Blood glucose monitoring control system and blood glucose monitoring control method

Abstract

The invention relates to a blood glucose monitoring control system and a blood glucose monitoring control method, and belongs to the technical field of blood glucose monitoring instruments. The system comprises a multispectral sensing module (comprising a 480nm / 660nm dual-wavelength camera and an mmu SORS sensor), a wearable multi-modal monitoring unit (integrating a blood glucose sensor and seven physiological parameter acquisition assemblies), a blood glucose metabolism dynamics modeling module (based on a nonlinear differential equation), a dynamic blood glucose prediction engine (TCN-Transform and ST-GCN fusion architecture), an edge calculation unit (FPGA accelerated RANSAC improved algorithm), a data processing unit (data processing unit), a data processing unit (data processing unit), a data processing unit (data processing unit), a data processing unit (data processing unit), a data processing unit (data processing unit) and a data processing unit (data processing unit), a data processing unit (data processing unit) and a data processing unit (data processing unit). A closed-loop feedback actuator (Bluetooth 5.2 insulin pump regulation and control module) and a data security module (homomorphic encryption under a federated learning framework). According to the system, through multi-source data fusion and reinforcement learning optimization, the blood glucose prediction error is remarkably reduced, the emergency response time is shortened, and the accuracy and real-time performance of diabetes management are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of blood glucose monitoring device technology, specifically to a blood glucose monitoring and control system and a blood glucose monitoring and control method. Background Technology

[0002] Current blood glucose monitoring technology faces three main technical bottlenecks: 1) Traditional invasive testing methods rely on finger-prick blood sampling or venous blood drawing, which not only causes pain and discomfort to users, but frequent operation may also lead to skin damage and psychological resistance; 2) Non-invasive devices based on a single detection principle (such as infrared spectroscopy, impedance measurement, etc.) have large fluctuations in measurement results due to individual differences such as human skin characteristics and subcutaneous tissue thickness. Clinical verification shows that their error rate is generally higher than 15%; 3) Existing algorithms mostly focus on static blood glucose value prediction, ignoring the dynamic correlation between food appearance (such as color and texture) and physiological indicators such as digestion rate and insulin secretion during the eating process.

[0003] To address the aforementioned issues, it is necessary to improve existing monitoring equipment and methods so that they can integrate multiple physiological parameters such as heart rate variability and skin temperature, and combine deep learning to construct high-precision metabolic dynamics models for real-time data processing. Ultimately, this will form an intelligent closed-loop control system of "detection-analysis-feedback," creating conditions for significantly improving monitoring comfort, accuracy, and predictive capabilities. Summary of the Invention

[0004] In view of this, the purpose of this invention is to provide a blood glucose monitoring and control system and a blood glucose monitoring and control method to solve some or all of the technical problems existing in the above-mentioned existing blood glucose monitoring devices or methods.

[0005] One of the present inventions is achieved through the following technical solution:

[0006] A blood glucose monitoring and control system, comprising:

[0007] The multispectral sensing module integrates a dual-wavelength optical sensor and a polarization filter array to acquire optical features of food and skin tissue.

[0008] The flexible monitoring unit includes an electrochemical sensor and a multi-channel physiological parameter acquisition system;

[0009] The metabolic modeling module enables the creation of personalized glycemic dynamics differential equation models.

[0010] A hybrid prediction engine that uses a fusion architecture of temporal convolutional networks and graph neural networks;

[0011] Edge computing units are equipped with processors that utilize improved random sampling algorithms;

[0012] Intelligent actuator, including an adjustable insulin infusion device;

[0013] The data security module implements biometric-based encrypted access control.

[0014] Furthermore, the multispectral sensing module includes:

[0015] Deep residual networks are used for image feature extraction.

[0016] Instance segmentation algorithms are used to identify food components;

[0017] Three-dimensional reconstruction technology is used to calculate nutrient parameters.

