A fuzzy information optimization processing technique for monitoring the transformer in neural-network on-line

In the electricity utilities around the world, a large number of power transformers are operating beyond their design life. The reliability and quality of power transformers is vital to system operation. In order to determine the condition of the insulation in transformers, many methods are developed. The most interesting methods for identifying fault conditions of the insulation for oil-filled transformer are dissolved gas analysis (DGA), acoustic analysis for the partial discharge (PD), liquid chromatography, and transfer function techniques. But people only apply single of them to monitoring transformer, and fail to combine the full of information from different methods. This paper establishes a theory prototype of neural network fuzzy information optimization processing technique. The learning rule and some key properties for the neural network are analyzed. It gives a fuzzy neural network learning rule. The paper integrates different diagnostic methods and information likes DGA, gas rate, acoustic analysis for the PD, temperatures in transformer, electric current, etc. that the data is not very clearly separable. There are different weights associated with each diagnosis decision with DGA, PD, and other technique. And a non-linear penalty function is used.