On-line and efficient monitoring of leakage faults in district heating network can effectively increase the quality of heat transmission and reduce energy consumption. However, the data feature extraction ability of conventional leakage fault diagnosis method is limited, and it is difficult to deal with the high dimensional nonlinear pressure flow monitoring data for complex heating network, which makes its diagnostic performance weak. Therefore, a fault diagnosis model of heating network leakage based on convolutional neural network (CNN) and Transformer was proposed. The proposed CNN-Transformer diagnostic model combines CNN and Transformer network to realize joint learning of different time scales and spatial features. The CNN network was used to extract local features, and the Transformer network was used to capture global features. The validity of the model was verified by simulating the fault data set of the annular heating pipe network system. The results show that the proposed CNN-Transformer diagnosis model based on multi-stage feature extraction and fusion mechanism of fault features significantly improves the accuracy of leak diagnosis. The CNN-Transformer method has the highest accuracy on the test set, with an accuracy increase of 13.21%, 7.49%, 6.1% and 4.62%, respectively, compared to other fault diagnosis methods including long short-term memory network, gate recurrent network, CNN and Transformer.