The Application Research Of Fault Diagnosis Technique Of Operating Transformers Based On Artificial Neutral Network

In order to monitor operating status of power transformer with new method, the neural network (NN) approach was introduced. Research and experiment were carried out in two ways: one is to utilize BP NN for dissolved gas analysis (DGA) based on transformer fault diagnosis, and the other is to apply ART NN & BP NN to recognize partial discharge sources.In the first application, teaching set of NN was set up with chromatography test records from many substations. The trained NN can process data composed of 7 kinds of gas contents (H2, CO, etc.) and determine the fault type in transformer subsequently. This new approach was compared with traditional IEC three-ratio method, and the results showed that it was superior.In the second application, much experimental work was done. Five electrode models were presented to simulate PD phenomena. Statistical operators were introduced to extract features from PD signals and they can well describe the properties of phase-position distributions. The original learning algorithm of ART2 NN is so complicated that a simplified ART (SART) NN was presented to handle statistical operators data that were used as input vector of NN. Satisfactory results showed that NN can recognize different PD patterns successfully.