Systems and methods for high-speed computationally efficient operation and training of machine learning-based simulation models characterizing a physical system

Abstract

Systems and methods are provided for a computer-implemented method for simulating operation of a physical system represented by a plurality of mesh cells. A plurality of pre-simulations of the physical system are performed, where a pre-simulation determines an output metric value for each mesh cell based on a plurality of input metric values. A dimension reduction is performed on results of the pre-simulations to generate condensed training data, and a simulation model is generated using a machine learning algorithm to train a plurality of sub-models for simulating output metric values for the plurality of mesh cells based on the condensed training data. The simulation model is configured to provide simulation inputs to the trained sub-models to generate an output metric value for each of the plurality of mesh cells.