Diagnosis for Transformer Faults Based on Combinatorial Bayes Network
Wenqing Zhao, Yanfang Zhang, Yongli Zhu · 2009
A novel specific transformer fault diagnostic method based on combinatorial Bayesian network method with AdaBoostMl is proposed in this paper and a combinatorial transformer diagnostic tree augmented naive Bayes (TAN) model is set up. AdaBoostMl algorithm can improve the classification performance. The different TAN classifiers can be seen as a series of basic classifiers and are iterated through boosting. Based on the discussion of fault classification methods and a bias analysis of dissolved gas data of thirteen usual transformer faults, a combinatorial Bayesian network using boosting algorithm is introduced to realize the multi-resolution recognition of the insulation faults, which not only can make the fault diagnosis be more exact. Moreover, by comparing with the other method like naive Bayes, the proposed model reduces the error ratio, and recognition results show that this model is effective.