In recent years, fire accidents caused by the fault of the electric bicycle charging circuit have occurred frequently. In order to identify the occurrence of arc fault in the battery charging stage, a feature identification method of AC series arc fault based on the combination of variational mode decomposition (VMD) and multi-scale sample entropy was proposed. The series arc fault experiments in the electric bicycle battery charging circuit was carried out by using a self-developed arc fault generator, and the measured circuit current signal was decomposed by VMD. The multi-scale sample entropy was calculated by using the decomposed IMF components and used as the series arc fault features. Combined with an optimized support vector machine, the series arc fault was identified. The results indicated that the method proposed in the paper can efficiently identify the arc fault in the battery charging stage.