Predictive Models for Type 1 Diabetes. A Case Study

Linear models were identified from glucose-insulin data of a type 1 diabetic patient. The models were used for simulation and prediction with 10, 30, 60, 90 and 120 min prediction horizon. The predictions were able to track the measured blood glucose, at least for the lower prediction horizons, so that hypo-and hyperglycemic events can be prognosed. The best predictions were achieved with ARX and ARMAX models, predictions for the subspace-based models were not as good as for the autoregressive models. Later a minimum variance controller was developed with an ARMAX model that was identified before. The controller could reduce the variance of the blood glucose concentration as well as eliminate hyperglycemic events. However it introduced the risk for hypoglycemia, which means that there is still some effort to be done. Linear models were identified from glucose-insulin data of a type 1 diabetic patient. The models were used for simulation and prediction with 10, 30, 60, 90 and 120 min prediction horizon. The predictions were able to track the measured blood glucose, at least for the lower prediction horizons, so that hypo-and hyperglycemic events can be prognosed. The best predictions were achieved with ARX and ARMAX models, predictions for the subspace-based models were not as good as for the autoregressive models. Later a minimum variance controller was developed with an ARMAX model that was identified before. The controller could reduce the variance of the blood glucose concentration as well as eliminate hyperglycemic events. However it introduced the risk for hypoglycemia, which means that there is still some effort to be done