The failure of capacitor voltage transformer（CVT）has seriously affected the monitoring of power grid equipment. In this paper

a kind of method based on K-Means clustering of sliding time window is proposed to detect whether the internal capacitance unit breakdown fault occurs in CVT at early stage. According to the situation that the breakdown of internal unit in CVT can lead to voltage variation of the secondary side

the voltage data at the secondary side of the CVT is decomposed into many segments in the time domain on the basis of sliding window and data within the block is subject to K-Means clustering process

so to determine whether the CVT transmission internal capacitance breakdown failure occursthrough two kinds of center of mass data difference compared with the threshold of the results. It is shown by the example analysis that this method has a definite effect on the diagnosis of whether the internal capacitance unit of CVT is punctured

which is more efficient and sensitive than the traditional physical detection method.After further improvement

it can be applied to intelligent power grid system.