Self-Repairing Predictive Control for Selective Catalytic Reduction Systems Based on Slow Feature Analysis and Closed-Loop Subspace Identification

Model–plant mismatch can impair predictive control of selective catalytic reduction (SCR) systems as operating conditions and process dynamics change. This study develops a self-repairing model predictive control (MPC) framework integrating slow-feature analysis (SFA), short-duration input excitation, and optimized predictor-based subspace identification (PBSIDopt). Its contribution is the coordinated monitoring–excitation–identification–validation–updating procedure rather than a new SFA or identification algorithm. A candidate model is accepted only after an independent multistep-prediction test and model/controller checks; the predictor, controller, and residual monitor are then updated together. Four representative simulations examine abrupt mismatch, two rates of gradual mismatch, and a different parameter-change direction. Relative to fixed-model MPC, the SFA framework reduced whole-run mean absolute tracking error by 5.63–42.91% and final-interval error by 24.07–69.08%. Compared with an exponentially weighted moving average (EWMA) monitor embedded in the same repair framework, SFA yielded lower whole-run tracking errors in three of the four representative scenarios. These illustrative simulation results support the feasibility of the integrated framework; they do not establish universal monitoring superiority or long-term industrial safety.