Applying Scientific Machine Learning Techniques to Power Electronics Parameter Prediction

Accurate parameters of the inverter-based resources (IBR) model are critical for the design of effective control systems. Both physics-based and data-driven methods present unique advantages and challenges in IBR parameter predictions. This paper proposes a hybrid method that uses machine learning tools coupled with a physics-based model to generate predictions of the system outputs. The scientific machine learning (SciML)-based method reduces the data size requirements and training time of machine learning and the computational needs of tra-ditional physics-based models. The proposed SciML method is accomplished in a novel way, wherein the loss equation of the machine learning techniques is modified to contain the known scientific equations for the system output. Simulation results on a buck converter demonstrate that the new method is able to produce more accurate predictions in less time than its non-assisted counterpart.