Abstract: Aiming at the problem that current distribution transformer overload control method is too passive,power big data and deep learning technology are combined to propose a distribution transformer overload prediction method suitable for large-scale distribution network analysis. Firstly,based on the load characteristics of distribution transformer,a heavy-load attention index is established and a secondary filtering system is established to screen out the distribution transformers and their dates with higher risk of heavy overload as the prediction objects. Secondly,fully considering internal and external factors,a convolutional neural network-gated recurrent unit deep learning model is established to achieve load rate prediction and convert it into an early warning level. An example shows the effectiveness of the proposed method.