To accelerate the digital transformation of the power system and ensure the high-quality and safe operation of the high voltage direct current (HVDC) system, it is necessary to fully mine and extract the massive data and rich management experience accumulated in daily regulation, operation and maintenance of such a system through intelligent technologies. Thus, a knowledge map is constructed to assist the staff in diagnosing and dealing with faults. This paper presents a fault diagnosis method for an HVDC system based on wavelet transform and deep learning. First, the fault recording data (single-phase grounding, phase-to-phase short circuit and valve group short circuit) of the converter station is converted into two-dimensional time-frequency images by wavelet transform, and data enhancement technology is used to further expand the sample data set. Then, the ResNet50 network is used to achieve the fault diagnosis of the HVDC system. The results are that the classification accuracy of the proposed method in the training set is 93%, and the classification accuracy of the test set is 82%. This proves the effectiveness of the proposed method and provides a new feasible route for fault diagnosis of the system. To further test the proposed method, it is compared with GoogleNet, VGG16, AlexNet, SVM, decision tree and KNN. The results show that the proposed method performs better in fault diagnosis of the HVDC system.