Worldwide, an estimated 350-536 million people are affected by diabetes mellitus. This disease alone caused around 5 million deaths in 2015, surpassing the total number of deaths from HIV/AIDS, tuberculosis and malaria. The chronically elevated blood sugar level increases the risk of dangerous long-term consequences for diabetics, such as kidney failure, retinal damage or heart attacks. In order to minimize the risk of these conditions, especially in type 1 diabetics who no longer have any endogenous insulin production, effective insulin therapy is required, which on the one hand avoids frequent hyperglycaemia and on the other hand protects the patient from acutely dangerous hypoglycaemic glucose levels. The best way to do this is to regulate blood glucose levels using the artificial pancreas. In such a device the patient's glucose level is monitored by continuously measuring sensors, and insulin is administered subcutaneously by an insulin pump. Numerous researchers are working on control strategies for these systems to improve the treatment of those affected. However, new methods must be tested in clinical trials for their reliability and safety. In the course of this, the importance of in silico tests is increasing in order to reduce the costs and risks of these trials. For this purpose, two methods were combined in this work in order to be able to make as reliable a statement as possible about the performance of new control strategies. First, an interval version of a mathematical metabolic model was created, which allows to determine the entire range of possible glucose values depending on the parameter variability of a patient. Subsequently, various procedures were used to identify parameters and parameter intervals from the measurement data of 37 patients and to test the meaningfulness of the predictions of the interval model. In the following, Deviation Analysis aproaches were used to make predictions about the glucose trajectories of patients. This method enables the integration of real measurement data in simulation studies to simulate the effect of different insulin doses on a real patient as realistically as possible. Finally, three different control strategies were tested and their effectiveness analyzed with these procedures. The effects of alternative bolus insulin quantities were investigated with both, Deviation Analyses and interval modeling. The results provide insight into the potential of the predictive models and control concepts used.