[0018] Furthermore, the prediction engine includes:

[0019] Multi-scale temporal feature extraction layer;

[0020] Cross-modal attention fusion mechanism;

[0021] Personalized metabolic profile database.

[0022] Furthermore, the closed-loop actuator includes:

[0023] A dose control algorithm optimized by reinforcement learning;

[0024] Dynamically adjusted multi-level response thresholds;

[0025] Real-time status recognition function.

[0026] Furthermore, the security module implements:

[0027] Encrypted transmission of parameters under the federated learning framework;

[0028] Biometric-based access control;

[0029] Real-time abnormal behavior detection.

[0030] Furthermore, the personalized glycemic dynamics differential equation model is used to collect data from multiple sources and construct a nonlinear differential equation, wherein the nonlinear differential equation is:

[0031] dBG / dt=α・BG(t)-β・I(t)・BG(t)+γ・G(t)

[0032] BG(t) represents blood glucose concentration; I(t) represents insulin concentration; G(t) represents the rate of exogenous glucose input; α represents the basal metabolic rate coefficient; β is the insulin sensitivity coefficient; γ is the glucose uptake efficiency coefficient; dBG / dt is the rate of change of blood glucose, estimated by the maximum likelihood method.

[0033] Furthermore, the nonlinear differential equation is established based on the secretory dynamics of pancreatic β cells, and the hyperparameter combination is optimized using a quantum genetic algorithm.

[0034] Furthermore, it also includes a user interaction module, which comprises:

[0035] AR nutrition annotation unit, adapted to HoloLens3 display terminal, displays food nutrition information and blood glucose prediction curves in an augmented reality manner;

[0036] The haptic feedback device sends out an alert via vibration when abnormal blood sugar is detected.

[0037] The second aspect of this invention is achieved through the following technical solution:

[0038] A blood glucose monitoring and control method of a blood glucose monitoring and control system as described above includes the following steps:

[0039] S1. Collect multi-source physiological data from 0.5Hz to 10Hz and perform dynamic threshold filtering;

[0040] S2. Construct a metabolic kinetic model and optimize hyperparameters;

[0041] S3. Achieve three-stage fusion through ST-GCN: early feature splicing, mid-term attention weighting, and late-term Choquet integral decision-making;

[0042] S4, AR visualization and tactile early warning feedback.

[0043] Furthermore, the three-stage fusion includes:

[0044] Early fusion stage: Directly splicing raw spectral features with time-series physiological parameter data;

[0045] Mid-term fusion stage: A cross-modal attention mechanism is used to calculate the weights of each parameter;

[0046] Late fusion stage: Choquet integral algorithm is applied for weighted decision-making.

[0047] Furthermore, it also includes a user interaction module, which includes:

[0048] AR nutrition annotation unit, adapted to HoloLens3 display terminal, displays food nutrition information and blood glucose prediction curves in an augmented reality manner;

[0049] The haptic feedback device sends out an alert via vibration when abnormal blood sugar is detected.

[0050] The beneficial effects of this invention are as follows:

[0051] 1. The multispectral sensing module can accurately identify food types and characteristics, providing an accurate food information basis for blood glucose prediction and control.

[0052] 2. The wearable multimodal monitoring unit enables the simultaneous acquisition of multiple physiological parameters, providing rich data support for blood glucose metabolism analysis.

[0053] 3. The nonlinear differential equations constructed by the glucose metabolism kinetics modeling module help to gain a deeper understanding of the laws governing glucose metabolism.

[0054] 4. The dynamic blood glucose prediction engine adopts an advanced fusion architecture, which has high prediction accuracy and can accurately generate blood glucose prediction curves for a period of time in the future, providing a reliable basis for insulin dosage adjustment.

[0055] 5. Edge computing units improve data processing efficiency and ensure system real-time performance.

[0056] 6. The closed-loop feedback actuator can dynamically adjust the insulin dose based on the prediction results, respond promptly to changes in blood glucose, and reduce the risk of complications.

