Three-phase motor and inverter loads occupy an important position in production and life. To reduce the occurrence of series arcs of fault, a method of arc of fault identification based on the combination of Variational Modal Decomposition (VMD) and entropy squared Euclidean distance with Kernel Extreme Learning Machine (KELM) optimized by the dung beetle algorithm is proposed. Experiments were designed for arc fault experiments under different inverter parameters, and the sparse indicator method was used to determine the VMD parameters, calculate the permutation entropy and approximate entropy of the decomposed eigenmode function, and the squared Euclidean distance of the permutation entropy and approximate entropy of the IMF4 was selected as the feature input to the Dung Beetle Optimizer-Kernel Extreme Learning Machine (DBO-KELM) model with an identification accuracy of more than 94%. Compared its recognition effect with other models, and found that the accuracy is improved compared to other models, and the detection effect is good. The feature still has a high accuracy rate in other classification models, indicating that the feature is effective in identifying series fault arcs under different operating conditions.