Physics-Embedded Data-Driven Modeling for Model Predictive Control of Battery Charging Converter

Model predictive control (MPC) is widely used for battery converter charging control due to its strong capability to explicitly handle multiple objectives and complex constraints in real time. However, its performance depends on the accuracy of the prediction model, which is often degraded by incomplete prior physical knowledge, parameter uncertainty, and aging effects, thereby necessitating data-driven modeling strategies. Moreover, periodic model updating and real-time control require both low training and inference time costs, which purely data-driven models struggle to achieve. To address these challenges, a physics-embedded dictionary-based system identification (PhD-SI) method based on the sparse identification of nonlinear dynamical systems (SINDy) method is proposed. It integrates physical knowledge with data in a compact nested linear-in-parameter matrix representation equipped with high efficient fractional-power monomial base functions with a closed-form sparse regression solution, enabling accurate system identification with low training and inference time costs under incomplete prior knowledge. Results on a battery charging case study demonstrate improved modeling accuracy, enhanced closed-loop control performance under disturbances compared with neural-network-based methods, and strong robustness to parameter uncertainty, indicating its potential to support battery charging system vendors in developing high-performance charging solutions.