Intelligent decision support for diagnosis of incipient transformer faults using self-organizing polynomial networks

To serve as an intelligent decision support for the transformer fault diagnosis, a new self-organizing polynomial networks (SOPNs) modeling technique is proposed and implemented in this paper. The technique heuristically formulates the modeling problem into a hierarchical architecture with several layers of functional nodes of simple low-order polynomials. The networks handle the numerical, complicated, and uncertain relationships of dissolved gas contents of the transformers to fault conditions. Verification of the proposed approach has been accomplished through a number of experiments using practical numerical diagnostic records of the transformers of Taiwan power (Taipower) systems. In comparison to the results obtained from the conventional dissolved gas analysis (DGA) and the artificial neural networks (ANNs) classification methods, the proposed method has been shown to possess far superior performances both in developing the diagnosis system and in identifying the practical transformer fault cases.