Abstract: In response to the inability of visual observation, infrared temperature measurement, leakage power frequency current, and DC component methods to quickly detect the internal moisture and insulation defects of metal oxide arrester（MOA） on 10 kV overhead lines, this paper proposes a method and principle based on K-means intelligent identification of defect types. Experimental samples of defects such as internal moisture and valve plate cracks in 10 kV MOA are made separately. Under the condition of pressurizing to the rated voltage of 10 kV without a partial discharge boosting device, a high frequency current sensor（HFCT） is used to collect raw partial discharge data, extract feature quantities, and establish a corresponding defect type database. By measuring 13 sets of data under different sub steady states of 10 kV MOA at different test points during live operation, the results show that this method can accurately identify internal insulation defects and moisture defects of 10 kV MOA, verifying the practicality of the K-means based high-frequency partial discharge 10 kV MOA fast live detection method, which has high economic and social value.