Intelligent compensation methods, systems, and storage media for blood glucose detection

Abstract

This invention belongs to the field of medical and health information technology, and discloses an intelligent compensation method, system, and storage medium for blood glucose detection. The method collects human acceleration and heart rate signals, fuses them to generate a comprehensive quantitative index of exercise intensity; compares the index with a threshold to determine the exercise state; extracts historical blood glucose curves for high-intensity states to determine a preliminary lag time estimation range; dynamically adjusts the preliminary range using real-time acceleration frequency fluctuations to generate a corrected lag time interval; combines real-time heart rate change rate, skin temperature, and historical blood glucose cycle trends to generate a dynamic blood glucose concentration prediction sequence; judges abnormal blood glucose states based on the distribution pattern and deviation characteristics of the prediction sequence; analyzes the correlation between abnormal states and exercise intensity indicators, adjusts the lag time estimation range and / or prediction frequency, and generates a compensation scheme. This invention achieves dynamic correction of blood glucose lag time in exercise scenarios, improving prediction accuracy and individual adaptability.

Description

Technical Field

[0001] This invention belongs to the field of medical and health information technology, and specifically relates to an intelligent compensation method, system and storage medium for blood glucose detection. Background Technology

[0002] Blood glucose monitoring and prediction technology is a core component of diabetes management, playing a crucial role in guiding patients' diet, medication, and exercise. With the development of wearable devices and artificial intelligence technology, methods for predicting blood glucose changes based on physiological parameters have become a research hotspot, especially in exercise scenarios. Accurately predicting blood glucose trends can help patients prevent hypoglycemia or hyperglycemia events, improving exercise safety.

[0003] Traditional blood glucose monitoring mainly relies on finger-prick blood sampling or continuous glucose monitoring (CGM). The former is a single-point measurement and cannot provide information on continuous changes; the latter, while providing a continuous blood glucose concentration curve, is "post-monitoring" and cannot predict future blood glucose changes. To address this issue, researchers have begun to explore using physiological parameters such as heart rate, acceleration, and skin conductance response, through machine learning or physiological models, to predict blood glucose concentration in advance.

[0004] In existing technologies, prediction models typically employ a fixed time delay parameter, assuming a constant lag between exercise intensity and blood glucose changes. For example, some studies have fixed the lag time for post-exercise blood glucose reduction at 15 minutes and built prediction models based on this. However, clinical observations show significant individual differences in the impact of exercise on blood glucose, and the lag time also dynamically changes for the same user under different exercise intensities and rhythms. Prediction methods with fixed lag times exhibit significantly increased prediction errors in scenarios involving high-intensity intermittent exercise or frequent fluctuations in exercise intensity, making them unsuitable for practical applications.

[0005] In summary, how to dynamically and continuously determine the lag time interval of blood glucose changes in a sports scenario based on real-time changes in exercise intensity and physiological parameters, and generate a high-precision blood glucose prediction sequence and adaptive compensation scheme based on this, is a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0006] This invention provides an intelligent compensation method, system, and storage medium for blood glucose detection, aiming to solve the technical problems mentioned in the background art.

[0007] In a first aspect, the present invention provides an intelligent compensation method for blood glucose detection, comprising:

[0008] Step S1: Collect human acceleration data and heart rate signals, and fuse them to generate a comprehensive quantitative index of exercise intensity;

[0009] Step S2: Compare the comprehensive quantitative index of exercise intensity with the preset threshold to determine whether it is a high-intensity exercise state or a low-intensity exercise state, and extract historical blood glucose curve reference data for the high-intensity exercise state to determine the preliminary lag time estimation range of blood glucose change.

[0010] Step S3: Combining the frequency fluctuations of the real-time collected acceleration data as correction data, dynamically offset the range of the preliminary lag time estimation. The dynamic offset adjustment includes: when the amplitude of the frequency fluctuation shows a sharp change, selectively shortening or extending the lower limit and upper limit of the preliminary lag time estimation range according to the direction and amplitude of the change, to generate a corrected blood glucose change lag time interval.

[0011] Step S4: Using the corrected blood glucose change lag time interval, combined with the real-time collected heart rate change rate and skin temperature as current physiological parameters, and historical blood glucose periodic change trend data, a dynamic blood glucose concentration prediction sequence is generated through time series analysis.

[0012] Step S5: Determine the abnormal state of blood glucose changes based on the distribution pattern and deviation characteristics of the dynamic blood glucose concentration prediction sequence;

[0013] Step S6: Analyze the correlation between the abnormal blood glucose change and the comprehensive quantitative index of exercise intensity, adjust the lag time estimation range and / or prediction frequency according to the correlation results, and generate a blood glucose lag compensation scheme that includes the adjusted lag time estimation range and / or prediction frequency.

[0014] Further, step S1 specifically includes:

[0015] Human acceleration data and heart rate signals are collected by wearable devices, and the acceleration data is processed by frequency domain analysis to extract the temporal fluctuation characteristics and frequency amplitude changes of the motion.

[0016] The heart rate signal is analyzed to extract the average heart rate, heart rate variability, and heart rate rise or fall rate as physiological parameter change characteristics of the heart rate signal.

[0017] By using a multidimensional mapping method, the temporal fluctuation characteristics, the frequency amplitude changes, and the physiological parameter changes of the heart rate signal are fused together to calculate the comprehensive quantitative index of exercise intensity.

