Improvement of Transformer Gas-in-Oil Diagnosis Based on Evidence Theory

Artificial intelligence (AI) algorithms are increasingly used to improve interpretation of gas-in-oil analysis. However, a satisfactory diagnosis may not be obtained if individual method is used. Each method has its own advantages and weaknesses, and the ability to identify fault is limited by its basic principle, training samples and other factors. Dempster-Shafer theory can be used to implement information fusion by combining the evidence bodies formed from several different AI methods. But the traditional Dempster-Shafer theory doesn´t consider the credibility and the correlation of evidences during the combining process. To improve the performance of transformer fault diagnosis, a novel diagnosis model based on modified Dempster-Shafer combination rule is proposed in this paper. The testing results show effectiveness of this proposed diagnosis model with enhancing the ability of transformer fault recognition.