Abstract: In underground small-current grounding power supply systems, the decomposition performance of the single-phase grounding fault line selection method based on Variational Mode Decomposition (VMD) depends heavily on the selection of parameters such as the penalty factor and the number of decomposition modes, which are difficult to set uniformly for different signals. To address this problem, a fault line selection method for mine small-current grounding based on optimized VMD and Random Forest (RF) was proposed. The Crested Porcupine Optimizer (CPO) was used to adaptively optimize the key parameters of VMD, including the penalty factor and the number of decomposition modes. A simulation model of underground power supply lines was established on the PSCAD/EMTDC platform. Zero-sequence current data under different fault conditions were obtained by changing the grounding resistance, initial fault phase angle, fault line, and fault location. The optimized VMD was applied to decompose the fault zero-sequence current signals. The modal components of each line were extracted, and their sample entropy was calculated to construct multidimensional feature vectors that reflected the complexity and nonlinear characteristics of the signals. The feature vectors were then input into the RF classifier for training and identification to achieve accurate determination of the fault line. The simulation results showed that the accuracy of the RF classifier was 98.3%, which was higher than that of Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM) network, and Extreme Learning Machine (ELM). The experimental results showed that the proposed method achieved a fault identification accuracy of 97.5%, unaffected by factors such as transition resistance, initial phase angle, and fault location, demonstrating high accuracy and applicability.