[0018] Further, step S2 specifically includes:

[0019] Obtain the comprehensive quantitative index of exercise intensity and compare it with the preset exercise intensity threshold;

[0020] If the comprehensive quantitative index of exercise intensity exceeds the preset exercise intensity threshold, it is determined to be a high-intensity exercise state; otherwise, it is a low-intensity exercise state. Based on the spectral energy distribution characteristics of real-time acceleration data, the current exercise metabolic mode is identified as either a continuous aerobic mode or an intermittent anaerobic mode.

[0021] For the high-intensity exercise state, based on the identified exercise metabolic pattern, reference data is extracted from the corresponding historical blood glucose curve sub-database, the response delay interval is analyzed, and the preliminary lag time estimation range of blood glucose changes is determined.

[0022] For low-intensity motion states, a preset default lag time interval is used as the initial lag time estimation range.

[0023] Furthermore, in step S3, the preliminary lag time estimation range is dynamically offset and adjusted, specifically including:

[0024] The frequency fluctuations of the acceleration data collected in real time are used as correction data;

[0025] If the amplitude of the frequency fluctuation shows a sharp change, the lower and upper limits of the preliminary lag time estimation range are shortened or extended according to the direction and amplitude of the change, generating a corrected blood glucose change lag time interval; wherein, the formula for calculating the offset is: offset = base value × amplitude change rate, the base value is set to one-quarter of the total length of the preliminary lag time estimation range, the amplitude change rate is the ratio of the current frequency amplitude change to the preset sharp change threshold, and the offset does not exceed one-half of the total length of the preliminary lag time estimation range.

[0026] Further, in step S4, the generation of a dynamic blood glucose concentration prediction sequence through time-series analysis specifically includes:

[0027] Based on the corrected blood glucose change lag time interval [L, U], the prediction time axis is divided into a metabolic response segment and a steady-state trend segment. The metabolic response segment corresponds to the time points within the interval [L, U], and the steady-state trend segment corresponds to the time points within the interval (U, U+ΔT], where ΔT is the preset prediction duration.

[0028] In the metabolic response phase, a weighted average method is used to predict short-period fluctuations in the heart rate change rate, skin temperature, and the periodic change trend data at continuous time points, generating the first subsequence;

[0029] In the steady-state trend segment, long-term baseline prediction is performed based on the low-frequency characteristics of historical blood glucose cyclical change trend data to generate a second subsequence;

[0030] The first subsequence and the second subsequence are concatenated and coupled along the time axis to generate a dynamic prediction sequence containing predicted blood glucose concentration values ​​for multiple future time points. Further, step S5 specifically includes:

[0031] The distribution patterns of peaks and troughs, as well as the abnormal deviations between predicted values ​​and historical normal values, are extracted from the dynamic blood glucose concentration prediction sequence.

[0032] If the distribution pattern of the peaks and troughs deviates from the preset normal range, or if the abnormal deviation exceeds the preset safety limit, then an abnormal trend correlation analysis is triggered.

[0033] Based on the results of the abnormal trend correlation analysis, the type and cause of abnormal blood glucose changes are determined.

[0034] Furthermore, in step S6, the blood glucose lag compensation scheme specifically includes:

[0035] Based on the correlation between the abnormal blood glucose levels and the comprehensive quantitative index of exercise intensity, the compensation interval division strategy is adjusted.

[0036] The personalized correction parameters are updated by adjusting the compensation interval division strategy; the personalized correction parameters include: the weight of the influence of exercise intensity on lag time, the upper limit adjustment coefficient and the lower limit adjustment coefficient of the lag time estimation range;

[0037] Based on the updated personalized correction parameters, the adjusted lag time estimation range and / or prediction frequency are determined, and a blood glucose lag compensation scheme containing the adjusted lag time estimation range and / or prediction frequency is generated.

[0038] Furthermore, it also includes step S7: using the adjusted lag time estimation range and prediction frequency determined in the blood glucose lag compensation scheme, the subsequently collected acceleration data and heart rate signals are reprocessed to predict blood glucose changes, and the actual blood glucose values ​​of the same period are obtained, the prediction accuracy is calibrated, and the final blood glucose change lag time adjustment value is determined.

[0039] Secondly, the present invention provides an intelligent compensation system for blood glucose detection, comprising:

[0040] The data acquisition module is used to collect human acceleration data and heart rate signals;

[0041] The intensity quantification module is used to integrate and generate a comprehensive quantitative index of exercise intensity.

[0042] The state determination module is used to compare the comprehensive quantitative index of exercise intensity with a preset threshold to determine whether it is a high-intensity exercise state or a low-intensity exercise state; based on the spectral energy distribution characteristics of real-time acceleration data, it identifies the current exercise metabolic mode; for the high-intensity exercise state, it extracts reference data from the corresponding historical blood glucose curve sub-library according to the identified exercise metabolic mode to determine the preliminary lag time estimation range of blood glucose change.

[0043] The hysteresis correction module is used to combine the frequency fluctuations of the real-time collected acceleration data as correction data to dynamically offset the range of the preliminary hysteresis time estimation. The dynamic offset adjustment includes: when the amplitude of the frequency fluctuation is determined to change drastically, selectively shortening or extending the lower limit and upper limit of the preliminary hysteresis time estimation range according to the direction and amplitude of the change, so as to generate a corrected hysteresis time interval for blood glucose changes.

[0044] The prediction sequence generation module is used to generate a dynamic prediction sequence of blood glucose concentration by using the corrected blood glucose change lag time interval, combined with the real-time collected heart rate change rate and skin temperature as current physiological parameters, and historical blood glucose periodic change trend data.

