Subspace-based Model Predictive Control with Data Prefiltering

Subspace-based model-free predictive control algorithms directly estimate the relevant components of a predictive controller. Due to disturbances and noise in the measured data, the estimation results were often poor, which limited the applications of subspacebased model-free predictive controllers. By assuming a priori knowledge of the disturbance characteristics, this paper proposes a subspace-based model-free predictive control algorithm that utilizes the noise model for the estimation of the predictive control gain matrices. Simulation results show improved control results.