A methodology for the comparison of traditional MPC and stochastic MPC in the context of the regulation of blood glucose levels in Type 1 diabetics

Type 1 diabetes is a major health issue. Approximately 8% of the world's population have diabetes of one form or another and 10% of these have Type 1 diabetes. The disease is extremely debilitating and difficult to manage. This is due to many factors including large inter- and intra-patient variability in the model parameters and the one-sided nature of the available control (i.e., insulin can be added but not removed). Consequences of poor blood glucose regulation include cardiovascular disease, coma and, in extreme cases, even death. Because of its importance, there has been a huge world wide research effort aimed at developing improved treatment strategies. Recently proposed solutions include many advanced control methods. At the forefront of these developments has been a major effort aimed at applying Model Predictive Control (MPC) to this problem. However, there is a difficulty central to the diabetes problem. Specifically, there exists substantial variability, on a daily basis, in both model parameters and the nature of the disturbance inputs (i.e., food consumption, exercise and stress). Hence, the prediction aspect inherent in MPC is fraught with difficulties. One way of addressing this issue would be to utilise stochastic MPC which, among other things, considers a set of possible future disturbance scenarios. This paper will present a methodology for comparing benefits of regular MPC versus stochastic MPC when applied to Type 1 diabetes treatment.