Development of glucose control algorithms for type 1 diabetes mellitus (T1DM) patients

In this study, novel open-loop and closed-loop control algorithms have been developed in order to keep the blood glucose (BG) value of Type 1 diabetes (T1DM) patients within the permissible ranges. The developed algorithms were tested on the UVA Padova T1DM simulator (T1DMS) and validated and the effectiveness of the algorithms was observed. In the first part of the study, a new open-loop control algorithm was developed for BG control of T1DM patients. The postprandial period with the greatest fluctuation is the time interval that determines the average BG value that patients will have. In this study, it was tried to obtain the best possible postprandial BG values with the help of the three parameters most frequently used by the patients. These three parameters are initial BG value, amount of carbohydrate to be taken at the meal, and difference time, which expresses the expected time until the start of the meal after insulin injection. In this study, it was predicted that due to the dominant linear effect in the system, all the extreme values of the BG parameter could be obtained by multiplying these three parameters with a patient-specific coefficient matrix. A patient-specific coefficient matrix was experimentally obtained from linear equations created with a kind of system diagnostic approach, using the records of these three parameters. Using this matrix, the ideal waiting time was determined. With the application of this period, the best postprandial BG values could be obtained. Thanks to the determined coefficient matrix, postprandial hypoglycemia, hyperglycemia and mean values of BG could be predicted with regression of 0.95, 0.99, and 0.98, respectively. In the second part of the study, closed-loop control algorithms were developed with a new approach for the BG control of T1DM patients, tested using T1DMS, and the algorithm efficiencies were observed. First of all, closed-loop BG regulation was tried to be done for 80 hours by using the classical Proportional-Derivative (PD) controller suggested in the literature. In the trials, it was observed that the limits of hypoglycemia and hyperglycemia were exceeded due to the incomplete linearity of the BG regulation mechanism and the coefficient optimization problems, and the desired result could not be obtained. The variable metabolic parameters of the BG regulation mechanism require a change in the control dynamics. On top of that, as a new approach, it has been tried to determine the controller coefficients with the help of system identification. First, a model of insulin metabolism was obtained by system identification, and then a BG regulation circuit equivalent to human metabolism was created using this mathematical model. By using this equivalent circuit, coefficient optimization was done with the help of a particle swarm optimization (PSO) algorithm. With this optimal controller, BG values were obtained in normal limits at a level that would meet expectations, and hypoglycemia was completely prevented. In addition, an Iterative Least Squares (RLS) based adaptive controller was added in parallel to the existing controller, since it would not be enough to optimize the controller coefficients once at the beginning so that the controller can adapt to the time-varying metabolic dynamics. When this new dual control structure was used, the rate of staying in the normal range of BG values increased to 91% and hypoglycemia did not occur. The same controller was retested in a new scenario of 200 hours, as a result, the time to stay in the normal range of BG was 90%, while the occurrence of hypoglycemia was completely eliminated. In this dual structure, the basic control function was made by the PD controller with its simple structure, and the most suitable coefficient optimization for the initial conditions was made with the PSO algorithm over the model obtained with the help of system identification. The adaptive control capability of the system to adapt to changing conditions has been achieved with the RLS-based controller. In this study, the effectiveness of PSO and RLS mechanisms on performance was also analyzed. It has been observed that the developed control structure can be used in all diabetes patients, especially in children who cannot perceive hypoglycemia, since the rate of realization of BG in the normal range is at the targeted level and hypoglycemia has been completely eliminated.