A Data Driven Control Strategy Adaptation Applied To Diabetes Meal Glucose Management

Diabetes is a disease resulting in the destruction of the insulin-secreting pancreatic beta cells or the body’s resistance to insulin. The consequent lack/reduced effectiveness of insulin, prevents body cells from utilizing glucose, leading to high blood glucose (BG) levels and associated long-term health complications. To avoid these health issues, patients with diabetes inject themselves with exogenous insulin fulfilling the fasting need and compensating for BG rises induced by carbohydrate intake. Overdosing insulin can lead to the life-threatening condition of hypoglycemia while underdosing to hyperglycemia and the subsequent health complications. Because of these implications and the numerous unknown factors determining post-meal BG, prandial insulin dosing is a major obstacle in diabetes management. This problem translates into a control problem with feedback and feedforward components that can be adapted, it is challenged by the extreme cost of control actions, the time-varying property of the system, and the abundant uncertainties. In this context, a data-driven method designed to minimize post-meal BG variance is proposed to adapt the controller components.