LCC The invention discloses an HVDC continuous commutation failure early warning network training method. The method comprises: acquiring a direct current side in the time window; the electrical information of the alternating current side and the trigger pulse information of the control system are used as initial characteristics for cascading fault judgment; the training method of the network comprises the following steps, firstly, training an auto-encoder based on original simulation data; an encoder part of the convolutional neural network classifier is taken as a shared layer input of theconvolutional neural network classifier, and then classification task training of the convolutional neural network is carried out, so that the feature dimension is reduced, the calculation amount in the neural network training process is reduced, and the accuracy of a classification task of the convolutional neural network is improved. According to the method, occurrence of serious continuous commutation failure accidents can be rapidly predicted, the problem of training difficulty caused by high-dimensional eigenvectors is solved, meanwhile, the calculation speed under engineering field application is increased, and the method has wide engineering application prospects.