By analyzing the composition and characteristics of fault transient zero sequence current in a resonant grounding system, a fault line selection method based on correlation clustering of transient non-power frequency components is proposed. First, the power frequency component and noise signal in the transient zero sequence current are separated accurately by combining the variational mode decomposition with the fast Fourier transform, and the transient non-power frequency component is obtained. Then, a local weighted regression algorithm is used to smooth the non-power frequency component of each outgoing line to obtain the overall trend of the line non-power frequency component, and calculate the correlation coefficient matrix of each outgoing line. Finally, the fault line selection is realized by a K-means clustering algorithm without threshold setting. The simulation results show that the fault line selection method proposed can accurately select the fault line, and the fault line selection result is not affected by fault resistance, fault initial phase angle and other factors, and has good anti noise ability. This work is supported by the National Natural Science Foundation of China (No. 51867005) and the Science and Technology Plan of Guizhou Province (No. [2018]5615).