An improved fuzzy C-means clustering algorithm for transformer fault

Normal operation of the transformer is an important guarantee of the reliability of the power system. The transformer fault diagnosis is an important basis for transformer maintenance. Three-ratio method is widely used in oil-immersed transformer fault diagnosis, however the encoded value is too severe to correspond to the type of failure. FCM (Fuzzy C-means clustering) algorithm is introduced to solve this problem, and its performance determines the correct rate of transformer fault diagnosis. This paper focuses on the defects of FCM algorithm and in view of diagnostic data characteristic of three-ratio method, the FCM algorithm is optimized in two ways. Firstly, the introduction of outlier factor detection algorithm weakens the adverse effects caused by outlier factors in the dataset during clustering process. Secondly, the penalty factor in the objective function of FCM algorithm is used to maximize the difference of cluster object and cluster center. The experiments are performed on multiple sets of samples of transformer fault diagnostic from IRIS standard data sets. The results verify that the improved FCM algorithm for transformer fault diagnosis has a higher correct rate than traditional methods.