Blood Glucose Control Based on Rapid Model Identification with Particle Swarm Optimization Method

Fully automatic controllers are designed to regulate blood glucose (BG) concentrations in people with Type 1 diabetes mellitus (T1DM) which must be controlled in a normal range. The model based on control algorithms may not obtain satisfied BG values if the prediction model of BG is mismatched. One of the challenges in glucose control is the lacking of accurate individual prediction models for T1DM patients because of limited modeling data. In order to solve this problem, in this paper, a rapid and economic modeling method in T1DM is first pointed out for glucose control. Using the idea of model migration with PSO, the ARX model structure is adaptively adjusted from person to person with limited data individually. Then the developed prediction model is used for glucose control using the zone model predictive control with the control-relevant constraints. The effectiveness of the proposed method is tested on the UVa/Padova metabolic simulator. It is shown that the proposed method has good control performance where 95% of the simulation time for glucose control can be regulated within the normal range. Further, the experimental results show that there are no statistically significant differences between the individual ARX model with sufficient data and the developed prediction model using paired t-test. The proposed glucose control method is effective and economic which can take place of repetitive subject-dependent control method especially when the modeling data are insufficient.