[0045] An anomaly detection module is used to determine abnormal blood glucose changes based on the distribution pattern and deviation characteristics of the dynamic blood glucose concentration prediction sequence.

[0046] The compensation scheme generation module is used to analyze the correlation between the abnormal blood glucose change state and the comprehensive quantitative index of exercise intensity, adjust the lag time estimation range and / or prediction frequency according to the correlation results, and generate a blood glucose lag compensation scheme that includes the adjusted lag time estimation range and / or prediction frequency.

[0047] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the above-described intelligent compensation method for blood glucose detection.

[0048] This invention offers the following advantages: First, by fusing acceleration and heart rate signals to generate a comprehensive quantitative index of exercise intensity, and identifying the exercise metabolic mode (continuous aerobic mode or intermittent anaerobic mode) based on the energy distribution characteristics of the acceleration spectrum, it matches differentiated preliminary lag time estimation ranges from the corresponding historical blood glucose curve sub-library. This overcomes the shortcomings of traditional methods that rely solely on exercise intensity and ignore differences in metabolic modes, significantly improving the accuracy of initial lag time estimation in complex scenarios such as high-intensity intermittent exercise. Second, by dynamically adjusting the preliminary lag time estimation range using real-time acceleration frequency fluctuations, and employing the offset calculation formula (offset = base value × amplitude change rate) and setting an upper limit for the offset, the lag time interval correction is proportional to the degree of exercise mutation and does not exceed the physiologically reasonable range. This solves the problem that fixed or segmented constant lag times cannot adapt to continuous changes in exercise intensity, further improving prediction timeliness. Third, based on the corrected lag time interval, the prediction time axis is divided into a metabolic response segment and a steady-state trend segment. In the metabolic response segment, a weighted average method is used to fuse heart rate change rate, skin temperature, and historical trends for short-cycle fluctuation prediction. In the steady-state trend segment, long-cycle baseline prediction is performed based on the low-frequency characteristics of historical periodic change trends. The two segments are smoothly spliced ​​together to generate a dynamic prediction sequence, achieving segmented coupled prediction at different physiological stages, balancing high sensitivity in the rapid change segment with low noise stability in the steady-state segment. Fourth, a correlation analysis is performed between abnormal blood glucose changes and comprehensive quantitative indicators of exercise intensity. Based on the correlation results, the lag time estimation range and prediction frequency are dynamically adjusted to generate an adaptive blood glucose lag compensation scheme. Through closed-loop calibration with subsequent data collection and actual blood glucose values, a complete closed-loop mechanism of "individual metabolic pattern matching—dynamic offset correction—segmented coupled prediction—abnormal correlation compensation—online iterative optimization" is formed, achieving highly personalized blood glucose prediction. Fifth, by integrating multiple physiological parameters such as heart rate change rate and skin temperature with historical cycle trends, the predicted sequence reflects both long-term individual patterns and real-time metabolic changes. The frequency domain analysis, time series analysis, and segmented prediction strategies employed have moderate computational complexity and can run in real time on wearable devices such as smartwatches, thus possessing broad industrial application prospects. Attached Figure Description

[0049] Figure 1 A flowchart of an intelligent compensation method for blood glucose detection provided by the present invention;

[0050] Figure 2 A comparison chart showing the prediction performance of an intelligent compensation method for blood glucose detection provided by this invention;

[0051] Figure 3 This invention provides a structural block diagram of an intelligent compensation system for blood glucose detection. Detailed Implementation

[0052] To further understand the content of this invention, a detailed description of the invention is provided in conjunction with the accompanying drawings and embodiments. The specific embodiments described herein are for illustrative purposes only and are not intended to limit the invention. It should also be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings.

[0053] The specific application scenario of this invention is applicable to wearable devices such as smartwatches, for the dynamic prediction and adaptive compensation of changes in a user's blood glucose concentration during exercise.

[0054] Firstly, such as Figure 1 This embodiment provides an intelligent compensation method for blood glucose detection, including:

[0055] Step S1: Collect human acceleration data and heart rate signals, and fuse them to generate a comprehensive quantitative index of exercise intensity.

[0056] Specifically, step S1 includes: collecting human acceleration data and heart rate signals through a wearable device; processing the acceleration data using frequency domain analysis to extract the temporal fluctuation characteristics and frequency amplitude changes of the motion; analyzing the heart rate signal to extract the average heart rate, heart rate variability, and heart rate rise or fall rate as physiological parameter change characteristics of the heart rate signal; and using a multidimensional mapping method to fuse the temporal fluctuation characteristics, the frequency amplitude changes, and the physiological parameter change characteristics of the heart rate signal to calculate the comprehensive quantitative index of exercise intensity.

[0057] Specifically, step S1 involves collecting human acceleration data and heart rate signals through wearable devices and fusing them to generate a comprehensive quantitative index of exercise intensity. The user's smartwatch or wristband has a built-in three-axis accelerometer and photoelectric heart rate sensor, continuously acquiring acceleration components and heart rate values ​​at a sampling frequency of, for example, 10 Hz. To eliminate noise introduced by device vibration or poor contact, the raw acceleration data is bandpass filtered, retaining the frequency band related to human movement within the range of 0.5 Hz to 5 Hz. This frequency band covers the main frequency components of typical walking, running, and sprinting movements. Subsequently, frequency domain analysis technology is used to divide the filtered acceleration data into time windows, each window lasting 4 seconds and overlapping 50% with adjacent windows. A fast Fourier transform is then used to convert the time-domain signal into a frequency-domain representation. Two features are extracted from the frequency domain spectrum: temporal fluctuation characteristics, i.e., the sequence of energy peak values ​​corresponding to the main frequency over time, reflecting the stability of the exercise rhythm; and frequency amplitude changes, i.e., the amplitude of the main frequency and its changes between consecutive windows, characterizing the instantaneous increase or decrease in exercise intensity. For example, when a user switches from jogging to sprinting, the main frequency jumps from 1.5 Hz to 2.3 Hz, the amplitude value increases synchronously, and the frequency amplitude change characteristic value produces a positive jump.

