Condition monitoring of power transformers with neural networks

In this paper, a neural network technique is applied to a system named NEUCOMS that is designed for the condition monitoring of power transformers. Through employing paired neural networks, NEUCOMS has the ability of analyzing data of electrical inspections as well as data of the chromatogram of oil-dissolved gases. It utilizes redundant input attributes to speed the training and reduce the size of the neural networks. Moreover, it exploits fuzzy techniques to preprocess the input data so that features with small values will not be blocked off by features with big values. Experiments show that NEUCOMS works well in real-world situations.