Parameter identification of diabetes models using sigma-point filters

OData support
Dr. Drexler Dániel András
Department of Control Engineering and Information Technology

Researches on artificial pancreas are extremely popular nowadays, because type 1 diabetes is a frequently occurring problem world wide. Type 1 diabetes is incurable and strongly affects life quality. Ultimately this condition can be fatal.

A key aspect of the artificial pancreas research is to accurately describe the human glycemic and metabolic processes with a usually non-linear model. We would have the opportunity to create virtual patients for the test environment by identify the parameters of the non-linear model. Furthermore, model-based control methods require a reliable model. Accurate modeling could speed up the artificial pancreas development. The non-linear model is inaccurate, loaded with noises and disturbances, so the simultaneous estimation of the state variables and model parameters could significantly improve the effectiveness of identification. I dealt with estimating the state variables in my thesis.

I started the state estimation with implementing a Kalman filter first, then I extended it to nonlinear systems as an Extended Kalman filter. The Extended Kalman filters disadvantage is the error coming from the linearization. To avoid this problem, I performed the state estimation with sigma point filters.

After the estimation of state variables I supplemented the filter algorithm to make it capable to estimate the model parameters.

The results were evaluated both on simulated and real clinical data sets.


Please sign in to download the files of this thesis.