Modeling and control to improve blood glucose concentration for people with diabetes
Diabetes mellitus is a chronical condition that features either the lack of insulin or increased insulin resistance. It is a disorder in the human metabolic system. To combat insufficiency of insulin released by pancreas, a closed-loop control system, also known as artificial pancreas (AP) in this application, have been created to mimic the functionality of a human pancreas. An AP is used to regulate blood glucose concentration (BGC) by managing the release of insulin. Therefore, an algorithm, which can administer insulin to reduce the variation of BGC and minimize the occurrences of hyper-/ hypoglycemia episodes, is the key component of an AP. The objective of the dissertation is to develop an optimal algorithm to better control BGC for people with diabetes.
For people with Type 2 diabetes, prevention or treatment of diabetes mellitus can typically be done via a change of lifestyle and weight management. A virtual sensing system that does not require many manual inputs from patients can ease the burden for people with Type 2 diabetes. This dissertation covers the development of a monitoring system for Type 2 diabetes.
To achieve the goal of tighter control of BGC for people with Type 1 diabetes, dynamic modeling methodology for capturing the cause-and-effect relationship between manipulated variable (i.e. insulin) and controlled variable (i.e. BGC) has been developed. Theoretically, this dissertation has established that physiologically based nonlinear parameterized wiener models being superior to nonlinear autoregressive moving average with exogenous inputs (NARMAX) models in capturing dynamic relationships in processes with correlated inputs. Based on these results, wiener models have been applied in the modeling of BGC for real subjects with Type 1 diabetes under free-living conditions. With promising results shown in wiener models, an extended physiologically based model (i.e. semi-coupled model) has been developed from wiener structure, which enables the development of a phenomenologically sound feedforward control law. The feedforward control law based on wiener models has been tested in simulated continuous-stirred-tank reactor (CSTR) that demonstrates tight control of controlled variables. Further simulation runs with a CSTR also shows feedforward predictive control (FFPC) can provide tighter control over model predictive control (MPC). Lastly, for the special application of BGC control for people with Type 1 diabetes, FFPC demonstrates tighter control than MPC under simulation environment. To account for unmeasured disturbances and inaccurate models for manipulated variable in real life scenarios, feedback predictive control (FBPC) is developed and proven to be a more effective control algorithm under both CSTR and diabetes simulation environment, which can establish the foundation for tightening BGC in real subject clinical studies.