Due to the increasing of cable-wire hybrid lines, the zero-sequence current waveform similarity between faulty lines and sound lines increases greatly, which makes fault line selection more difficult for nowadays power distribution networks. To solve these problems, a fault line selection method of complete ensemble empirical mode decomposition based on adaptive noise (CEEMDAN) and independent component analysis (ICA) is proposed. The zero-sequence current at the head end of each line is decomposed to obtain intrinsic mode functions (IMF) component, and then multi-dimensional source signal is constructed by ICA method. The fault signal is identified by kurtosis threshold criterion, then the zero-sequence current signals are reconstructed after setting the fault signal to zero, and the line with zero-variance of reconstructed zero-sequence signal is identified as fault line. Simulation results show that the proposed method is less affected by different fault conditions, of high reliability and strong anti-interference, and can be used for line selection for high resistance grounding multiple faults. Meanwhile, the “sort push-pull” strategy based on kurtosis value is proposed.