In order to improve the reliability of the coal mine power supply system, the simulation experiment in different voltage, current, power factor and environmental relative humidity conditions is carried out. The influence of different experimental parameters on the arc fault is analyzed, eigenvector constituted by the number of passing zero, normalized variance and covariance of current signals of adjacent five cycles on series arc fault are extracted, and a diagnosis model of series arc fault is established based on random forest classification algorithm. The training samples and test samples constituted by eigenvector of the normal operation and fault arc current signals are served as the input of random forest model, which are sorted to further diagnose whether the series arc fault is occurred. The results show that the method can effectively realize the diagnosis of series arc fault on mining electric connector.