Blood glucose dynamic prediction method and system fusing physiological data and behavior data

Abstract

This invention relates to the field of medical and health technology, and discloses a method and system for dynamic blood glucose prediction that integrates physiological and behavioral data. The method includes: collecting a patient's raw dataset using a wearable device; acquiring blood glucose value sequences using a blood glucose monitoring device; preprocessing the raw dataset; performing feature extraction operations on both the preprocessed data and the blood glucose values ​​in the blood glucose value sequences to obtain a raw high-dimensional feature dataset; evaluating the correlation of blood glucose values ​​using a time-series analysis model to obtain a weighted set of blood glucose value influences; weighting and fusing the multimodal feature vector set and the weighted set of blood glucose value influences to obtain a weighted vector set; dividing the weighted vector set into a training set and a validation set; validating the preliminary blood glucose prediction model; optimizing the preliminary blood glucose prediction model; and using the optimized blood glucose prediction model to predict blood glucose, thereby achieving dynamic blood glucose prediction. This invention addresses the problems of considering only one feature factor and low prediction accuracy in blood glucose prediction.

Description

Technical Field

[0001] This invention relates to the field of medical and health technology, and in particular to a method, system, electronic device, and computer-readable storage medium for dynamic blood glucose prediction that integrates physiological and behavioral data. Background Technology

[0002] With the rapid development of the healthcare industry, dynamic blood glucose prediction faces new challenges. This paper addresses the issues of limited predictive accuracy and reliance on a single factor by comprehensively considering various characteristics, including physiological and behavioral features.

[0003] Currently, blood glucose prediction is mainly achieved using limited physiological characteristics. Although traditional blood glucose prediction methods can obtain predicted blood glucose values, they suffer from problems such as considering only a single characteristic factor and low prediction accuracy. Therefore, optimizing dynamic blood glucose prediction methods is of great significance for improving prediction accuracy. Summary of the Invention

[0004] This invention provides a method for dynamic blood glucose prediction that integrates physiological and behavioral data, and a computer-readable storage medium. Its main purpose is to solve the problems of single characteristic factors and low prediction accuracy in blood glucose prediction.

[0005] To achieve the above objectives, the present invention provides a method for dynamic blood glucose prediction that integrates physiological and behavioral data, comprising:

[0006] The patient's raw dataset is collected using a pre-confirmed wearable device and a preset collection time interval. The blood glucose value sequence is obtained using a pre-confirmed blood glucose monitoring device and a collection time interval. The raw dataset includes multiple raw data, which includes physiological and behavioral data. The blood glucose value sequence includes multiple blood glucose values.

[0007] The original dataset is preprocessed to obtain a preprocessed dataset, which includes multiple preprocessed data.

[0008] Feature extraction operations were performed on both the preprocessed data in the preprocessed dataset and the blood glucose values ​​in the blood glucose value sequence to obtain the original high-dimensional feature dataset.

[0009] The original high-dimensional feature dataset is fused to obtain a multimodal feature vector set;

[0010] A pre-built time series analysis model is used to evaluate the correlation between blood glucose values ​​and multimodal feature vector sets, thereby obtaining feature importance evaluation values. Based on the feature importance evaluation values, a set of weights for the influence of blood glucose values ​​is obtained.

[0011] The multimodal feature vector set and the weight set of blood glucose value influence are weighted and fused to obtain a weighted vector set;

[0012] The weighted vector set is divided into a training set and a validation set. The training set is modeled using a pre-built time series modeling method to obtain a preliminary blood glucose prediction model. The training set accounts for 70% of the weighted vector set, and the validation set accounts for 30% of the weighted vector set. The time corresponding to the validation set is later than that of the training set.

[0013] The preliminary blood glucose prediction model is validated using a validation set to obtain validation evaluation results. Based on the validation evaluation results, the preliminary blood glucose prediction model is optimized to obtain an optimized blood glucose prediction model. The optimized blood glucose prediction model is used to predict blood glucose levels in patients to obtain predicted blood glucose values. Blood glucose risk warning reports are obtained using the predicted blood glucose values ​​of patients and preset blood glucose thresholds, thereby achieving dynamic prediction of blood glucose.

[0014] Optionally, the step of collecting the patient's raw dataset using a pre-verified wearable device and a preset collection time interval, and obtaining a blood glucose value sequence using a pre-verified blood glucose monitoring device and the collection time interval, includes:

[0015] The patient's raw dataset is collected using wearable devices and at specific time intervals. The raw dataset includes multiple raw data sets, which include physiological and behavioral data.

[0016] Blood glucose values ​​of patients are obtained using blood glucose monitoring equipment and at specific time intervals. The obtained blood glucose values ​​are then arranged in chronological order to obtain a blood glucose value sequence.

[0017] Optionally, the preprocessing of the original dataset to obtain a preprocessed dataset includes:

[0018] Perform missing data detection on the original dataset and obtain a missing data detection report;

[0019] Missing data is identified using the missing data detection report. Missing data includes single-point random missing data or consecutive missing data segments.

[0020] Complete the missing data by filling in the single point of random missing data in the missing data to obtain the completed dataset;

[0021] Identify the data missing time of consecutive missing data segments in the missing data. If the data missing time is confirmed to be less than a preset missing time threshold, then reconstruct the consecutive missing data segments in the missing data to obtain the completed data segments.

[0022] The completed dataset, the completed data segment, and the original dataset are integrated to obtain the repaired dataset;

[0023] The repair dataset is normalized using a preset normalization method to obtain a normalized dataset;

[0024] The normalized dataset and blood glucose value sequence are time-calibrated to obtain a preprocessed dataset.

[0025] Optionally, feature extraction is performed on both the preprocessed data in the preprocessed dataset and the blood glucose values ​​in the blood glucose value sequence to obtain the original high-dimensional feature dataset, including:

[0026] Extract the preprocessed physiological dataset and the preprocessed behavioral dataset from the preprocessed dataset;

[0027] Temporal feature extraction is performed on the preprocessed physiological dataset to obtain a set of physiological feature vectors;

[0028] The preprocessed behavioral dataset is segmented and aggregated to approximate the data, resulting in a set of behavioral feature vectors.

[0029] Feature extraction is performed on the blood glucose value sequence to obtain a blood glucose feature vector set;

[0030] The physiological feature vector set, behavioral feature vector set, and blood glucose feature vector set are correlated along the time dimension to obtain the original high-dimensional feature dataset.

[0031] Optionally, the fusion of the original high-dimensional feature dataset to obtain a multimodal feature vector set includes:

[0032] Multimodal feature extraction was performed on the original high-dimensional feature dataset to obtain blood glucose modality group, physiological modality group and behavioral modality group;

[0033] A fusion feature attention matrix is ​​constructed using the blood glucose modality group as the reference query vector, the physiological modality group as the key vector, and the behavioral modality group as the value vector, respectively.

[0034] Feature decoupling is performed on the fused feature attention matrix to obtain behavioral modality feature submatrices and physiological modality feature submatrices;

[0035] The behavioral modality feature submatrix and the physiological modality feature submatrix are dimensionality reduced respectively to obtain the behavioral modality feature vector set and the physiological modality feature vector set;

[0036] By summing the behavioral modality feature vector set and the physiological modality feature vector set, a multimodal feature vector set is obtained.

[0037] Optionally, the step of using a pre-built time-series analysis model to assess the correlation between blood glucose levels and multimodal feature vector sets, obtaining feature importance assessment values, and obtaining a set of weights for the influence of blood glucose levels based on the feature importance assessment values ​​includes:

[0038] The behavioral modality feature vector set and the physiological modality feature vector set in the multimodal feature vector set are time-aligned with the blood glucose value sequence to obtain the aligned feature vector set;

[0039] A time series analysis model was trained using aligned feature vector sets to obtain a blood glucose assessment model;

[0040] The importance assessment values ​​of behavioral characteristics and physiological characteristics were obtained using a blood glucose assessment model.

[0041] A weight set for the influence of behavioral blood glucose values ​​is generated based on the numerical assessment of the importance of behavioral characteristics, and a weight set for the influence of physiological blood glucose values ​​is generated based on the numerical assessment of the importance of physiological characteristics.

[0042] The weight set of behavioral blood glucose value influence is merged with the weight set of physiological blood glucose value influence to obtain the weight set of blood glucose value influence.

[0043] Optionally, the step of dividing the weighted vector set into a training set and a validation set, and using a pre-built time series modeling method to model the training set to obtain a preliminary blood glucose prediction model includes:

[0044] Arrange the weighted vector set in chronological order to obtain the weighted vector sequence;

[0045] Based on the proportion of the training set in the weighted vector sequence and the proportion of the validation set in the weighted vector sequence, the split point between the training set and the validation set is determined. The weighted vectors in the weighted vector sequence before the split point are used as the training set, and the weighted vectors after the split point are used as the validation set.

[0046] The multiple weighted vectors in the training set are normalized to obtain a normalized training set, which includes multiple normalized weighted vectors.

