Extremum Seeking Control for Personalized Zone Adaptation in Model Predictive Control for Type 1 Diabetes

Zone model predictive control has proven to be an effective closed-loop method to regulate blood glucose for people with type 1 diabetes (T1D). In this paper, we present a universal model-free optimization scheme for adapting the zone for T1D patients individually. The adaptation is based on a clinical glycemic risk index named relative regularized glycemic penalty index (rrGPI), which is calculated from glucose measurements by a continuous glucose monitor. The scheme's objective is to minimize rrGPI by simultaneously modulating a controller's blood glucose target zone's upper bound and lower bound. The adaptation mechanism is based on extremum seeking control, in which the zone boundaries are driven by gradient estimation obtained by continuously sinusoidally modulating and demodulating the rrGPI readings. To improve the adaptation method's robustness against uncertainties, a decaying feedback gain and a vanishing dither signal are employed. in-silico trials suggested that the personalized optimized zone can be reached within a week of adaptation. Both for announced and unannounced meals, the proposed method outperforms the fixed zone [80, 140] mg/dL, which has been employed in the authors' clinical trials. It is also shown that the developed method has strong robustness against real-life uncertainties.