Improvement of power transformer fault diagnosis by using sequential Kalman filter sensor fusion

Power transformers are one of the most important and costly equipment for the reliability and continuity of electrical power systems. For this reason, continuous monitoring of power transformers during normal operating conditions of the power grid and early fault diagnosis from existing parameters before a fault occurs is an important task. One of the most common analysis is the Dissolved Gas Analysis (DGA) method. In the DGA analysis, the concentrations of gases formed in the transformer insulating fluid are measured, classified and used to predict failures. It is observed that the classification and diagnosis are performed by classical and artificial intelligence-based methods in the relevant literature and applications. In this study, gas data classified by machine learning method is combined with sensor fusion methods to increase the diagnosis accuracy. It has been determined that the Sequential Kalman filter, which is first used differently from the literature, increases the estimation accuracy over 90% according to the results obtained by the Majority Voting and Dempster Shafer Evidence Theory fusion methods and the other results with IEC-TC-10 dataset mentioned in the literature.