simultaneously. For this purpose, first, mechanical defects of transformer winding were modeled on a laboratory sample. Then, types of cross-country faults, internal and external electrical faults, and transformers҆ inrush currents were simulated using EMTP software to extract differential current signals. The sampled signals were transferred to MATLAB software. Then, some salient features were extracted from the obtained signals by Time-Time transform. Finally, these features were used to train Random Tree, Random Forest, and K-Nearest Neighbors (K-NN) classifiers in order to identify and classify cross-country faults from other events in the Rapid-Miner software. The results showed that the intelligent combined method can identify and classify cross-country faults from other disturbances in transformers with an appropriate degree of accuracy; which is the main innovation of this research compared to other studies that have been limited to identifying and classifying only some of the mentioned faults.