Aiming at the problem that the fault prediction accuracy of distribution network transformer is low and cannot provide reliable basis for transformer operation and maintenance, digital twin and deep learning are introduced. By constructing the digital twin model of distribution network transformer, classifying the movement state of distribution network transformer; Innovatively deep learning fault samples to achieve intelligent prediction of transformer operation status in distribution networks. The experimental results show that the prediction results of this method are more accurate and the calculation time is shorter. This method has guiding significance for the knowledge system of digital twins in the application of distribution transformer fault prediction.