Output-Feedback RLS-Based Model Predictive Control

Abstract

This paper presents recursive-least-squares-based model predictive control (RLSMPC). RLSMPC uses only output feedback, and thus does not require full-state measurements. Online learning is performed through concurrent system identification, and thus no a priori model is needed. RLSMPC employs separate RLS algorithms for identification, offset determination, and control. Variable-rate forgetting is used to facilitate system identification and offset estimation.