In this paper, a CNN-SVM-based fault routing method is proposed for the problem of ground fault localisation in multi-terminal flexible DC distribution networks. The method combines the advantages of convolutional neural network (CNN) in feature extraction and the high efficiency of support vector machine (SVM) in fault classification, and achieves the routing of single-pole ground faults in distribution networks by analysing single-pole ground faults in multi-terminal flexible DC distribution networks. The current and voltage signals are transformed into time–frequency domain features by short-time Fourier transform and input into CNN for feature extraction, and finally the fault segments are classified by SVM. Simulation results show that the accuracy of the method reaches 99.52% and 98.89% on the training set and test set, respectively, showing good classification ability and strong generalisation performance. This study provides an efficient and reliable new scheme for ground fault routing in multi-terminal flexible DC distribution networks.