Identify Parameters of Digital Twins by Gray-Box Evolutionary Dynamic Optimization

As many digital twin papers were devoted to simulation models capable of predicting system states given some parameters, this work-in-progress paper focuses on its inverse problem, Parameter Identification (PI), which aims to determine the optimal parameters of simulation models that produce states that best match the physical system. We handle the task of PI as a dynamic optimization problem and propose a gray-box evolutionary dynamic optimization (EDO) framework to track parameters when the states vary with time. The main idea is to enhance EDO by building an inverse data-driven model to predict the location of the optimal parameters. We also present a case study of nuclear reactor digital twin, to explain how to apply the proposed framework.