AI physical simulation method and system for calculating differential operator on unstructured grid

The invention relates to an AI physical simulation method and system for calculating a differential operator on an unstructured grid, and the method comprises the steps: carrying out the padding of input initial condition data, and constructing a sparse unstructured grid; calculating a differential operator item on the unstructured grid, embedding the differential operator item in the equation into a network structure, and predicting a system state quantity under the unstructured coarse grid by adopting a message passing graph neural network; and embedding the boundary condition of the equation into the simulation model by adopting a mode of simultaneously padding in the hidden space, so that the simulation model always meets the boundary condition, and AI physical simulation is carried out on the fluid. According to the method, under the condition that sparse training samples are used, the simulation precision of the complex physical field on the non-structural coarse grid is remarkably improved. 本发明涉及一种在非结构