Performance of Mathematical Indices in Transformer Condition Monitoring Using k-NN Based Frequency Response Analysis

Despite the development of the use of frequency response analysis (FRA) in condition monitoring of power transformers, how to interpret the results of FRA measurements has not yet been standardized. Therefore, proposing new methods to interpret the results of FRA measurements in research works is followed with great interest by researchers. This paper proposes a k-nearest neighbor (k-NN) based method for condition monitoring of transformers using the results of FRA measurements. First, the necessary measurements are performed on healthy and faulty transformers (under different fault conditions) and the required database is created. Then, by extracting the peak (resonance) and trough (anti-resonance) points of the measured transfer functions from the transformer, several mathematical features for training and validation of k-NN are extracted. Finally, by applying the data obtained from actual transformers, the performance of k-NN in different states is evaluated and compared. The results show that the proposed method is able to determine the condition of the transformer (whether it is healthy or defective) with very good accuracy and if it is defective, identify the type of defect. In addition, in order to prove the ability of k-NN, a comparison is made with the results of the artificial neural network (ANN).