Structurally Tightened Robust MPC for Closed-Loop Artificial Pancreas with Uncertain Meals and Colored Sensor Errors

Blood glucose concentration (BGC) is a safety-critical variable in automated insulin delivery for Type 1 diabetes mellitus (T1DM), where actuator limits, glucose safety bounds, uncertain meals and imperfect CGM measurements must be handled. In postprandial control, meal feedforward boluses can consume insulin headroom and make robust feedback corrections infeasible. Conventional tube-based robust MPC often protects constraints by accumulating open-loop worst-case uncertainty over the horizon, which can be conservative when CGM errors are temporally correlated and meals are only partially announced. We propose a Structurally Tightened SRF-based Robust MPC (ST-SRF-MPC) for artificial pancreas regulation. The method separates basal-deviation meal feedforward from robust feedback, embeds an AR(1) bounded-error CGM model into the observer-error dynamics and computes horizon-indexed tightening margins from finite-horizon closed-loop reachable sets. A supervisory feedforward-scaling program is used for meal-announcement re-planning events to recover hard feasibility when possible, while preserving pump limits and the hypoglycemia lower bound on the MPC sampling grid; only the hyperglycemia upper bound is softened in fallback operation. Theoretical analysis establishes sampling-grid recursive feasibility under the stated bounded-uncertainty assumptions, while nonlinear BMM simulations show reduced tightening conservatism relative to constant tube margins and an enlarged scanned feasible operating region.