Dissolved gas analysis (DGA) in transformer oil is an important technology to identify transformer fault types

and fuzzy clustering is an effective analysis method. However

the traditional fuzzy clustering algorithm is sensitive to the randomly initialized cluster center and the effective measurement range of membership function is small

which makes it easy to fall into local extremum points

so the actual classification effect is not good. To address the shortcomings of the traditional FCM. Canopy algorithm was first used to conduct rough clustering of DGA data

and the results were used as the initial cluster center and optimal cluster number of subsequent FCM clustering

which reduced the subjectivity of artificial and random initialization parameters. Then

the iterative function of FCM membership was reconstructed by introducing the similarity index in the form of negative exponential function

which reduced the possibility of the algorithm falling into local extremum points. Finally

the effectiveness and practicability of the improved algorithm in transformer fault classification are verified by analyzing the fault gas data.