This research background introduces the importance of digital distribution networks, emphasizes the limitations of traditional power system management methods, and clarifies the practical application requirements for fault diagnosis and prediction in digital distribution networks. Using digital distribution network data for fault diagnosis and prediction, using CNN (Convolutional Neural Network) to achieve high-precision identification and classification, and applying DL (Deep Learning) technology to actual power system operation. In the Results and Discussion section, this study summarizes the main findings of fault diagnosis and prediction. The research results indicate that in the comparison of the three models, the accuracy of our model is superior to GA (Genetic Algorithm) and CF (Collaborative Filtering), with the highest accuracy reaching 91.25%. The accuracy is 13.68% and 6.18% higher than GA and CF. In the recall test, our model has a high recall rate. In the comparison of F1 scores, our model's F1 score is closest to 1, indicating that our model performs very well in identifying and classifying positive class samples.