Performance Assessment of Model-Based Artificial Pancreas Control Systems

Abstract

Many artificial pancreas control systems are based on models that predict glucose concentrations. The performance of these control systems depends on the accuracy of the models and may be affected when large dynamic changes in the human body or changes in equipment performance occur and move the operating conditions away from those used in developing the models and designing the control system. A controller performance assessment (CPA) module is developed to evaluate the performance of model-based controllers and initiate controller retuning if there is significant performance deterioration. The generalized predictive control (GPC) approach that utilizes models for glucose concentration predictions is used for illustrating the performance of the CPA. The module has six indexes that capture different aspects of model and controller performance, which can be analyzed to determine the specific component of the controller that caused performance deterioration. Four different kinds of controller errors were diagnosed by indexes and used for controller retuning. Thirty subjects in the UVa/Padova metabolic simulator are used in simulations to evaluate the performance of the CPA module. The results indicate that a GPC with the proposed CPA module has a safer range of glucose concentration variation and more reasonable insulin suggestions than a GPC without controller retuning guided by the CPA module.