[0057] 7. The data security module ensures the security of data transmission and access, and protects patient privacy.

[0058] In summary, this invention innovatively employs multispectral imaging technology to capture the optical characteristics of the skin's surface and deep tissues, simultaneously integrating multiple physiological parameters such as heart rate variability and skin temperature, and constructing a high-precision metabolic dynamics model through deep learning. This solution, coupled with edge computing devices, enables real-time data processing, ultimately forming an intelligent closed-loop control system of "detection-analysis-feedback," which can significantly improve monitoring comfort, accuracy, and predictive capabilities. Attached Figure Description

[0059] Figure 1 This is a framework diagram of a blood glucose monitoring and control system according to the present invention;

[0060] Figure 2 This is a flowchart of a blood glucose monitoring and control method according to the present invention. Detailed Implementation

[0061] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0062] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0063] like Figure 1 As shown, one embodiment of the present invention provides a blood glucose monitoring and control system. This system is a closed-loop blood glucose management system integrating real-time monitoring, intelligent prediction, automatic adjustment, and safety protection. It primarily provides comprehensive, personalized, and precise blood glucose management services for diabetic patients, effectively improving their quality of life and reducing the risk of diabetic complications. Its core components and functions are as follows:

[0064] This embodiment's multispectral sensing module integrates a dual-wavelength optical sensor and a polarization filter array. The dual-wavelength optical sensor can sensitively capture optical signals in different wavelength bands. Since different substances exhibit unique optical properties at specific wavelengths, this can be used to distinguish food components and identify the physiological state of skin tissue. The polarization filter array plays a crucial role in filtering ambient light interference. Under complex ambient light conditions, such as reflections from indoor lights or direct sunlight, the polarization filter array, through specific optical principles, allows only light with a specific polarization direction to pass through, thereby greatly enhancing the specificity of the acquired signal and making the obtained data purer and more accurate.

[0065] Two key types of information are acquired simultaneously using optical methods. First, acquiring optical features of food allows for precise identification of its components, such as the proportions of carbohydrates, proteins, and fats, as well as its structural characteristics. This is crucial for subsequent calculations of exogenous glucose input rates. Second, acquiring optical features of skin tissue can indirectly reflect blood composition and tissue metabolic status. For example, changes in certain skin tissue optical properties may indicate changes in blood glucose concentration or abnormalities in tissue metabolic processes.

[0066] Furthermore, this embodiment utilizes a deep residual network to process the acquired food / skin optical images. Traditional deep neural networks are prone to the vanishing gradient problem during training, making it difficult for the model to learn deep features. However, deep residual networks, by introducing shortcut connections, allow the network to directly transmit information from shallow layers to deeper layers during learning, avoiding excessive gradient decay during backpropagation. This enables the effective extraction of high-dimensional features from food and skin tissue images, such as texture details and spectral intensity distribution, providing richer and more accurate information for subsequent analysis.

[0067] In this embodiment, an instance segmentation algorithm is also employed: this algorithm, used in the multispectral sensing module, is used to accurately identify different components in food. It can segment the various components in a food image, clearly delineating the regions containing different components such as carbohydrates, proteins, and fats. Taking a mixed food containing rice, vegetables, and meat as an example, the instance segmentation algorithm can clearly identify each component and determine their specific location and range in the image, laying the foundation for subsequent accurate calculation of the content and nutritional parameters of each component.

[0068] In this embodiment, information acquired through multi-angle optical scanning is used to construct a three-dimensional model of the food. This model allows for the accurate calculation of parameters such as the food's volume and density. Combined with the food component information identified by the previous instance segmentation algorithm, key nutritional parameters such as calories and glucose content are further derived. For example, for an irregularly shaped fruit, its volume can be obtained through 3D reconstruction, and combined with the proportion of carbohydrates in its composition, the amount of glucose that the fruit may provide to the human body can be accurately calculated, thus providing accurate data support for the exogenous glucose input rate G(t) in the metabolic modeling module.

