IDENTIFICATION OF LINEAR DYNAMIC MODELS FOR TYPE 1 DIABETES: A SIMULATION STUDY

Models of type 1 diabetes with accurate prediction capabilities can help to achieve improved glycemic control in diabetic patients when used in a monitoring or model predictive control framework. In this research, empirical models are identified from a simulated physiological model. ARX and Box–Jenkins models of various orders are investigated and evaluated for their description of calibration and validation data that are characteristic of normal operation. In addition, model accuracy is determined for abnormal situations, or “faults.” The faults include changes in model parameters (insulin sensitivities), an insulin pump occlusion, underestimates in the carbohydrate content of meals, and mismatches between the actual and patient–reported timing of meals. The models describe normal operating conditions accurately, and can also detect significant faults.