The invention discloses an online monitoring method for abnormal operation of a
voltage transformer, and the method comprises the steps: firstly collecting the
secondary side voltage data of a CVT in real time, forming a
time sequence, building a bilateral sliding window with a to-be-measured point as a center, and sliding along a time axis to cover a
complete sequence; then, K-Means clustering initialization is executed on each sliding window, the clustering number K is set to be equal to 2, two initial centroids are randomly selected, and the distance between each piece of
voltage data and the two centroids is calculated; thirdly, dividing the voltage value in the window into the closest
centroid category, and recalculating a new
centroid until the position of the
centroid is stable or reaches the maximum number of iterations; and finally, calculating a current window clustering centroid distance in real time, comparing the current window clustering centroid distance with a dynamic threshold value obtained by historical data training, and judging whether the CVT has an internal fault or not. The method utilizes the existing voltage data, does not need an additional
measuring instrument, can adapt to different operation conditions, avoids the over-alarm problem of a traditional fixed threshold value method, and improves the accuracy and robustness of fault diagnosis.