基于动态线性化的预测控制及其嵌入式延时补偿方法

Finite control-set model predictive control (FCS-MPC) has emerged as a promising control scheme for power converters. However, its dependence on precise physical parameters presents challenges to practical implementation. To address this inherent model dependence issue, this article proposes a data-driven predictive control strategy for LC-filtered voltage source inverters, utilizing full-form dynamic-linearization (FFDL) technique. To mitigate the inherent computational delay in digital implementations, conventional model-based and model-free MPC typically adopt a sequential prediction framework. In contrast to this widely deployed sequential prediction structure, this article constructs a holistic input-output data model that directly captures the mapping dynamics from the current instant to the delay-compensated future. Consequently, neither physical parameters nor delay compensation procedures are required in this design, rendering a simplified implementation. Furthermore, by leveraging the comprehensive data structure of FFDL, the proposed strategy achieves direct output capacitor voltage control in the absence of inductor current sensors and specific circuit parameters, thereby enhancing system reliability and reducing computational cost. Comparative experimental results on a three-level neutral-point-clamped (3L-NPC) inverter validate the effectiveness of the proposed method, demonstrating its computational efficiency and strong robustness against parameter mismatches compared to existing schemes.