<p>This study focuses on developing a fully Automated Insulin Delivery System (AiDS) based on an event-triggered Model Predictive Control (ET-MPC) strategy aimed at individuals with type 1 diabetes mellitus (T1DM). The research addresses two key questions: (1) Can glycemic regulation be enhanced by integrating the recursive least squares (RLS) parameter estimator by dynamically adjusting the insulin sensitivity parameter, and (2) Can the embedded systems’ energy efficiency and operational autonomy by using an ET-MPC strategy? The ET-MPC strategy was enhanced with an RLS insulin sensitivity estimation. It was implemented within a hardware-in-the-loop (HIL) framework on a Raspberry Pi model 3B, integrated with a customized low-cost insulin pump and a battery driver. The virtual patient in the HIL emulation is the UVA/Padova simulator, which adds realistic physiological conditions to the emulation, such as highly nonlinear dynamics, measurement noise, parametric variations following the circadian cycle, and carbohydrate disturbances not announced to the controller. The prediction model was identified from 10 adults of the simulator and the states retrieved by the Kalman estimator. The study compared the performance of time-triggered (TT) and ET control strategies, both with and without the RLS insulin sensitivity estimator, over three-day HIL emulations. The integration of RLS improved glucose regulation under TT control, increasing the time within the target range from 75.97 to 91.08%. When the ET strategy is activated, it can reduce the control actions up to 60% while maintaining acceptable glycemic performance. Consequently, energy consumption was significantly reduced, with the device’s operational time extended up to threefold. However, higher thresholds in ET control compromised glucose control, highlighting the trade-off between energy savings and control performance. The RLS estimator enhances glycemic regulation, and the ET strategy achieves substantial energy savings, making the AiDS more efficient and portable. Future work should explore alternative estimation techniques for improved robustness and the real-time adaptation of the <InlineEquation ID="IEq1"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="42979_2025_3846_Article_IEq1.gif" Format="GIF" Height="10" Rendition="HTML" Resolution="72" Type="Linedraw" Width="10" /> </InlineMediaObject> <EquationSource Format="TEX">$\epsilon$</EquationSource> <EquationSource Format="MATHML"><math> <mi>ϵ</mi> </math></EquationSource> </InlineEquation> and <InlineEquation ID="IEq2"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="42979_2025_3846_Article_IEq2.gif" Format="GIF" Height="14" Rendition="HTML" Resolution="72" Type="Linedraw" Width="15" /> </InlineMediaObject> <EquationSource Format="TEX">$\lambda$</EquationSource> <EquationSource Format="MATHML"><math> <mi>λ</mi> </math></EquationSource> </InlineEquation> thresholds to optimize the trade-off between glycemic regulation and energy expenditure.</p>

Efficient AiDS Through Real-Time Insulin Sensitivity Estimation and Energy Management with Event-Triggered MPC

期刊论文
出版模式：

开放获取
发表日期：

2025年4月2日

查看全文

Jhon E. Goez-Mora,
Pablo S. Rivadeneira

摘要

This study focuses on developing a fully Automated Insulin Delivery System (AiDS) based on an event-triggered Model Predictive Control (ET-MPC) strategy aimed at individuals with type 1 diabetes mellitus (T1DM). The research addresses two key questions: (1) Can glycemic regulation be enhanced by integrating the recursive least squares (RLS) parameter estimator by dynamically adjusting the insulin sensitivity parameter, and (2) Can the embedded systems’ energy efficiency and operational autonomy by using an ET-MPC strategy? The ET-MPC strategy was enhanced with an RLS insulin sensitivity estimation. It was implemented within a hardware-in-the-loop (HIL) framework on a Raspberry Pi model 3B, integrated with a customized low-cost insulin pump and a battery driver. The virtual patient in the HIL emulation is the UVA/Padova simulator, which adds realistic physiological conditions to the emulation, such as highly nonlinear dynamics, measurement noise, parametric variations following the circadian cycle, and carbohydrate disturbances not announced to the controller. The prediction model was identified from 10 adults of the simulator and the states retrieved by the Kalman estimator. The study compared the performance of time-triggered (TT) and ET control strategies, both with and without the RLS insulin sensitivity estimator, over three-day HIL emulations. The integration of RLS improved glucose regulation under TT control, increasing the time within the target range from 75.97 to 91.08%. When the ET strategy is activated, it can reduce the control actions up to 60% while maintaining acceptable glycemic performance. Consequently, energy consumption was significantly reduced, with the device’s operational time extended up to threefold. However, higher thresholds in ET control compromised glucose control, highlighting the trade-off between energy savings and control performance. The RLS estimator enhances glycemic regulation, and the ET strategy achieves substantial energy savings, making the AiDS more efficient and portable. Future work should explore alternative estimation techniques for improved robustness and the real-time adaptation of the $\epsilon$ $ϵ$ and $\lambda$ $λ$ thresholds to optimize the trade-off between glycemic regulation and energy expenditure.