[0069] The hardware configuration of the flexible monitoring unit in this embodiment includes an electrochemical sensor and a multi-channel physiological parameter acquisition system. The electrochemical sensor has the ability to directly detect biochemical indicators such as blood glucose concentration. Through a specific chemical reaction, it converts substances such as glucose in blood or tissue fluid into electrical signals, thereby achieving real-time monitoring of core metabolic indicators such as blood glucose concentration (BG(t)) and insulin concentration (I(t)). The multi-channel physiological parameter acquisition system is more versatile, capable of acquiring various parameters such as heart rate, skin temperature, and activity level, reflecting the physiological state of the human body from multiple dimensions.

[0070] The electrochemical sensors within the system can collect key metabolic indicators such as blood glucose concentration (BG(t)) and insulin concentration (I(t)) in blood / tissue fluid in real time and with high precision, providing the most direct data for understanding the body's current metabolic state. For example, real-time monitored blood glucose concentration data can intuitively reflect whether the patient's blood glucose level is within the normal range, while insulin concentration data can reflect the body's insulin secretion status, which is of great significance for assessing insulin sensitivity.

[0071] The multi-channel acquisition system simultaneously acquires physiological parameters related to exercise status and basal metabolism. Heart rate variability reflects the body's stress state; it changes accordingly when the body is under stress or exercising. Skin temperature, to some extent, reflects the body's metabolic activity; skin temperature may be slightly elevated when the basal metabolic rate is high. Activity level directly reflects the body's exercise status; different activity levels have varying degrees of impact on glucose consumption and metabolism. These multi-source physiological parameters provide rich input information for metabolic models, helping to construct more comprehensive and accurate human metabolic models.

[0072] The entire monitoring unit features a flexible design that perfectly conforms to the curvature of human skin, significantly improving comfort during long-term wear. Compared to traditional rigid monitoring devices, the flexible design reduces motion interference caused by poor skin contact, ensuring continuous data acquisition. Its acquisition frequency covers 0.5Hz-10Hz, a carefully considered range that effectively balances real-time requirements with energy consumption, ensuring stable operation over extended periods and providing reliable data to the system.

[0073] This embodiment establishes a personalized glycemic dynamics differential equation model through a metabolic modeling module, quantifying the dynamic relationship between blood glucose changes and insulin and exogenous glucose. This model is the core mathematical model of the entire blood glucose monitoring and control system. It can accurately simulate the dynamic changes of blood glucose in the body based on individual physiological characteristics and real-time data, providing a solid theoretical foundation for subsequent blood glucose prediction and regulation.

[0074] The core equation is: dBG / dt = α・BG(t) - β・I(t)・BG(t) + γ・G(t). Here, dBG / dt represents the rate of change in blood glucose, directly reflecting the rise and fall of blood glucose per unit time, and is a key indicator for assessing blood glucose stability. BG(t) represents the blood glucose concentration at time t; it is the core variable in the model, reflecting the immediate blood glucose level at a given moment. I(t) is the insulin concentration at time t; insulin plays a crucial role in blood glucose regulation, and its concentration changes directly affect glucose metabolism. G(t) represents the rate of exogenous glucose input, mainly derived from the digestion and absorption of food; this parameter is closely related to dietary intake. α is the basal metabolic rate coefficient, reflecting the body's natural rate of glucose consumption in the absence of exogenous glucose input and insulin action, demonstrating the influence of basal metabolism on blood glucose. β is the insulin sensitivity coefficient; significant differences in insulin sensitivity exist between individuals. The larger the coefficient, the more sensitive the body is to insulin, and the higher the efficiency of insulin in lowering blood glucose; conversely, the smaller the coefficient, the lower the efficiency. γ is the glucose absorption efficiency coefficient, which reflects the proportion of glucose in food that is absorbed into the bloodstream by the human body. The glucose absorption efficiency varies among different individuals and different types of food.

