Method and system for reconstructing a physical field based on near-wall velocity field and sparse observation points

Abstract

The application discloses a method and system for reconstructing a physical field based on a near-wall velocity field and sparse observation points. The method obtains the reconstructed wall physical field through data acquisition, sparse observation point interpolation reconstruction, input feature construction, data block processing, model training and model inference. By integrating the near-wall velocity field dynamics information and the wall sparse observation point physical constraints, the CNN-Transformer hybrid architecture is used to realize the collaborative modeling of multi-scale features and global dependence, and the block processing effectively relieves the memory pressure of high-resolution data training, improves the training efficiency, and the overall process considers the reconstruction accuracy, physical consistency and calculation efficiency, solving the technical bottleneck that the traditional method is difficult to simultaneously satisfy multi-scale feature capture and global dependence modeling.

Images