[0058] The processing of heart rate signals first involves removing motion artifacts using a sliding median filter, followed by calculating the heart rate sequence per minute. From this sequence, three physiological parameters are extracted: the mean heart rate, representing the basal metabolic rate at rest or during exercise; the heart rate variability rate, i.e., the standard deviation of consecutive heartbeat intervals, reflecting the strength of autonomic nervous system regulation; and the rate of increase or decrease in heart rate, referring to the slope of the change in heart rate value per unit time (e.g., 30 seconds), used to capture the physiological response to sudden changes in exercise load. For example, within two minutes of starting a run, if the heart rate increases from 75 beats per minute to 115 beats per minute, with an increase rate of approximately 20 beats per minute, it indicates the immediate adaptation of the cardiovascular system to exercise intensity.

[0059] After obtaining the temporal fluctuation characteristics of the acceleration domain, the frequency amplitude changes, and the three physiological parameter changes of the heart rate domain, a multidimensional mapping method is used for fusion. Specifically, the multidimensional mapping method involves: performing min-max normalization on the five feature values ​​to unify their range to the [0,1] interval. The upper and lower limits of normalization for each feature are pre-set based on the 95% confidence interval of population statistical data. For example, five feature values ​​can be collected from no fewer than 100 subjects, outliers are removed, and the mean μ and standard deviation σ are calculated. The 95% confidence interval is [μ-1.96σ, μ+1.96σ], serving as the upper and lower limits of normalization. Real-time feature values ​​below the lower limit are normalized to zero, and values ​​above the upper limit are normalized to one, with linear interpolation within the interval. The normalized feature vectors are then weighted and summed to calculate a comprehensive quantitative index of exercise intensity. The weight ratio of the acceleration domain to the heart rate domain is 3:2, determined based on research on the correlation between exercise intensity and cardiac output in exercise physiology. The weighted summation value is the comprehensive quantitative index of exercise intensity, ranging between 0 and 1. A value closer to 1 indicates higher exercise intensity, while a value closer to 0 indicates resting or low-intensity activity. This index serves as the direct basis for determining the exercise state in subsequent step S2, unifying the two physically different original signals—acceleration and heart rate—into a standardized intensity measure, laying the quantitative foundation for threshold comparison and lag time estimation. For example, when a user runs at 12 km / h, the acceleration frequency amplitude reaches 0.7, and the heart rate rises at 18 beats per minute. The fused index is 0.78, far exceeding the preset threshold of 0.6. Based on this, it can be determined as a high-intensity exercise state, thereby triggering the extraction of historical blood glucose curve reference data and the determination of the preliminary lag time estimation range. Thus, the algorithm design of frequency domain analysis, physiological parameter extraction, and multidimensional mapping in step S1 is not isolated data processing, but rather mutually supportive and interactive with subsequent technical features such as exercise state determination and dynamic lag time correction, ensuring the foundation for accurate and continuous quantification of exercise intensity in complex exercise scenarios.

[0060] Step S2: Compare the comprehensive quantitative index of exercise intensity with the preset threshold to determine whether it is a high-intensity exercise state or a low-intensity exercise state, and extract historical blood glucose curve reference data for the high-intensity exercise state to determine the preliminary lag time estimation range of blood glucose change.

[0061] Specifically, step S2 includes: acquiring the comprehensive quantitative index of exercise intensity and comparing it with a preset exercise intensity threshold; if the comprehensive quantitative index of exercise intensity exceeds the preset exercise intensity threshold, it is determined to be a high-intensity exercise state, otherwise it is a low-intensity exercise state; based on the spectral energy distribution characteristics of real-time acceleration data, identifying the current exercise metabolic mode as a continuous aerobic mode or an intermittent anaerobic mode; for the high-intensity exercise state, according to the identified exercise metabolic mode, extracting reference data from the corresponding historical blood glucose curve sub-database, analyzing the response delay interval, and determining the preliminary lag time estimation range of blood glucose changes; for the low-intensity exercise state, using a preset default lag time interval as the preliminary lag time estimation range.

[0062] More specifically, step S2 determines the exercise state based on the comparison between the comprehensive quantitative index of exercise intensity and the preset threshold, and determines the preliminary lag time estimation range of blood glucose changes for different exercise states. The comprehensive quantitative index of exercise intensity is calculated in step S1, and its value is between 0 and 1, with higher values ​​indicating greater exercise intensity. The preset exercise intensity threshold is set to 0.6, which is determined based on the correlation statistics between exercise intensity and blood glucose lag time in large-scale population tests. When the index exceeds 0.6, the human metabolic rate increases significantly, the rate of glucose uptake by muscles accelerates, and the lag time of blood glucose changes shows a significant shortening trend. After obtaining the comprehensive quantitative index of exercise intensity output in step S1, the system compares it with 0.6: if the index is greater than 0.6, it is determined to be a high-intensity exercise state; if the index is less than or equal to 0.6, it is determined to be a low-intensity exercise state. For example, if a user runs at a speed of 12 kilometers per hour, the comprehensive quantitative index of exercise intensity is 0.78, which is greater than 0.6, and the system determines it to be a high-intensity exercise state; if a user walks at a speed of 4 kilometers per hour, the index is 0.35, and it is determined to be a low-intensity exercise state.

