Abstract: The series arc fault is one of the main causes of electrical fire. A feature frequency band screening method based on the change rate of wavelet packet decomposition(WPD) energy proportion and a fault feature extraction method based on finite impulse response(FIR) filter are proposed. The series arc fault experiments in widely used industrial three-phase motor with frequency converter load circuit are carried out. Nine-layer decomposition is performed on the current signal by using the WPD. The feature frequency band of the series arc fault is determined according to the change rate of energy proportion of the signal with and without the fault in each frequency band. The feature frequency band signal of the fault current is extracted by using the FIR filter. The absolute average value and kurtosis of the feature frequency band signal are used as the features of the series arc fault. A support vector machine(SVM) optimized by particle swarm and grid search optimization algorithms is used to identify the series arc fault. The results show that the common feature frequency bands of the series arc fault in three-phase motor with frequency converter load circuit are 1.56-1.76 kHz, 2.93-6.25 kHz, and 9.38-10.94 kHz. The proposed method can accurately detect the series arc fault in three-phase motor with frequency converter circuit.