Physics-Embedded Dictionary-Based Model Predictive Control for Electrical Vehicle Charging Systems

A Physics-embedded Dictionary-based System Identification (PhD-SI) method is presented for Model Predictive Control (MPC) in Electrical Vehicle (EV) charging systems. Compared with traditional Proportional-Integral-Derivative (PID) control, MPC excels at handling multi-objective tasks and making optimal decisions over longer prediction horizons and has better performance during critical transients. However, the effectiveness of MPC largely depends on the accuracy of the prediction model and the inference cost. As the physical model has high accuracy and low inference, it is a preferred choice for MPC. However, obtaining precise physical information is often challenging. On the other hand, fully data-driven methods suffer from limited generalizability. To bridge this gap, the physics-embedded data-driven approach, i.e., PhD-SI, is developed to identify the system dynamics for predictive control, leveraging both prior physical knowledge and learning from data. Additionally, the PhD-SI has an interpretable structure and a much cheaper time cost compared with the black box model, such as neural networks. A numerical example of the EV charging system demonstrates the effectiveness of the PhD-SI-based MPC, particularly in terms of computational efficiency and model generalizability.

Code for this research: ‘https://github.com/HanyangHe/EV-charging-project’