Skip to Main Content
This paper analyzes the robustness properties and modeling requirements for model-predictive control via Horizon Predictive Control (HPC). The theory of Structured Singular Values is used to determine optimal values for the correction horizon in HPC given user-provided uncertainty intervals and performance weights. Regarding system identification, control-relevant identification principles are used to provide guidelines for input signal design, prefiltered estimation, and uncertainty modeling. These results are tested experimentally using data from a methanol-isopropanol distillation column.