CONSTRUCT THE DIAGNOSIS SYSTEM POWER TRANSFORMER LATENT FAULTS BASED ON ARTIFICIAL NEURAL NETWORK AND DISSOLVED GAS IN OIL ANALYSIS METHOD

Abstract

In this paper, the application of neural networks is proposed to construct the potential fault system of the power transformer. The neural network inputs are the ratio of the gas components generated during the transformer’s work. Moreover, the output is the conclusions about its status. The diagnostic rule  is based on Dornemburg's proportional method with 5 ratios as using input components. Output conclusions include “normal”, “over temperature" or "discharge". Multi-layer Perceptron (MLP) network is used with 5-M-3 network structure. Through training with the number of neutrals of different hidden layers, we selected M = 16. This number gives the most accurate diagnostic results. Through experimentation with actual data, the results show that the diagnostic system makes credible conclusions.