Abstract:
Power transformer is one of the more important equipment in a power system. It is very essential to keep the equipment in good health at all points of time. Dissolved Gas Analysis (DGA) is a tool used in monitoring power transformers. National and international standards provide guidelines for interpreting DGA data and discerning the nature of fault. In many cases, the suggested guidelines are not able to classify the faults precisely. Soft computing tools such as Support Vector Machine Method and Extreme Learning Machine methods seem to offer more exact classification and hence are recommended in such cases. This paper proposes to apply these relatively new methods in fault classification. In an earlier contribution, the Authors considered only one data base, reported by the IEC was considered. However, a comparison of the applicability of the computational techniques over several data bases was found necessary. Also, a need was found to include combined electrical and thermal faults as a parameter in the classification. This contribution expressly considers in this aspect in some detail. To demonstrate the application of these computational techniques, IEC TC10 database (DB1) and a local database (DB2) are considered. Fault classification based on gas concentration as the input, as also the enthalpy of the corresponding gases are compared. The fault classification based on enthalpy is found to identify the fault more precisely.
