Abstract: When an early insulation fault occurs in a DC transmission line,the current fluctuation is small and the fault phenomenon is not obvious,so it is difficult to quickly identify and take protective measures. The topology of the photovoltaic power station line is complex,and it is difficult to accurately locate the fault location. In this paper, a method combining continuous wavelet transform and hybrid neural network model is proposed to identify and locate faults in the shortest possible time. The method first uses continuous wavelet transform to extract two-dimensional time-frequency matrix features from the transient zero-mode current signal,compresses it into a color image,and then sends the image to a neural network model for training. This hybrid neural network model improves recognition accuracy and reduces training time by combining convolutional neural networks and gated recurrent units. Finally,in order to verify the advantages of this method,four HVDC transmission lines are selected in a high noise environment to carry out four timefrequency analysis methods and three neural network model simulation comparisons,and then a separate simulation test is carried out to identify early insulation faults. The results show that this method can effectively identify the early insulation fault and locate the line where it occurs,and has strong anti-noise ability.