Key Gases in Transformer Oil – An Analysis using Self Organizing Map (SOM) Neural Networks

Power transformer (PT) is a typical resource to the power system network for the transmission and distribution of power. A typical PT operates all the days in a year at full-load capacity and the prime equipment to be affected by natural and abnormal conditions, hence susceptible to failures. The key gases (KG) evolved due to incipient faults (IF) in the operation of PT are named as $\mathrm{H}_{2},\mathrm{CH}_{4},\mathrm{C}_{2}\mathrm{H}_{6},\mathrm{C}_{2}\mathrm{H}_{4}$, and C
2
H
2
. These gases vary in composition according to the type of IF hence a menace to PT. The composition of KG in PT has an organized structure which can be found by clustering the KG data. The clustering in this investigation is achieved by an artificial neural network feature known as self-organizing map (SOM), an unsupervised machine learning application. The composition of the KG in PT has a 200 data which is used to generate a two-dimensional map, SOM using the KG as features called as nodes. SOM input planes show H
2
has higher weights in initial training followed by $\mathrm{C}_{2}\mathrm{H}_{2},\mathrm{CH}_{4},\mathrm{C}_{2}\mathrm{H}_{6},\mathrm{C}_{2}\mathrm{H}_{4}$. The SOM nodes were correlated with weights except few initial classes with distant nodes. Overall, the investigation supports in getting a clear picture of connections between the KG and thereby prioritize the pre-warning system to overcome failure of PT.