Intelligent fault types diagnostic system for dissolved gas analysis of oil-immersed power transformer

Transformer failures are often due to aging, deterioration or damage of the internal insulation materials. Combustible gases are also generated when the insulation materials are subjected to thermal or electrical stress. This study proposes a fault diagnosis system, which combines a multinomial logistic regression model and back-propagation neural networks, to determine the type of fault of a power transformer by analyzing the dissolved gases in the transformer. The compositions and amounts of the dissolved gases that are crucial or relevant to specific types of faults are selected by the multinomial logistic regression model as the inputs to the neural network to train the diagnosis system, so the diagnosis system can learn to diagnose the type of faults. The test results show that the recognition rate of the proposed intelligent fault type diagnosis system is about 10- 30% higher than those of a single-neural or multi-neural networks recognition system which does not incorporate the multinomial logistic regression model.