Abstract: In this paper, a fault diagnosis method is proposed for existing DC microgrids to address the challenges of speed and accuracy. The proposed method combines the strengths of convolutional neural network (CNN) and bidirectional long short-term memory (BiLSTM) network, incorporating an attention mechanism. Specifically, CNN is utilized to extract vertical detailed features from fault data at a specific moment, compressing the data length and reducing subsequent network training parameters to improve the speed of fault diagnosis. Furthermore, we construct a cascaded network with BiLSTM as the core, enabling the extraction of horizontal historical features from fault data during the fault evolution process. The attention mechanism is integrated to enhance the model's focus on the feature changes in fault data, thereby improving the accuracy of fault diagnosis. Simulation results demonstrate that the proposed method outperforms mainstream fault diagnosis methods in terms of accuracy and recognition speed. Additionally, the proposed method exhibits excellent diagnostic performance for fault record data under conditions of noise interference, imbalanced samples, and small sample sizes.