Abstract

Power transformers are major elements of the electric power transmission and distribution infrastructure. Transformer failure has severe economical impacts from the utility industry and customers. This paper presents analysis, design, development and experimental results of a robust failure diagnostic technique. Hopfield neural networks are used to identify variations in physical parameters of the system in a systematic way, and adapt the transformer model based on the state of the system. In addition, the Hopfield network is used to design an observer which provides accurate estimates of the internal states of the transformer that can not be accessed or measured during operation. The estimated physical parameters are then passed to a neural network to be classified into regions representing various parameter patterns. Analytical and experimental results of this adaptive observer for power transformer diagnostics are presented.

This content is only available via PDF.
You do not currently have access to this content.