[0063] For high-intensity exercise, the system first identifies the exercise metabolic pattern using the spectral energy distribution characteristics of real-time acceleration data. In step S1, the acceleration spectrum for each time window (e.g., 4 seconds in length, 50% overlap) has been obtained through frequency domain analysis. The system extracts the energy proportions of the low-frequency band (0.5-1.5Hz), the dominant frequency band (1.5-2.5Hz), and the high-frequency band (2.5-4Hz) from the spectrum. The energy concentration index E is defined as the ratio of dominant frequency band energy to total energy. If E ≥ 0.7 and the high-frequency band energy proportion is less than 0.15, it is determined to be a continuous aerobic mode (e.g., steady-paced running, long-distance jogging); if the energy ratio of the high-frequency band to the low-frequency band exceeds 1.2 and the dominant frequency band energy proportion is less than 0.5, it is determined to be an intermittent anaerobic mode (e.g., short-distance sprinting, high-intensity interval training). This identification process runs continuously, outputting a label for the current exercise metabolic pattern every 10 seconds. It should be noted that the continuous aerobic mode includes exercises primarily based on aerobic metabolism, such as steady-paced running, long-distance jogging, and cycling; the intermittent anaerobic mode includes exercises primarily based on anaerobic metabolism, such as short-distance sprints and high-intensity interval training. For mixed-mode exercises with complex spectral characteristics, such as strength training and ball sports, the system determines the mode by calculating the variance and kurtosis of the spectral energy distribution: if the energy variance exceeds a preset threshold and the spectrum exhibits a multi-peak distribution, it is determined to be a mixed mode, and an interpolation lag time interval between aerobic and anaerobic modes is used (for example, a weighted average of the aerobic mode interval and the anaerobic mode interval). For low-intensity recovery activities such as yoga and stretching, since their exercise intensity index is usually below 0.3, they have already been classified as low-intensity exercise and are treated using a default lag time interval, without the need for additional metabolic mode identification.

[0064] The system stores users' historical blood glucose curve data in a local or cloud database, categorized into two sub-databases based on exercise metabolic patterns: an aerobic sub-database and an anaerobic sub-database. Each historical curve records the blood glucose concentration value and corresponding timestamp from 15 minutes before exercise to 60 minutes after exercise, and is labeled with the metabolic pattern tag during exercise. Based on the real-time identified pattern, the system extracts reference data of blood glucose curves from the corresponding sub-database for at least five previous exercises of the same or similar intensity for that user. After time alignment and smoothing of these curves, the response delay interval is analyzed. The response delay interval is defined as the time range from the start of exercise to the first sustained decrease in blood glucose concentration, where sustained decrease refers to a rate of decrease in blood glucose values ​​exceeding 0.1 mmol / L per minute for three consecutive sampling points. The calculated delay times from all historical events are sorted in ascending order, with the 25th percentile as the lower limit and the 75th percentile as the upper limit, forming a preliminary lag time estimation range. Taking a type 1 diabetic patient as an example, historical data from their five sprint exercises in intermittent anaerobic mode showed that blood glucose levels began to decrease at 6, 7, 8, 9, and 7 minutes after exercise, with a calculated response delay range of [7, 9] minutes. However, data from their five steady-state runs in continuous aerobic mode showed that blood glucose levels decreased at 9, 10, 12, 11, and 13 minutes, with a range of [10, 12] minutes. The system automatically selects the corresponding range based on the currently identified metabolic mode, making the initial estimate range more consistent with the physiological response patterns under different exercise metabolisms.

[0065] For low-intensity exercise, due to the small fluctuations in blood glucose levels, long lag times, and significant individual differences, the system does not rely on historical data but instead uses a preset default lag time interval as the initial lag time estimation range. The default interval is set based on clinical experience: [10, 20] minutes for type 1 diabetes users, [12, 25] minutes for type 2 diabetes users, and [12, 20] minutes for healthy users or those in a cold start state. For example, a type 2 diabetes patient newly using the system, exercising at a slow pace while cycling, with a blood glucose level of 0.45, is classified as low-intensity, and the default interval [12, 25] minutes is directly used. It should be noted that while there is a lag time in low-intensity exercise, its engineering predictive value is low because blood glucose fluctuations are small, and the lag time is long and stable. The purpose of the default lag time interval is to ensure the integrity of the process, ensuring that subsequent steps can be executed in all exercise states. This embodiment mainly focuses on accurate prediction and compensation in high-intensity exercise.

[0066] The threshold comparison, metabolic pattern recognition, and dual-path initial value determination logic in step S2 transforms continuous exercise intensity indicators into discrete exercise state classifications, which are further refined into metabolic pattern subclasses. This makes the initial lag time estimation range both adaptable to exercise intensity stratification and adds metabolic pattern specificity. This technical feature solves the problem in existing technologies that cannot distinguish the differences in blood glucose lag time under different exercise metabolic patterns. It supports and interacts with the frequency domain analysis in step S1 and the dynamic offset adjustment function in step S3, together forming a complete means from exercise feature extraction to personalized lag time initialization. Its beneficial effect is that it significantly improves the accuracy of the initial estimation range in anaerobic-dominated scenarios such as high-intensity interval exercise, reduces the pressure of subsequent dynamic correction, and lays a reliable foundation for accurate prediction in step S4.

