Metaheuristic Patient Estimation Based Patient-Specific Fuzzy Aggregated Artificial Pancreas Design

Patient-specific artificial pancreas design has been receiving increasing attention lately. In this article, using the chaotic bat algorithm (CBA), Hovorka–Wilinska (H–W) model parameters are estimated from nominal H–W virtual patient data. Using this identified H–W model for the virtual patient, multiple empirical second-order plus delay time (SOPDT) models representing glucose–insulin dynamics are derived for the range of blood glucose concentrations (BGCs) considered. Clustering of these models using the k-means algorithm yields three distinct clusters. Implicitly enumerated multiparametric model predictive controllers (mpMPCs) are designed using the cluster representatives. A fuzzy logic aggregation (FLA) of prediction and control improves the design parsimony. An insulin on board (IOB) safety trigger is designed using FLA of multiple full-order linearized CBA-estimated H–W models. The FLA-based mpMPC along with IOB and meal estimation are implemented on an embedded platform and by hardware-in-the-loop (HIL) simulation. In silico trials of the regulation of multiple meal disturbances are performed on the nominal H–W patient in MATLAB through serial communication. With meal estimation errors and varying insulin sensitivity, a very good low blood glucose index (LBGI) of