RMP neural network based dissolved gas analyzer for fault diagnostic of oil-filled electrical equipment

Dissolved gas analysis (DGA) method is broadly used to diagnose incipient faults in oil-filled electrical equipment in service. This paper presents a reduced multivariate polynomial (RMP)-based neural network (NN) for the interpretation of DGA. RMP NN can be used as a pattern classifier and its parameters can be determined easily. Six inputs to the RMP NN with three-layer structures are made up of five gases. The effect of the order of RMP NN on diagnosis accuracy is analyzed in this study. The fault cases published have been used as training and testing patterns. The test results show that RMP NN has good diagnosis accuracy.