[0067] Step S3: Combining the frequency fluctuations of the real-time collected acceleration data as correction data, the preliminary lag time estimation range is dynamically offset and adjusted. The dynamic offset adjustment includes: when the amplitude of the frequency fluctuation is determined to change drastically, the lower limit and upper limit of the preliminary lag time estimation range are selectively shortened or extended according to the direction and amplitude of the change, so as to generate a corrected blood glucose change lag time interval.

[0068] In step S3, the preliminary lag time estimation range is dynamically offset and adjusted. Specifically, this includes: using the frequency fluctuation of the real-time collected acceleration data as correction data; if the amplitude of the frequency fluctuation shows a sharp change, then shortening or extending the lower and upper limits of the preliminary lag time estimation range according to the direction and amplitude of the change, generating the corrected blood glucose change lag time interval; wherein, the offset is calculated using the formula: offset = base value × amplitude change rate, the base value is set to one-quarter of the total length of the preliminary lag time estimation range, the amplitude change rate is the ratio of the current frequency amplitude change to the preset sharp change threshold, and the offset does not exceed one-half of the total length of the preliminary lag time estimation range.

[0069] Specifically, step S3 uses the frequency fluctuations of the real-time collected acceleration data as correction data to dynamically adjust the preliminary lag time estimation range determined in step S2, generating a corrected lag time interval for blood glucose changes. During exercise, the frequency fluctuations of the acceleration signal directly reflect changes in the exercise rhythm: during stable exercise, the dominant frequency and its amplitude remain relatively constant, while when the exercise rhythm changes abruptly, such as from jogging to sprinting or from acceleration to stopping, the dominant frequency and amplitude will rapidly change within several time windows. In step S1, the acceleration frequency amplitude value for each time window (4 seconds in length, 50% overlap) has been obtained through frequency domain analysis. This value sequence is updated in real time and temporarily stored in a circular buffer. Step S3 monitors this sequence in real time, calculates the difference between the frequency amplitude of the current window and the frequency amplitude of the previous window, and takes the average of the differences of three consecutive windows as a measure of the change amplitude. When this change amplitude measure exceeds a preset abrupt change threshold (e.g., 0.15, the normalized amplitude value), it is determined that the amplitude of the frequency fluctuation shows an abrupt change.

[0070] The direction of abrupt changes is determined by the sign of the difference: a positive difference indicates an increased amplitude, corresponding to a sudden increase in exercise intensity; a negative difference indicates a decreased amplitude, corresponding to a sudden decrease in exercise intensity. Based on the direction of change, the system dynamically adjusts the lower and upper limits of the initial lag time estimation range: if exercise intensity suddenly increases, the rate of glucose uptake by muscles accelerates, shortening the lag time of blood glucose changes; therefore, the offset is subtracted from both the lower and upper limits of the initial lag time estimation range. If exercise intensity suddenly decreases, the lag time will lengthen, so the offset is added to both the lower and upper limits. The offset is calculated using the following rule: Offset = Base value × Amplitude change rate, where the base value is set to one-quarter of the total length of the initial lag time estimation range, and the amplitude change rate is the ratio of the current frequency amplitude change to a preset abrupt change threshold. Simultaneously, the offset does not exceed half the total length of the initial lag time estimation range to ensure that the adjusted interval does not deviate excessively from the physiologically reasonable range. The lower and upper limits of the adjusted interval are rounded to integer values, generating the corrected blood glucose change lag time interval.

[0071] Taking a user performing high-intensity interval running as an example, step S2 determines the initial lag time estimation range of [10, 13] minutes based on their historical blood glucose curve (intermittent anaerobic mode sub-library). The real-time acceleration frequency amplitude is stable at around 0.7 for the first 30 seconds, then the user suddenly accelerates, and the frequency amplitude jumps to 0.92 in two consecutive windows, with a change amplitude of 0.22, exceeding the abrupt change threshold of 0.15. The change direction is positive, indicating a sudden increase in exercise intensity. The total length of the initial lag time estimation range is 3 minutes, with a base value set at 0.75 minutes. The amplitude change rate is 0.22 / 0.15≈1.47, and the offset is approximately 1.1 minutes. However, it is limited to not exceeding half of the total length (i.e., 1.5 minutes), so the actual offset is taken as 1.1 minutes. Therefore, the lower limit is reduced from 10 minutes to 9 minutes, and the upper limit is reduced from 13 minutes to 12 minutes, generating the corrected lag time interval [9, 12] minutes. If a user suddenly decelerates from a sprint to a slow walk, with the frequency amplitude decreasing from 0.92 to 0.58 (a change of 0.34 in negative direction), the offset is the base value of 0.75 multiplied by the amplitude change rate (0.34 / 0.15≈2.27). Due to the 1.5-minute upper limit, the actual offset is taken as 1.5 minutes. After adjustment, the upper and lower limits of the interval are each increased by 1.5 minutes, resulting in [11.5, 14.5] minutes, which is rounded to [12, 15] minutes.

