Data–driven NARX–based digital twin thread modelling of Boost converter inductance current under closed–loop output voltage control

This paper introduces a pioneering approach to optimizing input and output delays in Nonlinear Autoregressive with Exogenous Inputs (NARX) Neural Network (NN) models through the use of Mutual Average Information (MAI), enhancing the dynamic modeling of power converters. Our research advances the development of digital twins by employing a data– driven methodology that significantly improves model accuracy for systems under variable loads. The novelty of our work lies in applying the MAI for delays selection in NARX–NN models, a technique that ensures high fidelity in capturing the intricate dynamics of power converters. The validation of our approach is underscored by the use of an advanced simulation model that not only embodies key non–linearities and parasitic components but is also under closed–loop control with a PI compensator. This adds a layer of complexity and realism to our model, illustrating its potential for real-world application. By demonstrating the improved performance of digital twins in power electronics through this refined delay optimization process, our study opens new avenues for future research and application in the field.