In Silico Testing of a Closed-Loop Artificial Pancreas Based on Generalized Predictive Control

Abstract

Abstract The artificial pancreas (AP) system is a promising approach to maintain blood glucose concentration (BGC) of patients with type 1 diabetes (T1D) in a euglycemic range (70–180 mg/dl). A controller based on generalized predictive control (GPC) with an adaptive reference glucose trajectory and an adaptive softening factor (GPC+AA) was proposed for AP system in our previous research. It was built and tested with the University of Virginia (UVA)/Padova type 1 diabetes mellitus simulator (T1DMS), which embodies the biophysiological parameters of in-silico population accepted by the US Food and Drug Administration. It was showed that the GPC+AA controller realized effective control of the BGC of in-silico patients with T1D. Here, the effectiveness of the GPC+AA controller was further tested with strict conditions, including a pump with noise and error, a continuous glucose monitoring sensor and a high carbohydrate intake. Test results showed that the BGC of adolescent group was not effectively controlled with the GPC+AA controller. Then, the adaptive softening factor strategy was further optimized and a GPC controller with an adaptive reference glucose trajectory and an optimized adaptive softening factor (GPC+AO) was proposed here. It was demonstrated that the optimized adaptive softening factor strategy significantly improved the performance of the GPC controller, and the average BGC percentage within the euglycemic range of 9 adults and 10 adolescents increased to 80% and 70%, respectively. Thus, our GPC+AO controller is effective for patients with T1D and can be potentially applied in the AP systems.