[0072] The dynamic offset adjustment in step S3 directly maps the real-time motion rhythm change data—acceleration frequency fluctuations—to a correction amount for the lag time interval. The base value, amplitude change rate, and constraints in the offset calculation formula ensure that the correction amplitude is proportional to the degree of motion abrupt change and does not exceed a physiologically reasonable range. This technical feature solves the technical problem of premature or delayed prediction of fixed lag time when exercise intensity changes abruptly in the background technology. The corrected lag time interval serves as the time reference for generating the dynamic blood glucose concentration prediction sequence in step S4, enabling the prediction model to respond instantly to changes in exercise rhythm. It functionally supports and interacts with the preliminary estimation based on metabolic pattern recognition in step S2, jointly constituting a fine-grained dynamic calibration method for the lag time. The beneficial effect of this technical feature is that it significantly improves the timeliness and accuracy of blood glucose prediction in complex scenarios such as high-intensity interval exercise, and reduces prediction bias caused by lag time mismatch.

[0073] Step S4: Using the corrected blood glucose change lag time interval, combined with the real-time collected heart rate change rate and skin temperature as current physiological parameters, and historical blood glucose periodic change trend data, a dynamic blood glucose concentration prediction sequence is generated through time series analysis.

[0074] In step S4, the generation of a dynamic blood glucose concentration prediction sequence through time series analysis specifically includes: dividing the prediction time axis into a metabolic response segment and a steady-state trend segment based on the corrected blood glucose change lag time interval [L, U]. The metabolic response segment corresponds to the time points within the interval [L, U], and the steady-state trend segment corresponds to the time points within the interval (U, U+ΔT), where ΔT is a preset prediction duration. In the metabolic response segment, a weighted average method is used to predict short-cycle fluctuations in the heart rate change rate, skin temperature, and the periodic change trend data at continuous time points, generating a first sub-sequence. In the steady-state trend segment, a long-cycle baseline prediction is performed based on the low-frequency characteristics of historical blood glucose periodic change trend data, generating a second sub-sequence. The first sub-sequence and the second sub-sequence are spliced ​​and coupled along the time axis to generate a dynamic prediction sequence containing predicted blood glucose concentration values ​​for multiple future time points.

[0075] Specifically, step S4 utilizes the corrected blood glucose lag time interval generated in step S3, combined with real-time collected heart rate change rate, skin temperature, and historical blood glucose periodic change trend data, to generate a dynamic blood glucose concentration prediction sequence through multi-time-scale segmented prediction coupling. The heart rate change rate is calculated from the heart rate signal processed in step S1, representing the change in heart rate per minute over 30 consecutive seconds, measured in beats / minute / second, reflecting the immediate response intensity of the cardiovascular system to exercise load. Skin temperature is collected in real-time by a temperature sensor built into the wearable device, measured in degrees Celsius. Metabolic heat production during exercise causes skin temperature to rise, and its rate of change indirectly reflects the balance between muscle heat production and dissipation. Historical blood glucose periodic change trend data is extracted from the user's historical database, which stores typical blood glucose response curves according to time period (morning, afternoon, night) and exercise type (running, cycling).

[0076] The system first divides the prediction time axis into two stages based on the corrected lag time interval [L, U]: the time points within the interval [L, U] corresponding to the metabolic response segment (e.g., minutes 9 to 12), and the time points within the interval (U, U+ΔT) corresponding to the steady-state trend segment, where ΔT is the preset prediction duration (e.g., 30 minutes). In the metabolic response segment, a weighted average method is used to predict short-cycle fluctuations. The system uses the lag time interval length as the sliding window width, extracting the heart rate change rate sequence and skin temperature sequence within the time window before the current time point, and calculating the Pearson correlation coefficient r (absolute value, range 0-1) between these two sequences and the blood glucose change rate within the corresponding time period in the historical blood glucose periodic change trend curve. A higher correlation coefficient indicates a greater consistency between the current physiological parameter change pattern and the historical blood glucose response pattern. Simultaneously, a comprehensive quantitative index M of exercise intensity (range 0-1) is obtained. For each future time point t within the metabolic response segment (with a step size of 1 minute), the weighted average calculation formula for the predicted blood glucose value is constructed as follows:

[0077]

[0078] Where t is a future time point within the current time window (with a step size of 1 minute, starting from the lower limit L of the lag interval and ending at the upper limit U). The reference blood glucose value is obtained by multiplying the rate of blood glucose change at the corresponding time point in the historical periodic trend data by the time step and then adding the baseline value. The instantaneous blood glucose change estimate is obtained by mapping the current heart rate change rate and skin temperature change intensity; r is the correlation coefficient (absolute value, range 0~1) between the heart rate change rate, skin temperature and historical trend data calculated in step S4; M is the comprehensive quantitative index of exercise intensity (range 0~1). This weighted average method makes the contribution weights of historical trend and current physiological parameters proportional to r and M, respectively, so that the prediction results take into account both the individual's long-term pattern and real-time metabolic response. It should be noted that in the above formula, when r=0 and M=0, it means that the current physiological parameters are not related to the historical trend and the exercise intensity is 0. At this time, the prediction has no effective weight, and the system defaults to directly using the historical trend value. As a prediction result, avoid having a denominator of zero.

[0079] During the steady-state trend phase, because the immediate impact of heart rate change rate and skin temperature on blood glucose is significantly reduced, the system no longer relies on high-frequency physiological parameters. Instead, it performs long-term baseline prediction based on low-frequency characteristics of historical blood glucose cyclical change trend data. Specifically, the system selects the typical blood glucose change curve that best matches the current moment and exercise type from the historical database, extracts its average rate of change (e.g., change per minute) within the steady-state trend phase, and uses the predicted value at the end of the metabolic response phase as a starting point. This average rate is then linearly extrapolated to generate the second subsequence. For example, if a user is running in the afternoon and historical data shows that the rate of blood glucose recovery within the steady-state trend phase is 0.02 mmol / L per minute, then the predicted values ​​for subsequent points are calculated based on this rate, starting from the end of the metabolic response phase.

