Abstract: Correctly distinguishing abnormal data in the telemetry data collections is the key to improve the reliability of power distribution automation terminals in removing short circuit faults. On the basis of analyzing the basic characteristics of telemetry collected fault data, normal data, and abnormal data, a method for distinguishing abnormal data is proposed. It is to classify the original data of the same cycle collected by telemetry into several groups, calculate the similarity coefficient between adjacent groups, and uses the obvious dispersion of the similarity coefficient between adjacent groups when there is abnormal data, which is regarded as a criterion to identify abnormal data so as to improve the reliability of the distribution automation terminal to remove short circuit faults.