Study On Fault Detection And Data Recovery Method Based On Deep Learning For Continuous Glucose Monitoring

Continuous glucose monitor enables real-time blood glucose concentration measurement,providing a new solution to the increasingly severe problem of diabetes.However,decision-making related to patient safety,such as insulin dose calculation,depends on the blood glucose data provided by the continuous glucose monitor and the current continuous glucose monitor is not reliable enough due to the sensor failure.Therefore,it is especially important to improve data reliability through continuous glucose monitor signal fault detection and data recovery.Given that the generalization ability of the existing continuous glucose monitor signal fault detection and data recovery methods is insufficient,especially in complex fault scenarios,and deep learning algorithms perform well in feature extraction from complex data,which not only perform well in various image and text processing tasks,but also achieve remarkable results in physiological signal processing.Therefore,this paper establishes a network model based on deep learning to improve the performance of continuous glucose monitor signal fault detection and data recovery.In this paper,the principle of the main continuous glucose monitor signal fault detection algorithms and the mechanism of the main deep learning algorithms are studied.The traditional data-driven technique is combined with the powerful feature extraction ability of the deep learning model,which provides a high-performance solution to continuous glucose monitor failure.This paper refers to a large number of research concerned with the deep learning model designing and optimization as well as the existing continuous glucose monitor signal fault detection and data recovery algorithms to complete the architecture design and the hyperparameter optimization.By Model training and a large number of comparative experiments,three deep learning models,CNN_C5F1,GRU_H128L3 and VAE_C3F2 are proposed,which can provide more reliable fault detection and data recovery results.In this paper,the fault detection model and the data recovery model are combined to form a system with real-time processing capability of continuous glucose monitor signal.By a comprehensive analysis and comparison on fault detection performance,signal repair performance and real-time processing performance of CNN-VAE system and RNN-VAE system,RNN-VAE is determined as the final system scheme.The performance of the proposed real-time continuous glucose monitor signal fault detection and data recovery system is further verified by model testing under various fault scenarios and comparing key metrics,which shows that the generalization performance of the system outperforms traditional algorithms.