Physics-Informed Neural Network Model Predictive Control Research on Control Strategy of DC-DC Buck Converter

Power electronics devices often serve as interfaces that integrate renewable energy sources into power systems, playing a pivotal role in power systems. However, PE system models often face inaccuracies due to manufacturing defects, aging, or environmental factors affecting load conditions and component parameters. These factors compromise model precision, impacting system efficiency and reliability. Traditional model-based control methods face substantial challenges in achieving precise modeling under these conditions, which may lead to reduced efficiency or even render precise modeling unfeasible. This paper proposes an innovative hybrid control architecture that combines physics-informed neural network (PINN) with model predictive control (MPC). By leveraging the unique ability of PINN to integrate data-fitting capabilities with fundamental physical principles, accurate dynamic models of PE devices are constructed for MPC. This approach enhances MPC performance and improves the dynamic behavior and stability of PE systems. The efficacy of the proposed method is demonstrated through simulation results.