Enhancing Neural Networks-Based Classification of Incipient Faults in Power Transformers via Preprocessing

The power transformer is one of the most important equipment in an electric power system. If this equipment is out of order for some reason, the damage for both society and electric utilities are very significant. In this work, we present a comparative study of the application of Linear Networks, Multi-Layer Perceptrons - with three and four layers - and Radial Basis Functions Networks in the classification of incipient faults via Dissolved Gas Analysis (DGA) in power transformers. Besides, preprocessing techniques for databases have been discussed as well. The proposed procedures have been applied to real databases derived from chromatographic tests of power transformers. The results obtained by all techniques are compared and fully described.