[0047] The normalized weighted vectors in the normalized training set are divided by time step to obtain multiple time series samples. Each time series sample includes multiple normalized weighted vectors and multiple blood glucose values, and the normalized weighted vectors correspond one-to-one with the blood glucose values.

[0048] A preliminary blood glucose prediction model was obtained by using time series modeling methods and multiple time series samples.

[0049] Optionally, the validation of the preliminary blood glucose prediction model using a validation set to obtain validation evaluation results includes:

[0050] Multiple weighted vectors from the validation set are input into the preliminary blood glucose prediction model to obtain a set of predicted blood glucose values.

[0051] The average prediction error is calculated using the predicted blood glucose values ​​from the predicted blood glucose value set, the actual blood glucose values ​​corresponding to the weighted vector, and a pre-constructed average absolute percentage error calculation formula. The average absolute percentage error calculation formula is as follows:

[0052] ,

[0053] in, Indicates the average prediction error. This represents the number of weighted vectors in the validation set. Represents the first in the verification set The predicted blood glucose value corresponding to each weighted vector. Represents the first in the verification set The actual blood glucose values ​​corresponding to each weighted vector;

[0054] The verification and evaluation results are obtained by using the average prediction error and the preset error threshold. If the average prediction error is less than the error threshold, the verification and evaluation results are deemed qualified; if the average prediction error is greater than or equal to the error threshold, the verification and evaluation results are deemed unqualified.

[0055] Optionally, the step of optimizing the preliminary blood glucose prediction model based on the verification and evaluation results to obtain an optimized blood glucose prediction model, using the optimized blood glucose prediction model to predict the patient's blood glucose level, obtaining the patient's predicted blood glucose value, and using the patient's predicted blood glucose value and a preset blood glucose threshold to obtain a blood glucose risk warning report includes:

[0056] After confirming that the verification and evaluation results are unqualified, the optimized weight parameters of the preliminary blood glucose prediction model are calculated using the pre-constructed backpropagation weight parameter calculation formula, which is shown below:

[0057] ,

[0058] in, This indicates the optimization of weight parameters. The weight parameters represent the initial blood glucose prediction model. For learning rate, This indicates the number of samples used in each update. Represents the first in the verification set The prediction error corresponding to each weighted vector , Represents the first in the verification set The predicted blood glucose value corresponding to each weighted vector. Represents the first in the verification set The actual blood glucose value corresponding to each weighted vector. This represents the gradient of the prediction error with respect to the weight parameters of the preliminary blood glucose prediction model.

[0059] The preliminary blood glucose prediction model is adjusted using the weight parameters, and the adjusted preliminary blood glucose prediction model is used as the preliminary blood glucose prediction model. The process of verifying the preliminary blood glucose prediction model using the validation set and obtaining the verification evaluation result is repeated until the verification evaluation result is qualified. Then, the preliminary blood glucose prediction model corresponding to the weight parameters is confirmed as the optimized blood glucose prediction model.

[0060] The optimized blood glucose prediction model is used to predict the patient's blood glucose level, and the predicted blood glucose value of the patient is obtained.

[0061] The patient's predicted blood glucose value is compared with the blood glucose threshold. If the predicted blood glucose value is greater than or equal to the blood glucose threshold, a blood glucose risk warning report is obtained.

[0062] To achieve the above objectives, the present invention also provides a dynamic blood glucose prediction system that integrates physiological and behavioral data, comprising:

[0063] The data acquisition module is used to collect the patient's raw dataset using a pre-confirmed wearable device and a preset acquisition time interval, and to obtain a blood glucose value sequence using a pre-confirmed blood glucose monitoring device and an acquisition time interval. The raw dataset includes multiple raw data, which includes physiological data and behavioral data, and the blood glucose value sequence includes multiple blood glucose values.

[0064] The data processing module is used to preprocess the original dataset to obtain a preprocessed dataset, which includes multiple preprocessed data. Feature extraction operations are performed on the preprocessed data in the preprocessed dataset and the blood glucose values ​​in the blood glucose value sequence to obtain an original high-dimensional feature dataset. The original high-dimensional feature dataset is fused to obtain a multimodal feature vector set. The correlation between the multimodal feature vector set and blood glucose values ​​is evaluated using a pre-built time series analysis model to obtain feature importance evaluation values. Based on the feature importance evaluation values, a set of blood glucose value influence weights is obtained. The multimodal feature vector set and the set of blood glucose value influence weights are weighted and fused to obtain a weighted vector set.

[0065] The model creation module is used to divide the weighted vector set into a training set and a validation set. The training set is modeled using a pre-built time series modeling method to obtain a preliminary blood glucose prediction model. The training set accounts for 70% of the weighted vector set, and the validation set accounts for 30% of the weighted vector set. The time corresponding to the validation set is later than that of the training set.

[0066] The validation practice module is used to validate the preliminary blood glucose prediction model using a validation set, obtain validation evaluation results, optimize the preliminary blood glucose prediction model based on the validation evaluation results, obtain an optimized blood glucose prediction model, use the optimized blood glucose prediction model to predict blood glucose for patients, obtain predicted blood glucose values ​​for patients, and obtain blood glucose risk warning reports using predicted blood glucose values ​​for patients and preset blood glucose thresholds, thereby realizing dynamic prediction of blood glucose.

[0067] To address the above problems, the present invention also provides an electronic device, the electronic device comprising:

[0068] Memory, storing at least one instruction; and

[0069] The processor executes the instructions stored in the memory to implement the above-described method for dynamic blood glucose prediction by fusing physiological and behavioral data.

[0070] To address the aforementioned problems, the present invention also provides a computer-readable storage medium storing at least one instruction, which is executed by a processor in an electronic device to implement the aforementioned method for dynamic blood glucose prediction by fusing physiological and behavioral data.

[0071] To address the problems described in the background section, this invention utilizes a pre-verified wearable device and a preset acquisition time interval to collect a patient's raw dataset. It then uses a pre-verified blood glucose monitoring device and the acquisition time interval to obtain a blood glucose value sequence. The raw dataset includes multiple raw data points, encompassing physiological and behavioral data. The blood glucose value sequence includes multiple blood glucose values. Therefore, this invention comprehensively considers multiple characteristic factors, including physiological data, behavioral data, and blood glucose values, enabling a more comprehensive capture of the dynamic characteristics of blood glucose changes, thereby significantly improving the accuracy and reliability of blood glucose prediction. Furthermore, this invention preprocesses the raw dataset to obtain a preprocessed dataset, which includes... This invention includes multiple preprocessed data sets. Feature extraction is performed on both the preprocessed data in the preprocessed dataset and the blood glucose values ​​in the blood glucose value sequence to obtain an original high-dimensional feature dataset. The original high-dimensional feature dataset is then fused to obtain a multimodal feature vector set. A pre-built time series analysis model is used to evaluate the correlation between the multimodal feature vector set and blood glucose values ​​to obtain feature importance evaluation values. Based on the feature importance evaluation values, a weight set for the influence of blood glucose values ​​is obtained. The multimodal feature vector set and the weight set for the influence of blood glucose values ​​are then weighted and fused to obtain a weighted vector set. It can be seen that the embodiments of the present invention provide more optimized input data for blood glucose prediction, further improve prediction accuracy and model generalization ability, and provide more accurate decision support for blood glucose management. Next, this invention divides the weighted vector set into a training set and a validation set. A pre-built time series modeling method is used to model the training set to obtain a preliminary blood glucose prediction model. The training set accounts for 70% of the weighted vector set, and the validation set accounts for 30%. Furthermore, the time corresponding to the validation set is later than that of the training set. It is evident that this embodiment of the invention ensures the temporal order of the data by dividing the weighted vector set into a training set and a validation set. The pre-built time series modeling method is used to model the training set to obtain a preliminary blood glucose prediction model. This division method not only fully utilizes the temporal characteristics of the data but also evaluates the model performance through the validation set, ensuring the stability and reliability of the model.Furthermore, this invention uses a validation set to validate the preliminary blood glucose prediction model, obtaining validation evaluation results. Based on these results, the preliminary blood glucose prediction model is optimized to obtain an optimized blood glucose prediction model. This optimized model is then used to predict blood glucose levels in patients, yielding predicted blood glucose values. Using these predicted values ​​and preset blood glucose thresholds, a blood glucose risk warning report is generated, achieving dynamic blood glucose prediction. Thus, this embodiment of the invention uses a validation set to validate the preliminary blood glucose prediction model, ensuring its performance on unseen data. Based on the validation evaluation results, the preliminary blood glucose prediction model is further optimized to obtain an optimized model. Using this optimized model, more accurate predicted blood glucose values ​​are obtained for patients. Combined with preset blood glucose thresholds, a blood glucose risk warning report is generated, thereby achieving dynamic blood glucose prediction. This provides timely and accurate warning information for patients' blood glucose management, significantly improving the efficiency and effectiveness of blood glucose management. Therefore, this invention can solve the problems of considering only a single feature factor and low prediction accuracy in blood glucose prediction. Attached Figure Description

[0072] Figure 1 This is a flowchart illustrating a method for dynamic blood glucose prediction that integrates physiological and behavioral data, provided in an embodiment of the present invention.

