In order to improve power supply reliability

a detection method of incipient faults of power distribution lines is proposed. The proposed method takes the early fault waveform as the research object. By simulating the human cognitive process of the waveform

the early fault waveform is decomposed into different primitives

and the hierarchical probability learning method is used to establish the relationship between different primitive characteristics and different types of faults

so as to realize early fault type identification under small sample conditions. The experiments based on actual field operating data show that the accuracy of this method for early fault detection under small sample conditions is much better than that of deep learning methods under the same conditions. The research can provide a new idea for line fault diagnosis under the condition of small samples.