Physics-informed Machine Learning for Parameter Estimation of DC-DC Converter

Although various machine learning-based methods have been proposed for condition monitoring in power elec-tronics, they are challenging to be implemented in practice due to the accuracy, data availability, computation burden, explainability, etc. Physics-informed machine learning (PIML) has been emerging as a promising direction where the above challenges can be mitigated by incorporating domain knowledge. In this paper, we propose a PIML- based parameter estimation method for a DC-DC Buck converter, as an exemplary application of PIML in power electronics. By seamlessly integrating a deep neural network and the converter physical model, it can estimate multiple component parameters simultaneously with high accuracy and robustness, while based on a limited dataset. It expects to provide a new perspective to tailor existing ML tools for power electronic applications.