[0073] Figure 2 This is a functional block diagram of a dynamic blood glucose prediction system that integrates physiological and behavioral data, provided in an embodiment of the present invention.

[0074] Figure 3 This is a schematic diagram of the structure of an electronic device that implements the dynamic blood glucose prediction method that integrates physiological and behavioral data, according to an embodiment of the present invention.

[0075] Explanation of reference numerals in the attached figures:

[0076] 10. Electronic device; 11. Processor; 12. Memory; 13. Bus.

[0077] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0078] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0079] This application provides a method for dynamic blood glucose prediction that integrates physiological and behavioral data. The executing entity of this method includes, but is not limited to, at least one electronic device configured to execute the method provided in this application, such as a server or a terminal. In other words, the method for dynamic blood glucose prediction that integrates physiological and behavioral data can be executed by software or hardware installed on a terminal device or a server device, and the software may be a blockchain platform. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster.

[0080] Reference Figure 1 The diagram shown is a flowchart illustrating a method for dynamic blood glucose prediction that integrates physiological and behavioral data, according to an embodiment of the present invention. In this embodiment, the method for dynamic blood glucose prediction that integrates physiological and behavioral data includes:

[0081] S1. Collect the patient's raw dataset using a pre-confirmed wearable device and a preset collection time interval. Obtain the blood glucose value sequence using a pre-confirmed blood glucose monitoring device and a collection time interval. The raw dataset includes multiple raw data, including physiological and behavioral data. The blood glucose value sequence includes multiple blood glucose values.

[0082] It should be explained that the process of collecting the patient's raw dataset using a pre-verified wearable device and a preset collection time interval, and obtaining a blood glucose value sequence using a pre-verified blood glucose monitoring device and the collection time interval, includes:

[0083] The patient's raw dataset is collected using wearable devices and at specific time intervals. The raw dataset includes multiple raw data sets, which include physiological and behavioral data.

[0084] Blood glucose values ​​of patients are obtained using blood glucose monitoring equipment and at specific time intervals. The obtained blood glucose values ​​are then arranged in chronological order to obtain a blood glucose value sequence.

[0085] Furthermore, pre-identified wearable devices refer to portable devices used to monitor and record patients' physiological and behavioral data, such as smart bracelets. Preset data collection intervals refer to the time intervals at which wearable devices and blood glucose monitoring devices collect data, for example, every 5 minutes.

[0086] Understandably, the raw dataset refers to a collection of raw patient data gathered by wearable devices. This raw data includes physiological and behavioral data. Physiological data reflects the patient's physiological state, such as heart rate, blood pressure, body temperature, blood oxygen saturation, and respiratory rate. Behavioral data reflects the patient's daily activities and behavioral patterns, such as steps taken, exercise intensity, activity time, sedentary time, sleep duration, and sleep quality.

[0087] Further, the pre-confirmed blood glucose monitoring device refers to a device used to measure a patient's blood glucose. Optionally, a continuous glucose monitoring system (CGM) is used as the blood glucose monitoring device. The blood glucose value sequence is a sequence obtained by arranging multiple blood glucose values ​​in chronological order. The multiple blood glucose values ​​are blood glucose values ​​collected at different time points according to the collection time interval using the blood glucose monitoring device.

[0088] S2. Preprocess the original dataset to obtain a preprocessed dataset, which includes multiple preprocessed data.

[0089] It should be explained that the preprocessing of the original dataset to obtain the preprocessed dataset includes:

[0090] Perform missing data detection on the original dataset and obtain a missing data detection report;

[0091] Missing data is identified using the missing data detection report. Missing data includes single-point random missing data or consecutive missing data segments.

[0092] Complete the missing data by filling in the single point of random missing data in the missing data to obtain the completed dataset;

[0093] Identify the data missing time of consecutive missing data segments in the missing data. If the data missing time is confirmed to be less than a preset missing time threshold, then reconstruct the consecutive missing data segments in the missing data to obtain the completed data segments.

[0094] The completed dataset, the completed data segment, and the original dataset are integrated to obtain the repaired dataset;

[0095] The repair dataset is normalized using a preset normalization method to obtain a normalized dataset;

[0096] The normalized dataset and blood glucose value sequence are time-calibrated to obtain a preprocessed dataset.

[0097] Furthermore, missing data detection refers to identifying missing data points or segments by examining the integrity of the data in the original dataset at each time point. A missing data detection report is generated using this method, recording the location, type (single-point random missing or consecutive missing), and time range of the missing data in the original dataset. Missing data refers to data that is missing from the original dataset, which may have been missed due to reasons such as equipment malfunction, signal loss, or the patient not wearing the device.

[0098] Understandably, single-point random missing data refers to the absence of a single data point randomly occurring in the dataset, such as the loss of heart rate data at a certain time point. Continuous missing data segments refer to the absence of multiple consecutive data points in the original dataset, such as data loss due to prolonged periods of non-wearability of wearable devices. Completing single-point random missing data to obtain a completed dataset involves using a pre-built interpolation method to fill in the missing data points, resulting in single-point completed data. There may be one or more single-point random missing data points, and correspondingly, one or more single-point completed data points may also exist. These single-point completed data points are then aggregated to obtain the completed dataset. Optionally, a linear interpolation method is used as the interpolation method. A completed dataset refers to the dataset obtained by using an interpolation method to complete one or more single-point random missing data points, obtaining single-point completed data, and then aggregating the single-point completed data points. Data missing time refers to the duration of a continuous missing data segment, such as the length of time from the point when the data began to be missing to the point when the data was recovered. The preset missing time threshold is a pre-defined time length, such as 30 minutes, used to determine whether consecutive missing data segments can be reconstructed. If the missing time is less than the missing time threshold, the consecutive missing data segments can be reconstructed. Using the missing time threshold allows for the selection of suitable consecutive missing data segments for reconstruction, ensuring data reliability and improving data processing efficiency.

[0099] Furthermore, reconstructing consecutively missing data segments to obtain completed data segments means that if the missing time of consecutively missing data segments is less than a missing time threshold, then the consecutively missing data segments are reconstructed using a pre-built regression model to obtain completed data segments. Optionally, a linear regression model is used as the regression model. The completed data segment is a data segment obtained by reconstructing consecutively missing data segments using the regression model, and the completed data segment may contain one or more completed data segments.

[0100] Understandably, integrating the completed dataset, the completed data segment, and the original dataset to obtain the repaired dataset means adding the completed dataset and the completed data segment to the original dataset according to the time dimension, and confirming the supplemented original dataset as the repaired dataset.

[0101] It should be explained that the default normalization method scales the repaired physiological and behavioral data in the repaired dataset to a range of (0-1) to improve efficiency and accuracy in subsequent analysis and modeling. Optionally, Min-Max normalization can be used as the normalization method. The normalized dataset refers to the repaired dataset after normalization processing. Time calibration of the normalized dataset and blood glucose value sequence yields the preprocessed dataset. This involves aligning the normalized dataset and blood glucose value sequence in time to ensure consistent timestamps. If the missing time of consecutive missing data segments is greater than or equal to the missing time threshold, the blood glucose values ​​corresponding to those consecutive missing data segments are removed from the blood glucose value sequence. This allows for correlation analysis between the normalized physiological and behavioral data and blood glucose values, resulting in the preprocessed dataset.

[0102] S3. Perform feature extraction operations on both the preprocessed data in the preprocessed dataset and the blood glucose values ​​in the blood glucose value sequence to obtain the original high-dimensional feature dataset.

[0103] It should be explained that feature extraction operations are performed on both the preprocessed data in the preprocessed dataset and the blood glucose values ​​in the blood glucose value sequence to obtain the original high-dimensional feature dataset, including:

[0104] Extract the preprocessed physiological dataset and the preprocessed behavioral dataset from the preprocessed dataset;

[0105] Temporal feature extraction is performed on the preprocessed physiological dataset to obtain a set of physiological feature vectors;

[0106] The preprocessed behavioral dataset is segmented and aggregated to approximate the data, resulting in a set of behavioral feature vectors.

[0107] Feature extraction is performed on the blood glucose value sequence to obtain a blood glucose feature vector set;

[0108] The physiological feature vector set, behavioral feature vector set, and blood glucose feature vector set are correlated along the time dimension to obtain the original high-dimensional feature dataset.

[0109] Furthermore, the preprocessed physiological dataset is a preprocessed physiological dataset separated from the preprocessed dataset. Similarly, the preprocessed behavioral dataset is a preprocessed behavioral dataset separated from the preprocessed dataset. Extracting the preprocessed physiological and behavioral datasets from the preprocessed datasets allows for focused analysis on different types of features. Physiological and behavioral data typically possess different characteristics and influencing factors; extracting them separately allows for a deeper understanding of their impact on blood glucose levels.

