An Effort towards Offset-free Model Predictive Control of Artificial Pancreas Systems

Model predictive control (MPC) is one of the most commonly adopted algorithms for artificial pancreas (AP) systems. One of the unsolved issues is to achieve offset-free tracking during fasting periods given the mismatch in basal rate profiles. In this work, we introduce an MPC based on the extended state observer (ESO) to enable offset-free tracking behavior of AP. The proposed controller builds on a classical MPC structure but adds an ESO for total disturbance rejection. Specifically, ESO is added to estimate the uncertainties in the provided model and the mismatch of the insulin basal rate for the patient. Then we use the estimation to adaptively compensate for insulin injection and set dynamic reference values for the predictive model to isolate its effect on the control variables. This adaptive law has more flexibility to deal with changes in blood glucose in time. The performance of the proposed controller is evaluated through the 10-adult cohort of the US Food and Drug Administration (FDA) accepted Universities of Virginia (UVA)/Padova T1DM simulator. Compared with the classical MPC, the proposed controller achieves improved performance against basal rate mismatches.