Abstract Coronal mass ejections (CMEs) are eruptions originating from the solar corona, usually carrying a large amount of fast-flowing plasma into interplanetary space, accompanied by an enhanced magnetic field. CMEs can form shocks in the interplanetary space, accelerating charged particles to higher energies and forming solar energetic particle (SEP) events. CME-driven shocks have very complex 3D structures and keep evolving within the 3D solar wind environment. Therefore, it is crucial to obtain the 3D structure, position, and parameters of the shocks accurately and effectively, in order to study the acceleration and propagation process of SEPs and predict SEP events. In this work, a convolutional neural network (CNN) model is established based on deep learning technology, which can identify shock structures from 3D numerical simulation data of the CME propagation process. The recognition results of the CNN model are compared with traditional recognition methods to verify its reliability and overall performance. In addition, we simulated the CME event on 2021 December 4, and captured the associated shock for further testing. Notably, different CME initiation models were used during training and testing, further confirming the general applicability of the proposed CNN-based method. By calculating key shock parameters, we reveal the 3D characteristics of the CME-driven shock, including the spatial distribution and temporal evolution of the shock normal and velocity across the shock surface. The analysis demonstrates the advantages of our method in resolving the detailed structure and dynamics of the shock, offering a new perspective for understanding the origin and evolution of SEPs.
李 et al. (2026) studied this question.