[0110] It's important to explain that time-domain feature extraction (TDLE) involves extracting statistical features from preprocessed physiological datasets. These statistical features reflect the distribution, trends, and fluctuations of physiological characteristics and can be correlated with blood glucose levels to reveal their correspondence. TDLE is typically performed within a specific time window. Statistical features are numerical values ​​extracted from preprocessed physiological datasets that describe the distribution and characteristics of physiological data, such as mean heart rate, kurtosis, and rate of change. A time window refers to a specific time range selected for extracting statistical features. For example, when calculating the mean heart rate, a time window (e.g., every 5 minutes) can be specified to calculate the average heart rate within that window. A physiological feature vector set is a collection of multiple feature vectors extracted from preprocessed physiological data. Each feature vector contains multiple statistical features used to describe the characteristics of physiological data within a specific time window. Using physiological feature vector sets can effectively transform complex physiological data into meaningful feature representations, supporting subsequent analysis and modeling.

[0111] Understandably, segmented aggregation approximation is a data dimensionality reduction technique that divides time-series data into several segments and aggregates (e.g., takes the average) each segment, thereby simplifying the data representation. This reduces data dimensionality while preserving the main trends. Optionally, time-series analysis can be used as the method for segmented aggregation approximation. The behavioral feature vector set is a collection of feature vectors obtained after segmented aggregation approximation of the preprocessed behavioral dataset. Each feature vector contains aggregated features of the behavioral data, such as the average number of steps taken and the average activity intensity within each time segment. Feature extraction extracts features of blood glucose value changes from the blood glucose value sequence, such as the average blood glucose value, kurtosis, and average rate of change. This simplifies complex blood glucose time-series data into a set of meaningful features, reducing data dimensionality and allowing for a better understanding of the dynamic changes in blood glucose. Optionally, time-series analysis can be used as the method for feature extraction. The blood glucose feature vector set is a collection of features containing blood glucose changes within multiple time windows.

[0112] Furthermore, the physiological feature vector set, behavioral feature vector set, and blood glucose feature vector set are correlated along the time dimension to obtain the original high-dimensional feature dataset. This refers to aligning the physiological feature vectors, behavioral feature vectors, and blood glucose feature vectors in chronological order to form a high-dimensional dataset containing multiple features. The original high-dimensional feature data at each time point contains physiological, behavioral, and blood glucose features, which together describe the patient's state at that time point. The original high-dimensional feature dataset can be used for subsequent modeling and analysis.

[0113] S4. The original high-dimensional feature dataset is fused to obtain a multimodal feature vector set.

[0114] It should be explained that the process of fusing the original high-dimensional feature dataset to obtain a multimodal feature vector set includes:

[0115] Multimodal feature extraction was performed on the original high-dimensional feature dataset to obtain blood glucose modality group, physiological modality group and behavioral modality group;

[0116] A fusion feature attention matrix is ​​constructed using the blood glucose modality group as the reference query vector, the physiological modality group as the key vector, and the behavioral modality group as the value vector, respectively.

[0117] Feature decoupling is performed on the fused feature attention matrix to obtain behavioral modality feature submatrices and physiological modality feature submatrices;

[0118] The behavioral modality feature submatrix and the physiological modality feature submatrix are dimensionality reduced respectively to obtain the behavioral modality feature vector set and the physiological modality feature vector set;

[0119] By summing the behavioral modality feature vector set and the physiological modality feature vector set, a multimodal feature vector set is obtained.

[0120] Furthermore, multimodal feature extraction extracts features from different modalities, including blood glucose modality, physiological modality, and behavioral modality, from the original high-dimensional feature dataset. The purpose of multimodal feature extraction is to extract these different types of data separately for subsequent processing and analysis.

[0121] Understandably, the blood glucose modality set is a collection of feature vectors related to blood glucose extracted from the original high-dimensional feature dataset, including blood glucose values ​​themselves, their trends, and fluctuations. The physiological modality set is a collection of feature vectors related to physiology extracted from the original high-dimensional feature dataset, including feature vectors of physiological indicators such as heart rate, blood pressure, and body temperature. The behavioral modality set is a collection of feature vectors related to behavior extracted from the original high-dimensional feature dataset, including feature vectors of behavioral data such as steps taken, activity time, and sleep duration.

[0122] Understandingly, constructing a fused feature attention matrix using the blood glucose modality as the reference query vector, the physiological modality as the key vector, and the behavioral modality as the value vector refers to using the blood glucose modality as the query vector to guide the attention mechanism and determine which features will cause changes in blood glucose. The physiological modality is used as the key vector to match with the query vector to determine the correlation between physiological features and blood glucose features. The behavioral modality is used as the value vector to provide specific feature values, which are weighted and summarized based on the matching results of the query vector and the key vector to obtain the fused feature attention matrix. Using the fused feature attention matrix can better capture the features most relevant to changes in blood glucose, thereby improving the accuracy and robustness of subsequent analysis and modeling.

[0123] It should be explained that feature decoupling decomposes the fused feature attention matrix into multiple sub-matrices, each corresponding to a specific modality's features. The purpose of feature decoupling is to decompose the fused feature attention matrix into multiple sub-matrices, each corresponding to a specific modality (such as behavioral modality and physiological modality), which allows for clearer analysis and processing of the features of each modality and avoids interference between different modalities. The behavioral modality feature sub-matrix is ​​a feature sub-matrix related to the behavioral modality obtained using feature decoupling. The physiological modality feature sub-matrix is ​​a feature sub-matrix related to the physiological modality obtained using feature decoupling. Dimensionality reduction reduces the dimension of the feature sub-matrices while retaining their statistical characteristics. Optionally, the PCA algorithm is used as the dimensionality reduction method. The behavioral modality feature vector set is the set of behavioral modality feature vectors obtained after dimensionality reduction. The physiological modality feature vector set is the set of physiological modality feature vectors obtained after dimensionality reduction. The multimodal feature vector set is a set obtained by summing the behavioral modality feature vector set and the physiological modality feature vector set.

[0124] S5. Use a pre-built time series analysis model to evaluate the correlation between blood glucose values ​​and the multimodal feature vector set, obtain the feature importance evaluation value, and obtain the weight set of blood glucose value influence based on the feature importance evaluation value.

[0125] It should be explained that the process of using a pre-built time-series analysis model to assess the correlation between blood glucose levels and multimodal feature vector sets, obtaining feature importance assessment values, and then obtaining a set of weights for the influence of blood glucose levels based on these feature importance assessment values ​​includes:

[0126] The behavioral modality feature vector set and the physiological modality feature vector set in the multimodal feature vector set are time-aligned with the blood glucose value sequence to obtain the aligned feature vector set;

[0127] A time series analysis model was trained using aligned feature vector sets to obtain a blood glucose assessment model;

[0128] The importance assessment values ​​of behavioral characteristics and physiological characteristics were obtained using a blood glucose assessment model.

[0129] A weight set for the influence of behavioral blood glucose values ​​is generated based on the numerical assessment of the importance of behavioral characteristics, and a weight set for the influence of physiological blood glucose values ​​is generated based on the numerical assessment of the importance of physiological characteristics.

[0130] The weight set of behavioral blood glucose value influence is merged with the weight set of physiological blood glucose value influence to obtain the weight set of blood glucose value influence.

[0131] Furthermore, time alignment synchronizes the behavioral modality feature vector, physiological modality feature vector, and blood glucose value sequence over time. The aligned feature vector set is the set of feature vectors obtained after synchronizing the behavioral modality feature vector, physiological modality feature vector, and blood glucose value sequence over time. The pre-built time series analysis model is a model used to analyze time series data, capable of capturing trends, periodicity, and correlations in the time series. Optionally, a long short-term memory artificial neural network is used as the time series analysis model.

[0132] For example, the behavioral modality vector is (time: 0, steps: 100, activity time: 30), the physiological modality vector is (time: 0, heart rate: 70, blood pressure: 120), and the blood glucose value is (time: 0, blood glucose value: 5.0). After synchronization, the aligned feature vector in the aligned feature vector set is (time: 0, steps: 100, activity time: 30, heart rate: 70, blood pressure: 120, blood glucose value: 5.0).

[0133] It needs to be explained that training a time-series analysis model using alignment sets of behavioral feature vectors, physiological feature vectors, and blood glucose values ​​to obtain a blood glucose assessment model means using these alignment sets as input data to train the time-series analysis model to learn the relationship between behavioral features, physiological features, and blood glucose values. The blood glucose assessment model is a trained time-series analysis model capable of assessing the impact of behavioral and physiological features on blood glucose values. Feature importance assessment values ​​evaluate the contribution of each feature to blood glucose value prediction through the blood glucose assessment model, reflecting which features are more sensitive to changes in blood glucose values. Behavioral feature importance assessment values ​​evaluate the degree of influence of behavioral features (such as step count, activity time, etc.) on blood glucose values. Physiological feature importance assessment values ​​evaluate the degree of influence of physiological features (such as heart rate, blood pressure, etc.) on blood glucose values.

[0134] Furthermore, the behavioral blood glucose level influence weight set is a set of weights generated based on the importance assessment values ​​of behavioral characteristics, used to represent the influence of behavioral characteristics on blood glucose levels. The physiological blood glucose level influence weight set is a set of weights generated based on the importance assessment values ​​of physiological characteristics, used to represent the influence of physiological characteristics on blood glucose levels.

