Abstract: In order to improve the accuracy of switched current circuit fault diagnosis, a feature extraction and recognition method of switched current circuit based on wavelet packet optimization and optimization of BP neural network was proposed. Firstly, the wavelet packet decomposition of the original response signal of the switched current circuit was carried out. Then, the normalized energy value after the decomposition of the N layer was calculated, and the optimal wavelet packet basis was selected by using the characteristic deviation as the evaluation. Finally, the optimal fault feature vector was constructed. The extracted optimal fault characteristics were classified by BP neural network optimized by genetic algorithm. The results of this method were verified by the example circuit. The results show that all the soft faults are effectively classified, and the superiority of the method in the fault diagnosis of the switched current circuit is illustrated.