To detect the mechanical status of transformer winding more accurately

we propose that the optimal K-Means clustering algorithm is adopted to analyze the unstable and time-varying vibration signals of transformer under sudden short-circuit. Firstly

the characteristic vectors of each vibration signal are abstracted. Then the data density is introduced to optimize the K-means algorithm

which can more optimize the selection of initial centroids. The centroids of the characteristic vectors under different conditions can be calculated based on the optimized K-Means algorithm. The experimental results of vibration signals for a 110 kV transformer under sudden short-circuit show that the optimized K-Means algorithm has the advantage of high accuracy. The variation of centroid positions of characteristic vectors is capable of clearly reflecting the variation degree of winding deformation

which can provide the reference for the proposal of condition maintenance strategy.