[0135] S6. The multimodal feature vector set and the weight set of blood glucose value influence are weighted and fused to obtain a weighted vector set.

[0136] It's important to explain that weighted fusion involves weighting each feature vector in the multimodal feature vector set with its corresponding weight representing its impact on blood glucose levels. Each feature vector is multiplied by its weight to obtain the weighted feature vector. The purpose of weighted fusion is to adjust the contribution of feature vectors according to their importance, ensuring that features with a greater impact on blood glucose levels play a more significant role in subsequent analysis. The weighted vector set, obtained after weighted fusion, better reflects the feature information related to blood glucose levels, providing more optimized input data for subsequent blood glucose prediction models. Weighted fusion highlights features with a greater impact on blood glucose levels while suppressing features with a smaller impact, thus improving the accuracy and reliability of subsequent blood glucose prediction models.

[0137] S7. Divide the weighted vector set into a training set and a validation set. Use a pre-built time series modeling method to model the training set to obtain a preliminary blood glucose prediction model. The training set accounts for 70% of the weighted vector set, and the validation set accounts for 30% of the weighted vector set. The time corresponding to the validation set is later than that of the training set.

[0138] It should be explained that the process of dividing the weighted vector set into a training set and a validation set, and using a pre-built time series modeling method to model the training set to obtain a preliminary blood glucose prediction model includes:

[0139] Arrange the weighted vector set in chronological order to obtain the weighted vector sequence;

[0140] Based on the proportion of the training set in the weighted vector sequence and the proportion of the validation set in the weighted vector sequence, the split point between the training set and the validation set is determined. The weighted vectors in the weighted vector sequence before the split point are used as the training set, and the weighted vectors after the split point are used as the validation set.

[0141] The multiple weighted vectors in the training set are normalized to obtain a normalized training set, which includes multiple normalized weighted vectors.

[0142] The normalized weighted vectors in the normalized training set are divided by time step to obtain multiple time series samples. Each time series sample includes multiple normalized weighted vectors and multiple blood glucose values, and the normalized weighted vectors correspond one-to-one with the blood glucose values.

[0143] A preliminary blood glucose prediction model was obtained by using time series modeling methods and multiple time series samples.

[0144] Furthermore, the weighted vector sequence is formed by arranging the weighted vector set in chronological order, creating an ordered sequence. The training set, starting from the beginning of the weighted vector sequence, selects 70% of the weighted vectors as the training set for training the model. The validation set, consisting of the remaining weighted vectors in the sequence, is used to validate the model's performance. The split point, determined by the proportions, divides the weighted vector sequence into the training and validation sets. Dividing the weighted vector set into training and validation sets ensures that the model learns patterns in the data during training, while the validation set evaluates the model's generalization ability and prevents overfitting.

[0145] Understandably, normalization scales multiple weighted vectors in the training set to a range of (0-1). Normalization helps improve the convergence speed and performance of the model and avoids the impact of differences in feature value ranges on model training. Optionally, Min-Max normalization can be used as the normalization method.

[0146] It should be explained that the normalized training set is a training set that has undergone normalization processing, where each weighted vector is converted into a normalized weighted vector. Normalization helps improve the convergence speed and performance of the model. Time step partitioning involves dividing the weighted vectors in the normalized training set according to the time step, forming multiple time series samples. The time step refers to the number of time points contained in each sample; for example, if the time step is 10, then each sample contains 10 consecutive weighted vectors. Through time step partitioning, the model can learn the dynamic relationship between features and blood glucose values, improving its ability to model time series data. A time series sample refers to a sample that includes multiple normalized weighted vectors and their corresponding blood glucose values, used to construct a time series model, helping the time series model learn the dynamic relationship between behavioral data, physiological data, and blood glucose values. The time series modeling method is a method used to model time series data. Optionally, a Long Short-Term Memory (LSTM) artificial neural network is used as the time series modeling method. The preliminary blood glucose prediction model is a model trained using the time series modeling method and multiple time series samples, capable of predicting the patient's blood glucose values. The initial training objective of the blood glucose prediction model is to minimize the prediction error and improve the accuracy and reliability of the prediction.

[0147] S8. Validate the preliminary blood glucose prediction model using the validation set to obtain the validation evaluation results. Optimize the preliminary blood glucose prediction model based on the validation evaluation results to obtain the optimized blood glucose prediction model. Use the optimized blood glucose prediction model to predict the patient's blood glucose level to obtain the patient's predicted blood glucose value. Use the patient's predicted blood glucose value and a preset blood glucose threshold to obtain a blood glucose risk warning report, thereby achieving dynamic prediction of blood glucose.

[0148] It should be explained that the validation of the preliminary blood glucose prediction model using a validation set to obtain validation evaluation results includes:

[0149] Multiple weighted vectors from the validation set are input into the preliminary blood glucose prediction model to obtain a set of predicted blood glucose values.

[0150] The average prediction error is calculated using the predicted blood glucose values ​​from the predicted blood glucose value set, the actual blood glucose values ​​corresponding to the weighted vector, and a pre-constructed average absolute percentage error calculation formula. The average absolute percentage error calculation formula is as follows:

[0151] ,

[0152] in, Indicates the average prediction error. This represents the number of weighted vectors in the validation set. Represents the first in the verification set The predicted blood glucose value corresponding to each weighted vector. Represents the first in the verification set The actual blood glucose values ​​corresponding to each weighted vector;

[0153] The verification and evaluation results are obtained by using the average prediction error and the preset error threshold. If the average prediction error is less than the error threshold, the verification and evaluation results are deemed qualified; if the average prediction error is greater than or equal to the error threshold, the verification and evaluation results are deemed unqualified.

[0154] Furthermore, the predicted blood glucose value set is a collection containing multiple predicted blood glucose values. A predicted blood glucose value is obtained by inputting a weighted vector into the preliminary blood glucose prediction model. The actual blood glucose value is the blood glucose value collected by a blood glucose monitoring device, corresponding to the weighted vector. The average prediction error is an evaluation metric used to measure the error between the predicted and actual blood glucose values. The preset error threshold is a pre-defined upper limit for the error, used to determine whether the prediction performance of the preliminary blood glucose prediction model is satisfactory. If the average prediction error is less than the error threshold, the preliminary blood glucose prediction model is considered satisfactory; otherwise, it is considered unsatisfactory.

[0155] It should be explained that the process of optimizing the preliminary blood glucose prediction model based on the verification and evaluation results to obtain an optimized blood glucose prediction model, using the optimized blood glucose prediction model to predict the patient's blood glucose level, and using the patient's predicted blood glucose level and a preset blood glucose threshold to obtain a blood glucose risk warning report includes:

[0156] After confirming that the verification and evaluation results are unqualified, the optimized weight parameters of the preliminary blood glucose prediction model are calculated using the pre-constructed backpropagation weight parameter calculation formula, which is shown below:

[0157] ,

[0158] in, This indicates the optimization of weight parameters. The weight parameters represent the initial blood glucose prediction model. For learning rate, This indicates the number of samples used in each update. Represents the first in the verification set The prediction error corresponding to each weighted vector , Represents the first in the verification set The predicted blood glucose value corresponding to each weighted vector. Represents the first in the verification set The actual blood glucose value corresponding to each weighted vector. This represents the gradient of the prediction error with respect to the weight parameters of the preliminary blood glucose prediction model.

[0159] The preliminary blood glucose prediction model is adjusted using the weight parameters, and the adjusted preliminary blood glucose prediction model is used as the preliminary blood glucose prediction model. The process of verifying the preliminary blood glucose prediction model using the validation set and obtaining the verification evaluation result is repeated until the verification evaluation result is qualified. Then, the preliminary blood glucose prediction model corresponding to the weight parameters is confirmed as the optimized blood glucose prediction model.

[0160] The optimized blood glucose prediction model is used to predict the patient's blood glucose level, and the predicted blood glucose value of the patient is obtained.

[0161] The patient's predicted blood glucose value is compared with the blood glucose threshold. If the predicted blood glucose value is greater than or equal to the blood glucose threshold, a blood glucose risk warning report is obtained.

[0162] Furthermore, the optimized weight parameters are the weight parameters of the preliminary blood glucose prediction model calculated using the backpropagation weight parameter calculation formula. The weight parameters of the preliminary blood glucose prediction model are parameters used in the model to adjust the importance of input features, where the initial weight parameters are set to 1. The prediction error is the error between the actual blood glucose value and the corresponding predicted blood glucose value of the weighted vector. The learning rate is a hyperparameter preset when the preliminary blood glucose prediction model is established.

[0163] For example, assuming B=4, it means that the optimized weight parameters are calculated using the prediction error corresponding to the four weighted parameters, the corresponding actual blood glucose value, and the backpropagation weight parameter calculation formula.