[0075] In particular, this model is based on the secretory kinetics of pancreatic β cells, which are responsible for secreting insulin, and their secretory kinetics are a key physiological mechanism for blood glucose regulation. By simulating the regularity of insulin secretion by pancreatic β cells, the model more closely reflects the actual physiological mechanisms of the human body and can more accurately reflect the dynamic relationship between blood glucose and insulin.

[0076] During optimization, a quantum genetic algorithm is used to optimize the hyperparameter combination, specifically optimizing the initial ranges of hyperparameters such as α, β, and γ. Traditional optimization algorithms are prone to getting trapped in local optima when dealing with such complex nonlinear models, causing the model to fail to accurately adapt to individual differences. The quantum genetic algorithm, by utilizing the superposition states of qubits and quantum gate operations, increases the diversity of the search space, enabling it to more effectively find optimal solutions in complex parameter spaces. This significantly improves the model's adaptability to individual differences, allowing it to more accurately reflect the unique glycemic metabolism characteristics of each patient. For parameter estimation, dBG / dt is estimated using the maximum likelihood method. Based on probabilistic statistical principles, the maximum likelihood method can find the model parameter values ​​most likely to produce the observed data, thereby improving the accuracy of glycemic rate calculation.

[0077] The hybrid prediction engine architecture design in this embodiment includes a fusion architecture of Temporal Convolutional Network (TCN) and Graph Neural Network (GNN), namely, Spatiotemporal Graph Convolutional Network (ST-GCN). Temporal Convolutional Networks excel at processing time-series data, capturing the characteristics and patterns of data over time, such as the changing trends of indicators like blood glucose and insulin. Graph Neural Networks, on the other hand, perform exceptionally well in processing data with complex relationships; they can uncover the interrelationships between different parameters, such as the potential connections between diet, exercise, and physiological parameters. The fusion of these two technologies balances the extraction of both temporal and relational features, enabling the prediction model to more comprehensively analyze multi-source data.

[0078] The hybrid prediction engine in this embodiment is based on multi-source data, including spectral features, physiological parameters, and historical blood glucose / insulin data, and can predict blood glucose trends over a future period. Accurate blood glucose prediction is crucial for taking early intervention measures and preventing abnormal blood glucose fluctuations. It provides a reliable basis for insulin infusion decisions and helps patients better manage their blood glucose levels.

[0079] The edge computing unit in this embodiment is equipped with an improved random sampling algorithm processor. This processor is specifically optimized for multi-source heterogeneous data, such as optical images, physiological signals, and model parameters. In traditional data processing, conventional algorithms may perform full computation on massive amounts of multi-source data, resulting in excessive computational load and low efficiency. The improved random sampling algorithm processor, through a clever sampling strategy, reduces redundant computation while ensuring data representativeness, thus greatly improving data processing efficiency.

[0080] Edge computing units enable real-time data processing and model inference, achieving low-latency responses. In blood glucose monitoring and control scenarios, real-time data processing is crucial. For example, when a patient's blood glucose levels show a rapid rise or fall, the system needs to react quickly and adjust the insulin infusion dose promptly. Relying on cloud-based data transmission and computation may lead to decision-making delays due to network latency and other issues, hindering timely and effective blood glucose control. Edge computing units process data locally, avoiding cloud transmission delays and ensuring the system can respond to blood glucose changes in an extremely short time (e.g., less than 1 second). Simultaneously, improved algorithms reduce computational power consumption, extending the device's battery life. For portable blood glucose monitoring devices that require prolonged wear, this feature significantly improves the user experience and ensures continuous and stable operation.

[0081] The intelligent actuator in this embodiment includes at least core components, such as an insulin infusion device and a micro-pump. These devices possess precise dose control capabilities, accurately adjusting the insulin infusion rate according to system commands, and are key actuators for achieving closed-loop blood glucose control.

