Uncertainties and Modeling Errors of Type 1 Diabetes Models

Modeling and control are tightly connected if we want to guarantee safety and reliability. These are minimum requirements in the medical field. The more sophisticated methods usually require information beyond the available measurements, and one way or another incorporate all a priori knowledge. This can manifest in state estimation, model-based prediction, or robust design assuming the worst case, among others. The better the model the better the achievable control; however, all aspects of modeling are more difficult in the case of physiological systems compared to regular engineering applications. In the following, we will investigate how various errors resulting from modeling inaccuracies affect the prediction of the behavior in case of blood glucose prediction. Sigma-point filters are used to efficiently support Kalman filtering, while the error sources are introduced in a single uncertainty block.