Fault Detection Method Based On Multivariate Statistical Analysis For Continuous Glucose Monitoring

Diabetes,as one of the world's most serious public health problems,has seriously endangered people's health.Type I Diabetes Mellitus can't regulate glucose by themselves.The e external injection of insulin is needed to maintain the health of patients.The control for insulin injection depends on the reliable glucose data provided by continuous glucose monitor(CGM).However,glucose data lacks accuracy due to the faults of CGM.Fault detection for CGM is critical to the treatment of diabetes.The fluctuations of glucose reduces the performance of fault detection.How to avoid the influence of fluctuation on fault detection is the main research content of this paper.Specific research content and innovation can be summarized as follows:1.Aiming at the effect of glucose fluctuation on fault detection,an iterative PCA-based classification algorithm is proposed.The glucose data are separated according to the different characteristics.The smooth and linear correlation between the data and the fluctuation of glucose are eliminated.In addition,A classification-based fault detection model is built to detect faults on line.2.In order to avoid the omission of fault signals caused by misclassification of glucose data,this paper analyzes the relationship between the blood glucose levels based on the classified blood glucose data,and proposes an on-line blood glucose signal discrimination algorithm based on FDA classifier.And the Bayesian decision fusion is used to fuse the outputs of various blood glucose monitoring models to obtain a more accurate and reliable blood glucose meter fault detection result.3.Aiming at the problem that the single-object fault detection model is too costly to be modeled and has severe application conditions,a fault detection algorithm based on the general model is proposed.The general model contains a wealth of blood glucose fluctuations,can be directly applied to different monitoring objects.