[0082] In use, the system automatically adjusts the insulin infusion rate based on the blood glucose trend output by the prediction engine, achieving closed-loop blood glucose control. When the prediction engine predicts that blood glucose will rise above the target range, the intelligent actuator automatically increases the insulin infusion rate to promote glucose metabolism and lower blood glucose levels; conversely, when it predicts that blood glucose may drop too low, it reduces the insulin infusion rate to prevent hypoglycemia. Through this automatic adjustment mechanism, blood glucose is consistently maintained within the target range, achieving stable blood glucose control.

[0083] Of course, reinforcement learning can optimize the dosage control algorithm. This algorithm continuously learns the user's response to insulin, such as the magnitude and rate of blood glucose decrease after a certain dose of insulin infusion, and dynamically optimizes the dosage calculation strategy. In practical applications, different patients have different sensitivities and responses to insulin, and the insulin requirements of the same patient also change under different physiological states (such as after exercise or when ill). Reinforcement learning algorithms can automatically adjust the dosage calculation model to adapt to these changes during continuous interaction with the patient, ensuring the accuracy and safety of insulin infusion dosage and minimizing the risk of hypoglycemia.

[0084] During use, different blood glucose thresholds can be set based on the user's state, such as sleep, exercise, or eating. For example, during sleep, the body's basal metabolic rate decreases, and blood glucose consumption is relatively reduced. In this case, the hypoglycemic threshold can be appropriately widened to avoid undetected hypoglycemia caused by excessive insulin infusion during sleep. During exercise, the body's consumption of blood glucose increases, and blood glucose fluctuations are larger. Therefore, the hypoglycemic threshold can be appropriately raised, and the hyperglycemic threshold may also need to be adjusted accordingly to adapt to the characteristics of blood glucose changes during exercise. By dynamically adjusting multi-level response thresholds, the system can perform more reasonable and precise blood glucose control based on the user's actual state.

[0085] This embodiment utilizes data such as heart rate and activity level acquired by a multi-channel physiological parameter acquisition system, combined with machine learning algorithms, to identify the user's current activity state, such as resting, exercising, or eating. Different activity states have different effects on blood glucose; for example, blood glucose decreases during exercise and increases after eating. After identifying the user's current state, the system can match corresponding control strategies. When the user is exercising, the insulin infusion volume is appropriately reduced to prevent hypoglycemia; after eating, the insulin infusion plan is adjusted in advance according to the type and amount of food consumed to better control the postprandial blood glucose rise.

[0086] The data security module in this embodiment is primarily used to protect user privacy data, including blood glucose levels, biometrics, and metabolic models, during the collection, transmission, and storage processes. In today's digital age, the security of patient health data is paramount; leakage could severely damage patient privacy and rights.

[0087] In this embodiment, to enhance security, biometric identification technologies such as fingerprints and iris scans can be used to bind device access permissions to the user's biometric features. Only authorized users, i.e., users whose biometric features are successfully matched, can operate the device and view and modify related data. This method greatly improves device security, prevents unauthorized personnel from operating the device, and avoids the risks of data leakage and malicious operation due to device loss or theft, such as dangerous acts like unauthorized personnel tampering with insulin infusion dosages.

[0088] For abnormal behaviors, such as a sudden and significant increase in data transmission volume, abnormal transmission frequency, or unauthorized operation commands from the device, such as a sudden large infusion of insulin, which does not conform to normal usage logic, the system will immediately trigger a security alert. Alert methods may include sending notifications to the user's associated devices and displaying alarm information on the system management interface, so that timely measures can be taken to ensure system security and user health.

