Multi-physical field time domain model order reduction method in data-driven electromagnetic system

Abstract

The invention discloses a multi-physical field time domain model order reduction method in a data-driven electromagnetic system. The innovation of the method lies in that scientific machine learning and model order reduction are combined for the first time, and the method is used for solving the electromagnetic system problem. According to the method, firstly, the spatial dimension of a data set is subjected to order reduction through an orthogonal projection (POD) method, and the data volume can be reduced by more than three orders of magnitude; then, reducing the data volume in the time dimension through a self-adaptive frame selection method; then, a data driving method of scientific machine learning is adopted, and a new model is obtained by learning from the reduced data set; wherein the reduced-order model can replace different applications of a full-order model to realize more than thousands of times of calculation acceleration. The method has the advantages of being high in expansion and small in error, is particularly suitable for small sample data driving situations, and can greatly improve the time domain calculation efficiency of multiple physical fields in an electromagnetic system.

Description

technical field [0001] The invention relates to a data-driven multi-physical field time-domain model reduction method in electromagnetic systems such as high-power semiconductor devices, integrated micro-systems and integrated circuits, and is particularly aimed at the coupling of electric (magnetic)-thermal-mechanical multi-physical fields in the electromagnetic system Action scenarios to achieve accelerated computing. Background technique [0002] Coupling solutions of multi-physics in electromagnetic systems such as high-power semiconductor devices, integrated microsystems, and integrated circuits often face the problems of huge computational complexity, large memory footprint, and slow solution convergence. To this end, the present invention proposes a time-domain model reduction method for multi-physics in a data-driven electromagnetic system, which can realize accelerated calculation of multi-physics and significantly reduce memory requirements. [0003] In the prior ...

AI Technical Summary

Problems solved by technology

The invention is to solve the problem that the traditional slow coherent method is difficult to apply to the distributed power generation network

Aiming at the widely used three-phase droop control grid-connected inverter, the invention proposes a model reduction method suitable for droop grid-connected inverter network, which can apply the slow coherent algorithm to the droop control inverter network , and can improve the grouping results of the traditional slow coherent algorithm, and further improve the accuracy of the reduced-order model, but this invention cannot solve the problem of multi-physics finite element (FEM) simulation in electromagnetic systems such as high-power semiconductor devices, integrated microsystems and integrated circuits. The model reduction problem in

[0004] Literature (M.Wang, H.-X.Li, X.Chen, and Y.Chen, "Deep learning-based model reduction for distributed parameter systems," IEEE Transactions on Systems, Man, and Cybernetics: Systems, vol.46, no.12, pp.1664–1674, 2016) also uses a non-invasive method, which can work when the control equation is unknown, but there is a problem of high training cost

Images