Preliminary model-free results of a Hungarian robust artificial pancreas algorithm

Kovács, Levente and Szalay, Péter and Sas, I. P. and Eigner, György and Almássy, Z. and Felszeghy, E. and Kocsis, G. and Fövényi, J. and Körner, A. and Kautzky, L. and Soós, H. and Orbán, A. and Niederland, T. and Juhászné Tuifel, A. and Tóthné Sebestyén, T. and Soós, A. and Török, A. and Barkai, L. (2013) Preliminary model-free results of a Hungarian robust artificial pancreas algorithm. DIABETES TECHNOLOGY AND THERAPEUTICS, 15 (Suppl1). A-96. ISSN 1520-9156

Text ATTD_2013_Abstracts.pdf Restricted to Registered users only Download (219kB) | Request a copy

Abstract

Introduction: Individualized control therapies give better and better results in the artificial pancreas researches. However, due to neglected dynamics the robustness of the methodologies is still a challenge. The Hungarian Artificial Pancreas Working Group (MAP) investigates this problem from several years. On previous ATTD conferences a robust control algorithm based on the Sorensen-model and its validation results were presented.. Aim: The model-free property of the algorithm is investigated based on the Hovorka-model. Methods: The aforementioned two Type 1 diabetes model (T1DM) was used to generate the virtual patients. Their parameters were identified using data recorded of 203 weeks of 90 T1DM patients in clinical environment from the MAP’s insulin pump centers (aged 6-52 years). Results: Hypoglycaemia is efficiently avoided and hyperglycaemia is reduced more then 75% to the real datasets if parameter identification and the starting point of the algorithm is well determined. Hence, simulations were started 1-2 days before the start of the identification timeframe in order to minimize problems caused by initial states. However, this requires more apiori information from the patient side. Conclusions: Use of hard constraints proved their efficiency even in case of model-free investigations. However, a robust identification technique is required to support the proposed control algorithm. Despite the required improvements, preliminary results show that the methodology has the potential to globally handle patient groups and efficiently support individualized control (ex. MPC) protocols. Further steps: Fault detection analysis of different life situations (like stress, physical activity) is required together with a robust identification methodology.

Actions (login required)

Edit Item