A Tube-NMPC Approach for Robust Control of Glucose in Type 1 Diabetes Mellitus

Abstract

Blood glucose concentration (BGC) is a significant indicator of human health and plays an essential role in maintaining a healthy life. Type 1 diabetes mellitus (T1DM) is an immune-mediated deficiency disease that causes a substantial deficit in insulin secretion. However, hyperglycemia can lead to a range of complications. Additionally, uncertainties such as model mismatch, exercise, and stress can affect BGC, impacting overall health. To address these challenges, this study advocates a data-driven tube-based nonlinear model predictive control (Data-Driven Tube-NMPC) approach. Firstly, a genetic algorithm-based long short-term memory (GA-LSTM) neural network is formulated to build a prediction model, where its hyperparameters are optimized via a genetic algorithm (GA). Secondly, a tube-based nonlinear model predictive control (Tube-NMPC) strategy is proposed to introduce the concept of a tube, tightening constraints to reduce model mismatch and mitigate the effects of factors such as individual differences, exercise, and stress. Additionally, the feasibility, constraint satisfaction, and stability of the algorithm are analyzed. Finally, simulation results demonstrate that, compared to conventional model predictive control (MPC), this approach more effectively controls the BGC of T1DM patients in the presence of perturbations. Note to Practitioners—Due to the model mismatch problem and factors such as individual difference, exercise and pressure, the robust control of BGC during the human blood glucose process by artificial pancreatic system (APS) will be affected to some extent. In order to achieve the robust control of BGC in a limited time, this paper designs a Tube-NMPC strategy. The proposed control scheme effectively regulates BGC in T1DM patients after food intake, ensuring both safety and timeliness. In practice, practitioners should obtain accurate data. The prediction accuracy is improved by automatically determining the hyperparameters of the LSTM by a GA. Secondly, a Tube-NMPC algorithm is employed to incorporate the notion of tubes, thereby constricting constraints to mitigate the impact of uncertainty. Subsequently, the efficacy of the proposed control strategy is substantiated by validating the life trajectory of a patient diagnosed with T1DM.