Parameter identification of diabetes models using evolution algorithms and swarm intelligence

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

The diabetes mellitus is regarded as one of the most widespread diseases of the future, therefore particular attention should be paid to the research of this disease. The development of modern technology and the connection of the medical and the technical disciplines allow the creation of an artificial pancreas without human intervention. The human pancreas of type 1 diabetic patients could be replaced with the artificial one, keeping blood sugar level within normal range. Thereby the patient's life becomes easier, safer because the major risk factor of human error which is insulin overdose leading to coma or death, that is especially critical for kids, could disappear.

In my thesis, I was working on the improvement of an artificial pancreas closed loop control system. My first task was to study the most frequently referenced and used nonlinear model named after Hovorka [6], which describes human glucose and metabolic processes. Then I had to perform the identification of the model parameters resulting in “virtual patients” for the virtual test environment. This is a rather sophisticated problem, since the model is of high order and severely nonlinear. To solve the identification problem I implemented genetic algorithm and particle swarm optimization in MATLAB environment, because they can solve complex, nonlinear problems. The other part of my work was to learn sigma point filters which can be used for estimation of the variables of the Hovorka model. The critical points of these sigma point filters are the filter parameters. They determine the effectiveness of the estimation. In this semester I optimized these parameters with genetic algorithm and particle swarm intelligence, and I compared and evaluated the results.

In my thesis, the patient’s anonymity is handled with care.

Downloads

Please sign in to download the files of this thesis.