The Research Of Transformer Fault Diagnosis Based On Combining Multiple Classifiers

The transformer is important equipment in power system, its running state is directly related to the stability of power system. Due to the complexity of transformer fault, the diagnosis method is diverse.The present various single classifiers inadequacy in the transformer fault diagnosis, so in this paper, the author focuses on combining classifier complement multiple single classifiers to get better result. Four classifiers which are widely used in transformer fault diagnosis:support vector machines, neural network, nearest neighbor algorithm, Naive Bayes, and used the Meta-learning Strategy (including Stacking portfolio strategy and Cascading portfolio strategy)and Voting to construct two different combination classifier models. The experiments show that the transformer fault diagnosis model combination of the Meta-learning Strategy can make a superior precision than Voting. In this paper, the main works of this paper are expressed as follows:(1) Normalized and threshold of transformer oil dissolved gas data, and weighted K-nearest neighbor algorithm, used the upper and lower approximation to weight the Bayesian Classifier with tree augmented TAN classification algorithm, established transformer fault diagnosis models base on K-nearest neighbor algorithm, Support Vector Machine, Bayesian Classifier with tree augmented TAN classification algorithm and neural network respectively.(2) Weighted Voting by using the prior probability, structured the transformer fault diagnosis model combination of the Voting, and compared the transformer fault diagnosis model with the Voting without weighted; changed the number of the base classifier of multiple classifiers combination transformer fault diagnosis model, and gave a comparative analysis.(3) Used the Stacking Meta-learning Strategy to construct transformer fault diagnosis combination model, changed the combination model with Stacking Meta-learning Strategy’s the input variable type, and gave a comparative analysis; changed the’s number of base classifiers with Stacking Meta-learning Strategy, and gave a comparative analysis.(4) Used the Cascading Meta-learning Strategy to construct transformer fault diagnosis combination model, changed the combination model with Cascading Meta-learning Strategy’s base classifier sequence, and gave a comparative analysis; changed the’s number of base classifiers with Cascading Meta-learning Strategy, and gave a comparative analysis.(5) Compared the transformer fault diagnosis model used the Meta-learning Strategy with Voting.