One of the most important disease of civilization is diabetes, or diabetes mellitus, which is a chronic disease of the body’s metabolic system. The cause of the disease is the absence of the insulin hormone, which is produced in the pancreas (absolute insulin failure), or the insensitivity of the body’s cells for insulin (insulin resistance, relative insulin failure) or both. As a consequence of absolute or relative insulin failure the level of the blood glucose rises, and this is, what the main cause of the symptoms of diabetes is.
The modeling of the body’s metabolism is an active field of research since the 70’s, so there were made a lot of mathematical models from the glucose-insulin metabolism.
In my thesis I fitted a model predictive controller, which uses the Mosca-Zhang constraints on two of the above mentioned models. I present the set-up of these controllers. Since state-space control algorithms are primarily used in the literature, and in practice there is only one measurable state variable, so I needed to fit an extended Kalman filter on the nonlinear models. This filter predicts the other state variables, wherewith the model predictive control calculates the optimal system input.
I compared the performance of these model predictive controllers with the performance of a modern robust controller (a generalized LQ or H2/H∞ method) fitted on the same models.