Probabilistic Neural Network Based Incipient Fault Identification Using DGA Dataset

AbstractTransformers are expensive and important equipment in the whole power system. Fault in transformers not only affect electric utilities but also customers by interrupting the power supply. The maintenance and servicing of transformers is usually time consuming and costly. If we can somehow predict the incipient faults, its maintenance and servicing can be planned before faults actually happens. Dissolve gas analysis is an important method which can detect the incipient fault conditions. In earlier times faults were monitored by conventional methods, which were time taking and sometimes even need transformer to be out of service, which was a big drawback. Artificial Intelligence is a technique that can help in detecting the incipient fault conditions in lesser time with high accuracy as compared to the conventional techniques available. Key Gas, Dual Triangle, Rogers Ratio, IEC Ratio and, Doernenburg Method, percentage of gases like %C2H2, %C2H4, %CH4, and combination of gases are applied as an input to ANN models (PNN and MLP). MLP and PNN techniques are trained and compared to get the input parameter that gives most accurate fault predictions