▎ 摘　　要

NOVELTY - The system has a processor which is programmed to receive a coarse mesh input including a first set of nodes. The coarse mesh is input into a computational fluid dynamics (CFD) solver with physical parameters to obtain a coarse mesh solution. A fine mesh input is received from a second set of nodes, where the second set of nodes includes more nodes than the first set of nodes. The fine mesh input is concatenated to the physical parameters and the concatenation is performed through a graphene convolutional layer to obtain a hidden fine mesh layer. The coarse mesh is Upsampled to obtain a coarse mesh upsampling that includes the same number of nodes as the second set of nodes. A hidden fine grid layer is concatenated with the coarse grid upsampling. A prediction is outputted in response to the concatenation of the hidden fine mesh layer and the coarse mesh upsamples. USE - System for combining differential partial differential equation solvers and neural graph networks for fluid flow prediction. ADVANTAGE - The hybrid neural network combines a traditional graph convolution network with an embedded differentiable fluid dynamics simulator on the network itself. The processor subsystem is designed to operate during the operation of the system to provide an iterative function to replace a stack of layers of the neural network to be trained. The system can use gradients and some deep learning optimization algorithm to set the parameters and the coarse grid positions. DETAILED DESCRIPTION - An INDEPENDENT CLAIM is included for a method for combining differential partial differential equation solvers and neural graph networks for fluid flow prediction. DESCRIPTION OF DRAWING(S) - The drawing shows a schematic diagram of the system for training a neural network. System for training a neural network (100) Processor subsystem (160) Data storage interface (180) Data memory (190) Training data (192)