Power transformers are one of the most important and costly equipment of electrical networks. Possible malfunctions that may occur in a power transformer may cause power outages as well as large energy losses. Therefore, it is important to detect the fault in transformers in advance. One of the commonly used methods for fault diagnosis in transformers is to make analyses based on gas concentrations that occur at the time of failure. This method is called dissolved gas analysis (DGA) which is based on measuring gas formation in the transformer insulating fluid during or before the fault. In this study, gas data obtained from DGA was used as the inputs of the chosen machine learning algorithms, and their diagnostic performances were measured. First, the International Electrotechnical Commission Technical Committee (IEC TC)-10 data set which is very popular in the literature was used, and then the real data set obtained from the Turkish Electricity System was applied. Since the data set consists of different sizes, it greatly affects the performance of classification algorithms. Different data preprocessing methods were applied to increase the performance of the algorithms, and how they affect the performance of the algorithms was examined.
Cite this article as: M. Demirci, M. Saka, H. Gözde and M. C. Taplamacıoğlu, “Fault analysis with data pre-processing methods in power transformers,” Turk J Electr Power Energy Syst., 2023; 3(3), 133-139.