In order to detect series fault arc more accurately, this paper proposes a series fault arc detection method based on wavelet packet decomposition, PCA downscaling and optimization Transformer-BiLSTM. Firstly, the current data collected from different loads under normal and series fault arc conditions are extracted by wavelet packet decomposition to obtain the energy entropy, multi-scale information entropy and energy spectrum to form the feature matrix, and then the feature matrix is subjected to principal component analysis and dimensionality reduction by PCA, and finally the dimensionality reduction results are used as inputs to the model. In order to fully learn the interdependence between the input features, Transformer is combined with BiLSTM; in addition, considering the significant influence of hyperparameters on the model performance, an improved particle swarm optimization algorithm based on the golden sinusoid is proposed to globally optimize the hyperparameters of the model. The experimental results show that the method proposed in this paper performs well in identifying series arc faults with multiple single and combined loads, with an average accuracy rate of more than 99%, which verifies the effectiveness and superiority of the method in this paper.