Aiming at the problem that the existing high voltage direct current（HVDC） transmission line fault identification method has low recognition accuracy and cannot accurately identify low resistance and high resistance faults at the same time, a fault identification method of HVDC transmission lines based on improved variational mode decomposition(VMD) - multi-scale permutation entropy(MPE) and general regression neural network(GRNN) was proposed. Firstly, the VMD improved by the whale algorithm was used to decompose the fault current signal, and the appropriate IMF components were selected to calculate the multi-scale permutation entropy and the IMF energy and ratio to extract the fault feature to form the fault feature vector. Secondly, the feature vector was input into the GRNN network for training with testing, the GRNN network’s high classification ability for small sample data was used to identify different types of faults. The results show that the proposed method has high accuracy in identifying different types of faults on HVDC transmission lines, regardless of whether low resistance or high resistance faults occur accurate identification, strong ability to withstand transition resistance, outstanding performance in small sample fault identification, and high reliability.