[0164] Understandably, the optimized blood glucose prediction model is a performance-compliant model obtained after multiple iterations and confirmation of optimized weight parameters. This optimized model accurately predicts a patient's blood glucose levels. The predicted blood glucose level is the value obtained by using the optimized model to predict the patient's blood glucose. A preset blood glucose threshold is a pre-defined upper limit for blood glucose levels, used to determine if a patient has a risk of blood glucose deficiency. If the predicted blood glucose level is greater than or equal to the threshold, the patient is considered to have a risk of blood glucose deficiency. The blood glucose threshold needs to be set according to the patient's actual situation; for example, the fasting blood glucose threshold is set to 6.1 mmol / L, and the blood glucose threshold two hours after a meal is set to 7.8 mmol / L. A blood glucose risk warning report is generated when the patient's predicted blood glucose level is greater than or equal to the threshold, used to remind the patient or healthcare personnel of the risk of blood glucose deficiency.

[0165] To address the problems described in the background section, this invention utilizes a pre-verified wearable device and a preset acquisition time interval to collect a patient's raw dataset. It then uses a pre-verified blood glucose monitoring device and the acquisition time interval to obtain a blood glucose value sequence. The raw dataset includes multiple raw data points, encompassing physiological and behavioral data. The blood glucose value sequence includes multiple blood glucose values. Therefore, this invention comprehensively considers multiple characteristic factors, including physiological data, behavioral data, and blood glucose values, enabling a more comprehensive capture of the dynamic characteristics of blood glucose changes, thereby significantly improving the accuracy and reliability of blood glucose prediction. Furthermore, this invention preprocesses the raw dataset to obtain a preprocessed dataset, which includes... This invention includes multiple preprocessed data sets. Feature extraction is performed on both the preprocessed data in the preprocessed dataset and the blood glucose values ​​in the blood glucose value sequence to obtain an original high-dimensional feature dataset. The original high-dimensional feature dataset is then fused to obtain a multimodal feature vector set. A pre-built time series analysis model is used to evaluate the correlation between the multimodal feature vector set and blood glucose values ​​to obtain feature importance evaluation values. Based on the feature importance evaluation values, a weight set for the influence of blood glucose values ​​is obtained. The multimodal feature vector set and the weight set for the influence of blood glucose values ​​are then weighted and fused to obtain a weighted vector set. It can be seen that the embodiments of the present invention provide more optimized input data for blood glucose prediction, further improve prediction accuracy and model generalization ability, and provide more accurate decision support for blood glucose management. Next, this invention divides the weighted vector set into a training set and a validation set. A pre-built time series modeling method is used to model the training set to obtain a preliminary blood glucose prediction model. The training set accounts for 70% of the weighted vector set, and the validation set accounts for 30%. Furthermore, the time corresponding to the validation set is later than that of the training set. It is evident that this embodiment of the invention ensures the temporal order of the data by dividing the weighted vector set into a training set and a validation set. The pre-built time series modeling method is used to model the training set to obtain a preliminary blood glucose prediction model. This division method not only fully utilizes the temporal characteristics of the data but also evaluates the model performance through the validation set, ensuring the stability and reliability of the model.Furthermore, this invention uses a validation set to validate the preliminary blood glucose prediction model, obtaining validation evaluation results. Based on these results, the preliminary blood glucose prediction model is optimized to obtain an optimized blood glucose prediction model. This optimized model is then used to predict blood glucose levels in patients, yielding predicted blood glucose values. Using these predicted values ​​and preset blood glucose thresholds, a blood glucose risk warning report is generated, achieving dynamic blood glucose prediction. Thus, this embodiment of the invention uses a validation set to validate the preliminary blood glucose prediction model, ensuring its performance on unseen data. Based on the validation evaluation results, the preliminary blood glucose prediction model is further optimized to obtain an optimized model. Using this optimized model, more accurate predicted blood glucose values ​​are obtained for patients. Combined with preset blood glucose thresholds, a blood glucose risk warning report is generated, thereby achieving dynamic blood glucose prediction. This provides timely and accurate warning information for patients' blood glucose management, significantly improving the efficiency and effectiveness of blood glucose management. Therefore, this invention can solve the problems of considering only a single feature factor and low prediction accuracy in blood glucose prediction.

[0166] like Figure 2 The diagram shown is a functional block diagram of a dynamic blood glucose prediction system that integrates physiological and behavioral data, provided in an embodiment of the present invention.

[0167] The blood glucose dynamic prediction system 100, which integrates physiological and behavioral data, described in this invention, can be installed in an electronic device. Depending on the functions implemented, the blood glucose dynamic prediction system 100 may include a data acquisition module 101, a data processing module 102, a model creation module 103, and a validation practice module 104. The module described in this invention can also be referred to as a unit, which refers to a series of computer program segments that can be executed by the processor of an electronic device and perform a fixed function, and which are stored in the memory of the electronic device.

[0168] The data acquisition module 101 is used to acquire the patient's original dataset using a pre-confirmed wearable device and a preset acquisition time interval, and to obtain a blood glucose value sequence using a pre-confirmed blood glucose monitoring device and an acquisition time interval. The original dataset includes multiple original data, and the original data includes physiological data and behavioral data. The blood glucose value sequence includes multiple blood glucose values.

[0169] The data processing module 102 is used to preprocess the original dataset to obtain a preprocessed dataset, wherein the preprocessed dataset includes multiple preprocessed data. Feature extraction operations are performed on the preprocessed data in the preprocessed dataset and the blood glucose values ​​in the blood glucose value sequence to obtain an original high-dimensional feature dataset. The original high-dimensional feature dataset is fused to obtain a multimodal feature vector set. The blood glucose value correlation of the multimodal feature vector set is evaluated using a pre-built time series analysis model to obtain a feature importance evaluation value. Based on the feature importance evaluation value, a blood glucose value influence weight set is obtained. The multimodal feature vector set and the blood glucose value influence weight set are weighted and fused to obtain a weighted vector set.

[0170] The model creation module 103 is used to divide the weighted vector set into a training set and a validation set, and to model the training set using a pre-built time series modeling method to obtain a preliminary blood glucose prediction model. The training set accounts for 70% of the weighted vector set, the validation set accounts for 30% of the weighted vector set, and the time corresponding to the validation set is later than that of the training set.

[0171] The verification practice module 104 is used to verify the preliminary blood glucose prediction model using a verification set, obtain verification evaluation results, optimize the preliminary blood glucose prediction model based on the verification evaluation results, obtain an optimized blood glucose prediction model, use the optimized blood glucose prediction model to predict blood glucose for patients, obtain predicted blood glucose values ​​for patients, and obtain blood glucose risk warning reports using predicted blood glucose values ​​for patients and preset blood glucose thresholds, thereby realizing dynamic prediction of blood glucose.

[0172] In detail, the modules in the blood glucose dynamic prediction system 100 that integrates physiological and behavioral data described in this embodiment of the invention employ the same methods as described above. Figure 1 The method used is the same as the dynamic blood glucose prediction method that integrates physiological and behavioral data described in the article, and can produce the same technical effect, so it will not be elaborated here.

[0173] like Figure 3 The diagram shown is a schematic representation of an electronic device for implementing a method for dynamic blood glucose prediction that integrates physiological and behavioral data, according to an embodiment of the present invention.

[0174] The electronic device 1 may include a processor 10, a memory 11 and a bus 12, and may also include a computer program stored in the memory 11 and capable of running on the processor 10, such as a blood glucose dynamic prediction method program that integrates physiological data and behavioral data.

[0175] The memory 11 includes at least one type of readable storage medium, such as flash memory, portable hard drive, multimedia card, card-type memory (e.g., SD or DX memory), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 11 can be an internal storage unit of the electronic device 1, such as a portable hard drive. In other embodiments, the memory 11 can be an external storage device of the electronic device 1, such as a plug-in portable hard drive, smart media card (SMC), secure digital card (SD), flash card, etc., equipped on the electronic device 1. Furthermore, the memory 11 includes both internal storage units and external storage devices of the electronic device 1. The memory 11 can be used not only to store application software and various types of data installed on the electronic device 1, such as the code of a blood glucose dynamic prediction method program that integrates physiological and behavioral data, but also to temporarily store data that has been output or will be output.

[0176] In some embodiments, the processor 10 may be composed of integrated circuits, such as a single packaged integrated circuit or multiple integrated circuits with the same or different functions, including combinations of one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and various control chips. The processor 10 is the control unit of the electronic device, connecting various components of the entire electronic device through various interfaces and lines. It executes programs or modules stored in the memory 11 (e.g., a blood glucose dynamic prediction method program that integrates physiological and behavioral data), and calls data stored in the memory 11 to perform various functions of the electronic device 1 and process data.

[0177] The bus 12 can be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus, etc. The bus 12 can be divided into an address bus, a data bus, a control bus, etc. The bus 12 is configured to realize the connection and communication between the memory 11 and at least one processor 10, etc.

[0178] Figure 3 Only electronic devices with components are shown; it will be understood by those skilled in the art that... Figure 3The structure shown does not constitute a limitation on the electronic device 1, and may include fewer or more components than shown, or combine certain components, or have different component arrangements.

