Method, device, equipment, medium and program product for training of physical field neural network with cross-scale separable grid representation

By assigning multiple resolutions to the grid basis functions and generating personalized fusion coefficients, the problem that traditional grid structures cannot take into account both global and local details is solved. This achieves the coordinated optimization of grid basis functions and network parameters, improving the expressive power and reconstruction accuracy of physical field modeling.

CN122242260APending Publication Date: 2026-06-19TSINGHUA UNIVERSITY
0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
TSINGHUA UNIVERSITY
Filing Date
2026-04-02
Publication Date
2026-06-19

AI Technical Summary

⚠Technical Problem

Traditional single-scale grid structures struggle to simultaneously capture both the global trends and local details of the physical field. The grid basis functions and neural network weights are independent, impacting the model's representation efficiency and reconstruction accuracy.

⚗Method used

A cross-scale separable grid representation method is adopted to assign multiple resolutions to the grid basis functions. Common features are captured by neural networks and personalized fusion coefficients are generated to achieve the collaborative optimization of grid basis functions and network parameters.

🎯Benefits of technology

It improves the model's ability to express complex wing geometry and its reconstruction accuracy, and enhances the model's adaptive expression ability and overall representation efficiency.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 1
    Figure 1
  • Figure 2
    Figure 2
Patent Text Reader

Abstract

This application provides a method, apparatus, device, medium, and program product for training a physical field neural network with a cross-scale separable grid representation. The method includes: inputting a sample set of aircraft wing shapes into a first neural network; assigning M resolutions to M grid basis functions to balance global and local feature learning; capturing common features among samples through the weights of the first network; determining M specific fusion coefficients for each sample and processing them through the M grid basis functions to obtain M grid basis function features; fusing the grid basis function features according to the fusion coefficients to obtain sample-specific modulation features; obtaining predicted features based on the modulation features and common features and reconstructing the sample; and updating the parameter values ​​of the first network weights and grid basis functions based on the difference between the reconstructed result and the sample. This application can improve the parameter efficiency and model expressive power of physical field modeling.
Need to check novelty before this filing date? Find Prior Art