Control-Oriented Linear Parameter-Varying Model for Glucose Control in Type 1 Diabetes

PATRICIO COLMEGNA; RICARDO S. SÁNCHEZ PEÑA; RAVI GONDHALEKAR

Buenos Aires

Conferencia; IEEE Multiconference on Systems and Control; 2016

IEEE

The contribution of this paper is a controller designoriented model of insulin-glucose dynamics in Type 1Diabetes Mellitus (T1DM). The novelty of the proposed modelis to more effectively include the time-varying nature, andalso the inter-patient variability, associated with the glucosecontrolproblem. Importantly, this is achieved in a mannerthat straightforwardly facilitates well-known and standardcontroller synthesis procedures. In that way, an average LinearParameter-Varying (LPV) model that captures the dynamicsfrom the insulin delivery input to the subcutaneous-glucoseconcentration output is constructed based on the Universitiesof Virginia (UVA)/Padova metabolic simulator. In addition, asystem-oriented reinterpretation of the classical ad-hoc 1800rule is applied to adapt the model?s gain.In order to quantify the effectiveness of this approach, thenu-gap between this new modeling strategy and the UVA/Padovamodel is computed at different glucose levels, together with theRoot Mean Square Error (RMSE) between the glucose deviationpredicted by both models. For comparison purposes, both open-(RMSE) and closed-loop (nu-gap metric) metrics are also determinedfor other control-oriented models previously presented.From the results, it could be concluded that the proposedmodel is more suitable for controller design than the othercontrol-relevant models, due to the fact that it provides abetter fit and a smaller distance to the UVA/Padova model in aclosed-loop setting. Furthermore, it would allow one to obtaina LPV controller for both postprandial and overnight controlin a straightforward way, considering its affine dependence onthe glucose level, which is measured in real-time.