[0089] The core function of the user interaction module in this embodiment is to enable intuitive interaction between the user and the system, enhance the user's perception and participation in blood glucose status management, and strengthen the user's initiative and compliance in the blood glucose management process. Specifically, it includes:

[0090] AR Nutrition Labeling Unit: Compatible with HoloLens 3 display terminals, this unit uses augmented reality technology to overlay food nutritional information and a 2-hour blood glucose prediction curve onto the real-world scene. When users eat, they can directly see the nutritional information of the food in front of them, such as carbohydrate content and calories, through the HoloLens 3 device. They can also intuitively see the predicted blood glucose levels 1 hour and 2 hours after consuming the food. For example, if a user sees a serving of rice, the AR nutrition labeling unit will display the carbohydrate content of that serving and predict the blood glucose level that may rise to XX mmol / L 1 hour after consumption, as well as the possible blood glucose level 2 hours later. This provides users with an intuitive and real-time reference for dietary decisions, helping them better understand the impact of diet on blood glucose and thus make more rational dietary choices.

[0091] In addition, this embodiment also includes a tactile feedback device. When the system predicts an abnormal blood sugar level, such as impending hypoglycemia or hyperglycemia, the tactile feedback device issues tiered warnings through vibrations of different frequencies. For emergencies, such as a rapid drop in blood sugar to the danger zone of hypoglycemia, a high-frequency vibration will alert the user, drawing their attention. For less urgent but still concerning abnormal blood sugar levels, such as a slow rise in blood sugar approaching the hyperglycemia threshold, a low-frequency vibration may be used. This tactile feedback method is particularly suitable for visually / hearingly impaired users, and in scenarios where visual or auditory cues are inconvenient to view, such as during exercise or driving. Tactile feedback can promptly remind users to monitor their blood sugar levels and take appropriate measures, such as supplementing carbohydrates to combat hypoglycemia, adjusting exercise intensity, or preparing medication to combat hyperglycemia.

[0092] like Figure 2 As shown, the method for blood glucose monitoring and control using the above system includes the following steps:

[0093] This method describes the operating logic of the aforementioned system, achieving closed-loop control through four steps, as follows:

[0094] S1. Acquire multi-source physiological data from 0.5Hz to 10Hz, and perform dynamic threshold filtering.

[0095] The system uses a multispectral sensing module and a flexible monitoring unit to collect spectral data, blood glucose, insulin, heart rate and other multi-source data at a frequency of 0.5Hz-10Hz. It also uses a dynamic threshold (which adjusts the noise threshold in real time according to the user's status) to filter out abnormal values ​​(such as instantaneous abnormal signals caused by motion interference) to ensure data quality.

[0096] S2. Construct a metabolic kinetic model and optimize hyperparameters.

[0097] Based on the cleaned data from S1, personalized glycemic dynamics differential equations are substituted into the model, and hyperparameters such as α, β, and γ are optimized through quantum genetic algorithms to make the model more closely match the user's metabolic characteristics (e.g., for users with low insulin sensitivity, the β value is smaller).

[0098] S3. Achieve three-stage fusion (early, mid, and late stages) through ST-GCN.

[0099] To improve prediction accuracy, a spatiotemporal graph convolutional network (ST-GCN) is used to fuse features from multiple data sources. The three stages are as follows:

[0100] Early fusion stage: Directly splicing original spectral features (such as food composition) with time-series data of physiological parameters (such as blood glucose change curves) to preserve the original information;

[0101] Mid-term integration phase: Employ cross-modal attention mechanisms to calculate the weights of each parameter (e.g., the weight of food spectral features increases during the postprandial phase, and the weight of basal metabolic parameters increases at night).

[0102] Late fusion stage: The Choquet integral algorithm (a weighted decision-making method for handling fuzzy information) is applied to make a comprehensive decision on the intermediate fusion results and output the final blood glucose prediction value.

[0103] S4.AR Visualization and Tactile Warning Feedback

[0104] The user interaction module enables a closed-loop feedback mechanism—the AR unit visualizes nutritional information and blood glucose prediction curves, helping users intuitively understand the impact of diet on blood glucose; the haptic device vibrates to warn of abnormal blood glucose levels, reminding users to intervene in a timely manner (such as supplementing carbohydrates or pausing exercise).

