Research on Transformer Fault Type Detection Based on Model Fusion

In order to further improve the accuracy of power transformer fault type detection, a model fusion transformer fault type detection method based on Self-organizing Map-Local Outlier Factor-Isolation ForestSelf (SOM-LOF-iForest) is proposed. First, according to the fault gas concentration data, SOM algorithm, LOF algorithm, and iForest algorithm are used to detect the fault type, and then use Analytic Hierarchy Process to fuse and improve the decision weight values of the SOM algorithm, LOF algorithm, and iForest algorithm to establish the model fusion and to get a new model of fault decision making. The results show that model fusion method overcomes the problem of the diagnostic accuracy of a single model in a certain aspect, and has the advantage of high detection accuracy compared to the existing power transformer fault type detection method.