关键词: 变压器, 油气识别, 故障诊断, 神经网络, 改进BP算法

Abstract:

Transformer faults can generally be divided to two types: the discharge fault and the thermal fault. Both of these two faults could cause specific effect on the transformer oil. The fault types can be determined by analyzing the gas content in the transformer oil in this paper. A new method is presented for transformer fault diagnosis, which includes an improved BP algorithm and information fusion. Five characteristic gases dissolved in the transformer oil are applied as the inputs of the neural network, and six states of the transformer as the outputs. With the momentum factor added, the learning rate coefficient of the neutral network is enhanced. The improved BP algorithm features the advantage of adaptive learning. The results of simulation indicate that the new method of transformer fault diagnosis is reliable and effective.

Key words: transformer;dissolved gas-in-oil analysis;fault diagnosis;neural network;improved BP algorithm