[0179] For example, although not shown, the electronic device 1 may also include a power supply (such as a battery) to power the various components. Preferably, the power supply can be logically connected to the at least one processor 10 through a power management system, thereby enabling functions such as charging management, discharging management, and power consumption management through the power management system. The power supply may also include one or more DC or AC power supplies, recharging systems, power fault detection circuits, power converters or inverters, power status indicators, and other arbitrary components. The electronic device 1 may also include various sensors, Bluetooth modules, Wi-Fi modules, etc., which will not be described in detail here.

[0180] Furthermore, the electronic device 1 may also include a network interface. Optionally, the network interface may include a wired interface and / or a wireless interface (such as a Wi-Fi interface, a Bluetooth interface, etc.), which is typically used to establish communication connections between the electronic device 1 and other electronic devices.

[0181] Optionally, the electronic device 1 may further include a user interface, which may be a display, an input unit (such as a keyboard), and optionally, a standard wired interface or a wireless interface. Optionally, in some embodiments, the display may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen, etc. The display may also be appropriately referred to as a screen or display unit, used to display information processed in the electronic device 1 and to display a visual user interface.

[0182] It should be understood that the embodiments described are for illustrative purposes only and are not limited to this structure in the scope of the patent application.

[0183] The blood glucose dynamic prediction method program, which integrates physiological and behavioral data and is stored in the memory 11 of the electronic device 1, is a combination of multiple instructions. When run in the processor 10, it can achieve the following:

[0184] The patient's raw dataset is collected using a pre-confirmed wearable device and a preset collection time interval. The blood glucose value sequence is obtained using a pre-confirmed blood glucose monitoring device and a collection time interval. The raw dataset includes multiple raw data, which includes physiological and behavioral data. The blood glucose value sequence includes multiple blood glucose values.

[0185] The original dataset is preprocessed to obtain a preprocessed dataset, which includes multiple preprocessed data.

[0186] Feature extraction operations were performed on both the preprocessed data in the preprocessed dataset and the blood glucose values ​​in the blood glucose value sequence to obtain the original high-dimensional feature dataset.

[0187] The original high-dimensional feature dataset is fused to obtain a multimodal feature vector set;

[0188] A pre-built time series analysis model is used to evaluate the correlation between blood glucose values ​​and multimodal feature vector sets, thereby obtaining feature importance evaluation values. Based on the feature importance evaluation values, a set of weights for the influence of blood glucose values ​​is obtained.

[0189] The multimodal feature vector set and the weight set of blood glucose value influence are weighted and fused to obtain a weighted vector set;

[0190] The weighted vector set is divided into a training set and a validation set. The training set is modeled using a pre-built time series modeling method to obtain a preliminary blood glucose prediction model. The training set accounts for 70% of the weighted vector set, and the validation set accounts for 30% of the weighted vector set. The time corresponding to the validation set is later than that of the training set.

[0191] The preliminary blood glucose prediction model is validated using a validation set to obtain validation evaluation results. Based on the validation evaluation results, the preliminary blood glucose prediction model is optimized to obtain an optimized blood glucose prediction model. The optimized blood glucose prediction model is used to predict blood glucose levels in patients to obtain predicted blood glucose values. Blood glucose risk warning reports are obtained using the predicted blood glucose values ​​of patients and preset blood glucose thresholds, thereby achieving dynamic prediction of blood glucose.

[0192] Specifically, the processor 10's implementation method for the above instructions can be found in [reference needed]. Figures 1 to 3 The descriptions of the relevant steps in the corresponding embodiments are not repeated here.

[0193] Furthermore, if the modules / units integrated in the electronic device 1 are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. The computer-readable storage medium can be volatile or non-volatile. For example, the computer-readable medium may include: any entity or system capable of carrying the computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, or a read-only memory (ROM).

[0194] The present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor of an electronic device, can perform the following:

[0195] The patient's raw dataset is collected using a pre-confirmed wearable device and a preset collection time interval. The blood glucose value sequence is obtained using a pre-confirmed blood glucose monitoring device and a collection time interval. The raw dataset includes multiple raw data, which includes physiological and behavioral data. The blood glucose value sequence includes multiple blood glucose values.

[0196] The original dataset is preprocessed to obtain a preprocessed dataset, which includes multiple preprocessed data.

[0197] Feature extraction operations were performed on both the preprocessed data in the preprocessed dataset and the blood glucose values ​​in the blood glucose value sequence to obtain the original high-dimensional feature dataset.

[0198] The original high-dimensional feature dataset is fused to obtain a multimodal feature vector set;

[0199] A pre-built time series analysis model is used to evaluate the correlation between blood glucose values ​​and multimodal feature vector sets, thereby obtaining feature importance evaluation values. Based on the feature importance evaluation values, a set of weights for the influence of blood glucose values ​​is obtained.

[0200] The multimodal feature vector set and the weight set of blood glucose value influence are weighted and fused to obtain a weighted vector set;

[0201] The weighted vector set is divided into a training set and a validation set. The training set is modeled using a pre-built time series modeling method to obtain a preliminary blood glucose prediction model. The training set accounts for 70% of the weighted vector set, and the validation set accounts for 30% of the weighted vector set. The time corresponding to the validation set is later than that of the training set.

[0202] The preliminary blood glucose prediction model is validated using a validation set to obtain validation evaluation results. Based on the validation evaluation results, the preliminary blood glucose prediction model is optimized to obtain an optimized blood glucose prediction model. The optimized blood glucose prediction model is used to predict blood glucose levels in patients to obtain predicted blood glucose values. Blood glucose risk warning reports are obtained using the predicted blood glucose values ​​of patients and preset blood glucose thresholds, thereby achieving dynamic prediction of blood glucose.

[0203] In the embodiments provided by this invention, it should be understood that the disclosed devices, systems, and methods can be implemented in other ways. For example, the system embodiments described above are merely illustrative, and actual implementations may have other classification methods.

[0204] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0205] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in the form of hardware plus software functional modules.

[0206] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0207] Furthermore, it is clear that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units or systems stated in a system claim may also be implemented by a single unit or system through software or hardware. The term "second class" is used to indicate names and does not indicate any specific order.

[0208] 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.

Claims

1. A method for dynamic blood glucose prediction that integrates physiological and behavioral data, characterized in that, The method includes: The patient's raw dataset is collected using a pre-confirmed wearable device and a preset collection time interval. The blood glucose value sequence is obtained using a pre-confirmed blood glucose monitoring device and a collection time interval. The raw dataset includes multiple raw data, which includes physiological and behavioral data. The blood glucose value sequence includes multiple blood glucose values. The original dataset is preprocessed to obtain a preprocessed dataset, which includes multiple preprocessed data. The preprocessing of the original dataset to obtain a preprocessed dataset includes: Perform missing data detection on the original dataset and obtain a missing data detection report; Missing data is identified using the missing data detection report. Missing data includes single-point random missing data or consecutive missing data segments. Complete the missing data by filling in the single point of random missing data in the missing data to obtain the completed dataset; Identify the data missing time of consecutive missing data segments in the missing data. If the data missing time is confirmed to be less than a preset missing time threshold, then reconstruct the consecutive missing data segments in the missing data to obtain the completed data segments. The completed dataset, the completed data segment, and the original dataset are integrated to obtain the repaired dataset; The repair dataset is normalized using a preset normalization method to obtain a normalized dataset; The normalized dataset and blood glucose value sequence are time-calibrated to obtain a preprocessed dataset; Feature extraction operations were performed on both the preprocessed data in the preprocessed dataset and the blood glucose values ​​in the blood glucose value sequence to obtain the original high-dimensional feature dataset. The process involves performing feature extraction operations on both the preprocessed data in the preprocessed dataset and the blood glucose values ​​in the blood glucose value sequence to obtain the original high-dimensional feature dataset, including: Extract the preprocessed physiological dataset and the preprocessed behavioral dataset from the preprocessed dataset; Temporal feature extraction is performed on the preprocessed physiological dataset to obtain a set of physiological feature vectors; The preprocessed behavioral dataset is segmented and aggregated to approximate the data, resulting in a set of behavioral feature vectors. Feature extraction is performed on the blood glucose value sequence to obtain a blood glucose feature vector set; The physiological feature vector set, behavioral feature vector set and blood glucose feature vector set are associated with the feature vector set along the time dimension to obtain the original high-dimensional feature dataset. The original high-dimensional feature dataset is fused to obtain a multimodal feature vector set; A pre-built time series analysis model is used to evaluate the correlation between blood glucose values ​​and multimodal feature vector sets, thereby obtaining feature importance evaluation values. Based on the feature importance evaluation values, a set of weights for the influence of blood glucose values ​​is obtained. The multimodal feature vector set and the weight set of blood glucose value influence are weighted and fused to obtain a weighted vector set; The weighted vector set is divided into a training set and a validation set. The training set is modeled using a pre-built time series modeling method to obtain a preliminary blood glucose prediction model. The training set accounts for 70% of the weighted vector set, and the validation set accounts for 30% of the weighted vector set. The time corresponding to the validation set is later than that of the training set. The preliminary blood glucose prediction model is validated using a validation set to obtain validation evaluation results. Based on the validation evaluation results, the preliminary blood glucose prediction model is optimized to obtain an optimized blood glucose prediction model. The optimized blood glucose prediction model is used to predict blood glucose levels in patients to obtain predicted blood glucose values. Blood glucose risk warning reports are obtained using the predicted blood glucose values ​​of patients and preset blood glucose thresholds, thereby achieving dynamic prediction of blood glucose.

