Conjugate gradient finite element model solving method and system based on sparse convolution preprocessing
By employing the conjugate gradient finite element model method with sparse convolution preprocessing and training a preprocessing sub-generator using a sparse convolution U-net network, the problem of low efficiency in solving large-scale sparse linear equation systems is solved, achieving efficient and stable solutions in various structural systems.
Patent Information
- Authority / Receiving Office
- CN Β· China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- BEIJING UNIV OF TECH
- Filing Date
- 2026-04-09
- Publication Date
- 2026-07-03
AI Technical Summary
When solving large-scale sparse linear equation systems in structural engineering, the applicability of traditional preprocessing methods is limited. Deep learning preprocessing research lacks adaptability and generalization ability in engineering scenarios, making it difficult to construct stable and efficient preprocessors, resulting in low computational efficiency.
A conjugate gradient finite element model method with sparse convolution preprocessing is adopted. The preprocessor generator is trained by sparse convolution U-net network to reduce the condition number of stiffness matrix and construct high-quality preprocessors. Combined with sparse convolutional neural network, the convergence speed is significantly improved and the solution time is reduced.
Achieving stable acceleration of 150%-420% in various structural systems improves solution efficiency and forms a unified and scalable preprocessing framework for structural engineering, applicable to beam element, solid element and hybrid element structures, and reduces computational costs.
Smart Images

Figure CN122333882A_ABST