In order to improve the deep learning training efficiency of the large reflector antenna active adjustment technique,this paper synthesizes the characteristic that each actuator can only adjust the panel connected to it,and proposes a divided-ring antenna active adjustment deep learning training modeling method. The method organizes panel node data according to actuator ring positions,using panel displacements as input features and actuator adjustments as output labels. Through systematic sorting,reorganization, and normalization, the ring-divided data are transformed into grid-structured tensors suitable for convolutional processing. Multi-layer convolutional neural networks are then constructed for surface adjustment prediction, optimized through a hybrid strategy combining simulated annealing and the Adam algorithm. Through the dataset divided-ring preprocessing, active adjustment neural network construction and model training for the case of an 8 m reflector antenna, the analytical results show that the proposed method can effectively shorten the training time, and the final model’s prediction accuracy is greatly improved,which demonstrates the feasibility and effectiveness of the proposed method.