The health status of power transformers can be effectively reflected by on-line monitoring the dissolved gas in oil. However

if the monitoring device works in abnormal conditions

it is difficult to ensure the accuracy of the condition assessment. At present

tedious manual on-site verification is adopted to identify abnormal monitoring devices

it is hard for the criterion fusion method to identify minor step anomaly

and it takes a long time to complete the identification. To settle the problems

we put forward an abnormally fast identification method of dissolved gas device based on Kernel Principal Component Analysis（KPCA）.This method can be adopted to identify abnormal data quickly according to Hotelling-T2 and Squared Prediction Error statistics and identify abnormal types with abnormal data characteristics. The test results show that KPCA can be adopted to effectively identify step mutation anomalies with a minimum amplitude of 5% that are difficult to be recognized by threshold method. To identify the abnormal working state of 677 monitoring devices in a regional power grid with this method

the test accuracy rate is 95.7%

which is not far from accuracy of the criterion fusion method 96.9%

but the time consuming is far less than that by the criterion fusion method. Therefore

the abnormal state of dissolved gas in oil monitoring devices can be identified quickly when the accuracy is not significantly reduced.