All-DC wind power systems can effectively address issues such as reactive power and voltage fluctuations caused by AC cable capacitance, making them a growing focus of research. To tackle challenges such as the large number of collection branches, small differences in fault characteristics between adjacent lines, and difficulties in threshold setting, this paper proposes a fault line selection method based on the convolutional power energy ratio for all-DC wind power systems. First, the characteristics of transient power amplitude in the frequency domain for faulted and non-faulted lines are analyzed, revealing that within a characteristic frequency band, the transient power of faulted lines is greater than that of the non-faulted lines. Second, convolution power, as a time-domain fault characteristic quantity, is constructed to effectively extract the frequency domain power characteristics. Additionally, to improve the ability to withstand transition resistance, a time-domain convolutional power energy ratio is proposed between each collection line and the outlet of the collection bus. Analysis shows that the time-domain convolutional power energy ratio can effectively identify the faulty line within the characteristic frequency band. Combined with fault initiation and fault pole selection criteria, a complete fault line selection scheme is developed. Finally, PSCAD/EMTDC simulation results show that the proposed method can correctly identify faulty lines in all-DC wind power systems under fault resistance of 80 Ω and noise interference of 20 dB, without the need for simulation-based threshold tuning.