Computational grid parallel region division method and device based on reinforcement learning

Abstract

The invention provides a computational grid parallel region division method and device based on reinforcement learning. The method comprises the following steps: carrying out initial region division on a computational grid, taking a division result as an initial state, and taking a region to which each unit in the computational grid belongs as an action to construct an action value function according to a reinforcement learning algorithm; taking the single-step operation speed of the solver on the cluster as an award, and performing iterative computation on the action value function accordingto the award to obtain a convergence function value; according to the iterated state, action and convergence function value, realizing grid division of the parallel region. According to the method, the computational grid parallel area is divided through reinforcement learning training, the load balance condition in the program running process is effectively improved, and the time for inter-area message communication is shortened.

Description

technical field [0001] The present invention relates to the field of computers, and more specifically, to a method and device for parallel area division of computing grids based on reinforcement learning. Background technique [0002] CFD (Computational Fluid Dynamics) refers to a technical method that uses mathematical methods to simulate related phenomena such as flow, heat transfer, and chemical reactions of design fluids in a computer. The basic characteristics of CFD are numerical simulation and computer experiment. It starts from the basic physical law and largely replaces the expensive fluid dynamics experimental equipment, which has a huge impact on scientific research and engineering technology. With the development of science and technology, especially the rapid development of computer hardware technology and industry, it has laid a solid foundation for the development and promotion of CFD technology. At present, CFD technology has been applied to aerospace, marin...

AI Technical Summary

Problems solved by technology

In actual situations, due to the different computing capabilities of hardware devices of different nodes, when using parallel area division software, the geometric elements corresponding to different areas need to be manually given weight coefficients for uneven division, and inappropriate weight coefficients will increase significantly Unbalanced load at runtime, resulting in the inability to fully utilize the computing power of the device used, resulting in a significant increase in computing time

Images