[0105] This embodiment achieves accurate blood glucose monitoring, intelligent prediction, and adaptive regulation through a closed-loop architecture of "perception-modeling-prediction-execution-feedback" combined with technologies such as multimodal data fusion, personalized modeling, edge computing, and reinforcement learning. At the same time, data security and user experience are guaranteed through security and interaction modules, making it particularly suitable for daily blood glucose management for diabetic patients.

[0106] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A blood glucose monitoring and control system, characterized in that, include: The multispectral sensing module integrates a dual-wavelength optical sensor and a polarization filter array to acquire optical features of food and skin tissue. The flexible monitoring unit includes an electrochemical sensor and a multi-channel physiological parameter acquisition system; The metabolic modeling module enables the creation of personalized glycemic dynamics differential equation models. A hybrid prediction engine that uses a fusion architecture of temporal convolutional networks and graph neural networks; Edge computing units are equipped with processors that utilize improved random sampling algorithms; Intelligent actuator, including an adjustable insulin infusion device; The data security module implements biometric-based encrypted access control.

2. The blood glucose monitoring and control system according to claim 1, characterized in that, The multispectral sensing module includes: Deep residual networks are used for image feature extraction. Instance segmentation algorithms are used to identify food components; Three-dimensional reconstruction technology is used to calculate nutrient parameters.

3. The blood glucose monitoring and control system according to claim 1, characterized in that, The prediction engine includes: Multi-scale temporal feature extraction layer; Cross-modal attention fusion mechanism; Personalized metabolic profile database.

4. A blood glucose monitoring and control system according to claim 1, characterized in that, The closed-loop actuator includes: A dose control algorithm optimized by reinforcement learning; Dynamically adjusted multi-level response thresholds; Real-time status recognition function.

5. A blood glucose monitoring and control system according to claim 1, characterized in that, The security module implements: Encrypted transmission of parameters under the federated learning framework; Biometric-based access control; Real-time abnormal behavior detection.

6. A blood glucose monitoring and control system according to claim 1, characterized in that, The personalized glycemic dynamics differential equation model is used to collect data from multiple sources and construct a nonlinear differential equation, which is as follows: dBG / dt=α・BG(t)-β・I(t)・BG(t)+γ・G(t) BG(t) represents blood glucose concentration; I(t) represents insulin concentration; G(t) represents the rate of exogenous glucose input; α represents the basal metabolic rate coefficient; β is the insulin sensitivity coefficient; γ is the glucose uptake efficiency coefficient; dBG / dt is the rate of change of blood glucose, estimated by the maximum likelihood method.

7. A blood glucose monitoring and control system according to claim 6, characterized in that, The nonlinear differential equation is based on the secretory dynamics of pancreatic β cells, and the hyperparameter combination is optimized using a quantum genetic algorithm.

8. A blood glucose monitoring and control system according to claim 1, characterized in that, It also includes a user interaction module, which comprises: AR nutrition annotation unit, adapted to HoloLens3 display terminal, displays food nutrition information and blood glucose prediction curves in an augmented reality manner; The haptic feedback device sends out an alert via vibration when abnormal blood sugar is detected.

9. A control method for a blood glucose monitoring and control system as described in any one of claims 1-8, characterized in that, include: S1. Collect multi-source physiological data from 0.5Hz to 10Hz and perform dynamic threshold filtering; S2. Construct a metabolic kinetic model and optimize hyperparameters; S3. Achieve three-stage fusion through ST-GCN: early feature splicing, mid-term attention weighting, and late-term Choquet integral decision-making; S4, AR visualization and tactile early warning feedback.

10. The control method according to claim 9, wherein the three-stage fusion includes: Early fusion stage: Directly splicing raw spectral features with time-series physiological parameter data; Mid-term fusion stage: A cross-modal attention mechanism is used to calculate the weights of each parameter; Late fusion stage: Choquet integral algorithm is applied for weighted decision-making.

Images