A direct current (DC) system is key to maintaining the normal operation of a substation. To address the difficulties of detecting ground faults in DC system ring networks, a ground fault detection method combining Harris Hawk optimization (HHO) and wavelet neural network (WNN) is proposed based on the traditional low-frequency signal injection method. The proposed method first collects the status of the ring network operation by low-frequency signal injection and then applies the Mallat wavelet decomposition algorithm to decompose the initial signal of each branch into four layers. The low-frequency coefficient waveforms are obtained, and the waveform energy is calculated to construct the WNN input sample set. Finally, the HHO algorithm is used to optimize the WNN parameters, and the optimized and trained WNN is applied to the branch circuit detection of ground faults. The simulation results show that the proposed method can effectively detect ground faults of ring networks in the DC system.