Linear parameter varying (LPV) based robust control of type-I diabetes driven for real patient data

A roadmap of possibilities and boundaries of H controller design for the artificial pancreas problem is presented.LPV-modeling technique is used to handle direct the nonlinear model for the H linear control method.Weighting functions selection is presented based on knowledge accumulated from real patient measurements. Due to increasing prevalence of diabetes as well as increasing management costs, the artificial control of diabetes is a highly important task. Model-based design allows finding more effective solutions for the individual treatment of diabetic patients, but robustness is an important property that can be hardly guaranteed by the already developed individualized control algorithms. Modern robust control (known as H) theory represents an efficient possibility to solve robustness requirements in a general way based on exact mathematical formulation (Linear Matrix Inequalities) combined with knowledge-based expertise (through real patient data, uncertainty weighting functions can be formulated). When the difference between the nominal model and real patient dynamics is bounded and known, this approach becomes highly reliable. However, this requirement poses the greatest limitation since a model always represents an approximation of the complex physiological process. Consequently, the uncertainty formulation of the neglected dynamics becomes crucial as robust methods are very sensitive to them. In order to formulate them, large amount of real patient data and medical expertise is needed to cover the different life-style scenarios (especially the worst-case ones) that define the control space by the accumulated knowledge. On the other hand, Hbased methods represent linear control techniques; hence their direct nonlinear application is important for a physiological process. The paper presents a roadmap of using modern robust control in diabetes focusing on nonlinear model-based interpretation: how the weighting functions should be selected based on (knowledge-based) medical expertise, the direct nonlinear applicability of the method taking additional advantage of the recently emerged Linear Parameter Varying (LPV) methodology, robust performance investigation and switching control possibilities. During the control characteristics discussion, the trade-off between the medical knowledge-based empiricism and exact control engineering formulation is introduced through different examples computed under MATLAB on real diabetic patient data.