2. The method for dynamic blood glucose prediction by integrating physiological and behavioral data as described in claim 1, characterized in that, The process of collecting raw patient datasets using pre-verified wearable devices and preset collection time intervals, and obtaining blood glucose value sequences using pre-verified blood glucose monitoring devices and collection time intervals, includes: The patient's raw dataset is collected using wearable devices and at specific time intervals. The raw dataset includes multiple raw data sets, which include physiological and behavioral data. Blood glucose values ​​of patients are obtained using blood glucose monitoring equipment and at specific time intervals. The obtained blood glucose values ​​are then arranged in chronological order to obtain a blood glucose value sequence.

3. The method for dynamic blood glucose prediction by integrating physiological and behavioral data as described in claim 2, characterized in that, The process of fusing the original high-dimensional feature dataset to obtain a multimodal feature vector set includes: Multimodal feature extraction was performed on the original high-dimensional feature dataset to obtain blood glucose modality group, physiological modality group and behavioral modality group; A fusion feature attention matrix is ​​constructed using the blood glucose modality group as the reference query vector, the physiological modality group as the key vector, and the behavioral modality group as the value vector, respectively. Feature decoupling is performed on the fused feature attention matrix to obtain behavioral modality feature submatrices and physiological modality feature submatrices; The behavioral modality feature submatrix and the physiological modality feature submatrix are dimensionality reduced respectively to obtain the behavioral modality feature vector set and the physiological modality feature vector set; By summing the behavioral modality feature vector set and the physiological modality feature vector set, a multimodal feature vector set is obtained.

4. The method for dynamic blood glucose prediction by integrating physiological and behavioral data as described in claim 3, characterized in that, The process involves using a pre-built time-series analysis model to assess the correlation between blood glucose levels and multimodal feature vector sets, obtaining feature importance assessment values, and then obtaining a set of weights representing the influence of blood glucose levels based on these assessment values. This includes: The behavioral modality feature vector set and the physiological modality feature vector set in the multimodal feature vector set are time-aligned with the blood glucose value sequence to obtain the aligned feature vector set; A time series analysis model was trained using aligned feature vector sets to obtain a blood glucose assessment model; The importance assessment values ​​of behavioral characteristics and physiological characteristics were obtained using a blood glucose assessment model. A weight set for the influence of behavioral blood glucose values ​​is generated based on the numerical assessment of the importance of behavioral characteristics, and a weight set for the influence of physiological blood glucose values ​​is generated based on the numerical assessment of the importance of physiological characteristics. The weight set of behavioral blood glucose value influence is merged with the weight set of physiological blood glucose value influence to obtain the weight set of blood glucose value influence.

5. The method for dynamic blood glucose prediction by integrating physiological and behavioral data as described in claim 4, characterized in that, The process involves dividing the weighted vector set into a training set and a validation set, and then using a pre-built time series modeling method to model the training set to obtain a preliminary blood glucose prediction model, including: Arrange the weighted vector set in chronological order to obtain the weighted vector sequence; Based on the proportion of the training set in the weighted vector sequence and the proportion of the validation set in the weighted vector sequence, the split point between the training set and the validation set is determined. The weighted vectors in the weighted vector sequence before the split point are used as the training set, and the weighted vectors after the split point are used as the validation set. The multiple weighted vectors in the training set are normalized to obtain a normalized training set, which includes multiple normalized weighted vectors. The normalized weighted vectors in the normalized training set are divided by time step to obtain multiple time series samples. Each time series sample includes multiple normalized weighted vectors and multiple blood glucose values, and the normalized weighted vectors correspond one-to-one with the blood glucose values. A preliminary blood glucose prediction model was obtained by using time series modeling methods and multiple time series samples.

6. The method for dynamic blood glucose prediction by integrating physiological and behavioral data as described in claim 5, characterized in that, The validation of the preliminary blood glucose prediction model using a validation set yields validation evaluation results, including: Multiple weighted vectors from the validation set are input into the preliminary blood glucose prediction model to obtain a set of predicted blood glucose values. The average prediction error is calculated using the predicted blood glucose values ​​from the predicted blood glucose value set, the actual blood glucose values ​​corresponding to the weighted vector, and a pre-constructed average absolute percentage error calculation formula. The average absolute percentage error calculation formula is as follows: ， in, Indicates the average prediction error. This represents the number of weighted vectors in the validation set. Represents the first in the verification set The predicted blood glucose value corresponding to each weighted vector. Represents the first in the verification set The actual blood glucose values ​​corresponding to each weighted vector; The verification and evaluation results are obtained by using the average prediction error and the preset error threshold. If the average prediction error is less than the error threshold, the verification and evaluation results are deemed qualified; if the average prediction error is greater than or equal to the error threshold, the verification and evaluation results are deemed unqualified.

7. The method for dynamic blood glucose prediction by integrating physiological and behavioral data as described in claim 6, characterized in that, The process involves optimizing the preliminary blood glucose prediction model based on the validation and evaluation results to obtain an optimized blood glucose prediction model. This optimized model is then used to predict the patient's blood glucose level, resulting in a predicted blood glucose value. Finally, a blood glucose risk warning report is generated using the predicted blood glucose value and a preset blood glucose threshold. After confirming that the verification and evaluation results are unqualified, the optimized weight parameters of the preliminary blood glucose prediction model are calculated using the pre-constructed backpropagation weight parameter calculation formula, which is shown below: ， in, This indicates the optimization of weight parameters. The weight parameters represent the initial blood glucose prediction model. For learning rate, This indicates the number of samples used in each update. Represents the first in the verification set The prediction error corresponding to each weighted vector , Represents the first in the verification set The predicted blood glucose value corresponding to each weighted vector. Represents the first in the verification set The actual blood glucose value corresponding to each weighted vector. This represents the gradient of the prediction error with respect to the weight parameters of the preliminary blood glucose prediction model. The preliminary blood glucose prediction model is adjusted using the weight parameters, and the adjusted preliminary blood glucose prediction model is used as the preliminary blood glucose prediction model. The process of verifying the preliminary blood glucose prediction model using the validation set and obtaining the verification evaluation result is repeated until the verification evaluation result is qualified. Then, the preliminary blood glucose prediction model corresponding to the weight parameters is confirmed as the optimized blood glucose prediction model. The optimized blood glucose prediction model is used to predict the patient's blood glucose level, and the predicted blood glucose value of the patient is obtained. The patient's predicted blood glucose value is compared with the blood glucose threshold. If the predicted blood glucose value is greater than or equal to the blood glucose threshold, a blood glucose risk warning report is obtained.

8. A system for dynamically predicting blood glucose using the method for fusing physiological and behavioral data as described in any one of claims 1 to 7, characterized in that, The system includes: The data acquisition module is used to collect the patient's raw dataset using a pre-confirmed wearable device and a preset acquisition time interval, and to obtain a blood glucose value sequence using a pre-confirmed blood glucose monitoring device and an acquisition time interval. The raw dataset includes multiple raw data, which includes physiological data and behavioral data, and the blood glucose value sequence includes multiple blood glucose values. The data processing module is used to preprocess the original dataset to obtain a preprocessed dataset, which includes multiple preprocessed data. Feature extraction operations are performed on the preprocessed data in the preprocessed dataset and the blood glucose values ​​in the blood glucose value sequence to obtain an original high-dimensional feature dataset. The original high-dimensional feature dataset is fused to obtain a multimodal feature vector set. The correlation between the multimodal feature vector set and blood glucose values ​​is evaluated using a pre-built time series analysis model to obtain feature importance evaluation values. Based on the feature importance evaluation values, a set of blood glucose value influence weights is obtained. The multimodal feature vector set and the set of blood glucose value influence weights are weighted and fused to obtain a weighted vector set. The model creation module is used to divide the weighted vector set into a training set and a validation set. The training set is modeled using a pre-built time series modeling method to obtain a preliminary blood glucose prediction model. The training set accounts for 70% of the weighted vector set, and the validation set accounts for 30% of the weighted vector set. The time corresponding to the validation set is later than that of the training set. The validation practice module is used to validate the preliminary blood glucose prediction model using a validation set, obtain validation evaluation results, optimize the preliminary blood glucose prediction model based on the validation evaluation results, obtain an optimized blood glucose prediction model, use the optimized blood glucose prediction model to predict blood glucose for patients, obtain predicted blood glucose values ​​for patients, and obtain blood glucose risk warning reports using predicted blood glucose values ​​for patients and preset blood glucose thresholds, thereby realizing dynamic prediction of blood glucose.

Images