Abstract

Model-free predictive control (MFPC) is developed to address the issue of the weak robustness in model predictive control (MPC) based on a physical model. In predictive control algorithms, the knowledge of future values of the reference is vital and this information is conventionally obtained using the Lagrange extrapolation algorithm. However, during implementation numerical errors introduced by inaccurate time-shifts compromise the adaptability of the data-driven model and its accuracy to reflect the plant's dynamics and operating states. To overcome these challenges, a time-series-based model-free predictive current control (MF-PCC) is proposed that employs the ordinary kriging (OK) algorithm in the continuous-control-set (CCS) type, and applied to a permanent magnet synchronous motor (PMSM) driving system as a current controller. The prediction errors for the time-series model with the typical prediction and ultra-localized time-series structure are analyzed in the CCS-type. Based on the this time-series model, state variables are predicted by the time-shift using the OK algorithm, replacing the Lagrange-based approach in the conventional methods, to reduce the prediction error. Both simulation and experimental results demonstrate the effectiveness of the proposed method, highlighting its superiorities in current quality and model accuracy, as well as its enhanced robustness.