Abstract: In the field of computational fluid dynamics, the high-order discontinuous Galerkin（DG）method has the advantages of high-order format construction, geometric flexibility, adaptive calculation capability, and parallel computing efficiency. However, when the high-order DG algorithm is used to perform numerical simulation on the high-speed flow, numerical oscillations will occur near the shock wave, which will affect the stable progression of the DG calculation.This oscillation phenomenon is similar to the Gibbs noise in the field of image processing.With the development of modern science and technology, there are new ideas and options for using artificial intelligence methods to suppress numerical oscillations near shock waves.This paper proposes an intelligent denoising method for three-dimensional Gibbs phenomena to address the non-physical oscillation problem of high-order formats in the region near shock waves.In this paper, the idea of zero-shot learning and the graph attention model are adopted. Through cross-domain transfer and spatial modeling of graph structure, the high-efficiency suppression of numerical oscillations near the shock wave in three-dimensional cases is realized. numerically simulating the three-dimensional hypersonic cylindrical flow and three-dimensional forward step and three-dimensional hemisphere cases, it is verified that embedding the three-dimensional Gibbs phenomenon intelligent denoising model constructed this paper into the DG method calculation process can effectively eliminate the numerical noise near the shock wave, and the numerical oscillations near the shock wave are suppressed in an intelligent way, and the stability and shock wave capturing accuracy of the three-dimensional high-order DG are guaranteed.