[0080] The system smoothly splices the first subsequence generated in the metabolic response segment and the second subsequence generated in the steady-state trend segment on the time axis: at the splicing point (i.e., at the upper limit U of the lag time interval), the weighted average of the endpoint value of the first subsequence and the starting value of the second subsequence is taken as the transition value to avoid jumps. Finally, a complete dynamic prediction sequence containing multiple future time points (e.g., one point per minute, for a total of 30 points) is generated. Taking a type 1 diabetic patient as an example, the corrected lag interval determined in step S3 is [9, 12] minutes, ΔT = 30 minutes, and the total predicted duration is 42 minutes. Within the metabolic response segment, the historical correlation coefficient r = 0.85, and the exercise intensity M = 0.78. The blood glucose level was shown to decrease at a rate of 0.08 mmol / L per minute. The data shows a decrease of 0.10 mmol / L per minute. Weighted predictions indicate that blood glucose will begin to decrease from 6.2 mmol / L at minute 9 and drop to 5.7 mmol / L at minute 12. Within the steady-state trend, the historical recovery rate is 0.02 mmol / L per minute, starting from 5.7 mmol / L and reaching 6.3 mmol / L at minute 42. These data are then assembled to form a complete predicted sequence, providing the data basis for step S5.

[0081] Step S4 divides the prediction time axis into a metabolic response segment and a steady-state trend segment, employing short-cycle weighted averaging and long-cycle baseline extrapolation respectively, fully utilizing the structural boundaries of the lag time interval. This technique addresses the technical problem of existing time-series prediction methods failing to distinguish physiological driving differences at different stages within the lag time interval, leading to overfitting of high-frequency noise in the steady-state segment. It mutually supports and interacts with the lag time interval output in step S3 and the motion intensity quantification and metabolic pattern recognition functions in steps S1 to S2, collectively forming a refined prediction method from interval division to segmented coupling. Its beneficial effects include maintaining high sensitivity during rapid blood glucose changes, reducing computational load and improving stability during the steady-state trend stage, significantly enhancing the physiological rationality and overall accuracy of the predicted sequence.

[0082] Step S5: Determine the abnormal state of blood glucose changes based on the distribution pattern and deviation characteristics of the dynamic blood glucose concentration prediction sequence.

[0083] Specifically, step S5 includes: extracting the distribution patterns of peaks and troughs from the dynamic blood glucose concentration prediction sequence, as well as the abnormal deviation magnitude between the predicted value and the historical normal value; if the distribution patterns of peaks and troughs deviate from the preset normal range, or the abnormal deviation magnitude exceeds the preset safety limit, then triggering an abnormal trend correlation analysis; and determining the type and cause of the abnormal blood glucose change state based on the results of the abnormal trend correlation analysis.

[0084] More specifically, step S5 determines the abnormal state of blood glucose changes based on the distribution pattern and deviation characteristics of the dynamic blood glucose concentration prediction sequence generated in step S4. The dynamic blood glucose concentration prediction sequence consists of multiple prediction points with a step size of minutes, each prediction point corresponding to a blood glucose concentration value at a future time. The system extracts two features from this sequence: first, the distribution pattern of peaks and troughs; and second, the abnormal deviation magnitude between the predicted value and historical normal values. Peak and trough detection uses the sliding window extreme value method, that is, identifying local maximum values ​​(peaks) and local minimum values ​​(troughs) in the sequence, and recording the time position and value of each peak and trough. The distribution pattern specifically includes the time interval from peak to trough (duration of decline), the time interval from trough to peak (duration of recovery), and the blood glucose change magnitude between adjacent peaks and troughs. The preset normal range is determined based on the user's historical blood glucose response statistics under the same exercise intensity over the past thirty days. For example, the normal range for the duration of decline is 8 to 25 minutes, the normal range for the magnitude of decline is 0.5 to 2.5 mmol / L, and the normal range for the duration of recovery is 10 to 40 minutes. If the duration of the current predicted sequence decreases to 5 minutes and the decrease exceeds 3.0 mmol / L, or the duration of the rebound increases to more than 60 minutes, then the peak-valley distribution pattern is determined to deviate from the preset normal range.

[0085] Abnormal deviation amplitude refers to the difference between the predicted blood glucose value at each time point in the prediction sequence and the historical normal value. The historical normal value is the median or mean of the user's blood glucose concentration during the same time period and under similar exercise conditions. The system calculates the deviation amplitude for all time points in the prediction sequence and counts the duration of consecutive deviations and the maximum deviation value. The preset safety limits are typically set as a low blood glucose threshold (3.9 mmol / L) and a high blood glucose threshold (13.9 mmol / L), as well as a rate of change safety limit (e.g., a decrease of more than 0.3 mmol / L per minute). When the blood glucose value at any prediction point is lower than the low blood glucose threshold or higher than the high blood glucose threshold, or when the rate of decrease at three consecutive prediction points exceeds the safety limit, it is determined that the abnormal deviation amplitude exceeds the preset safety limit.

[0086] Once any of the above conditions are met, the system triggers anomaly trend correlation analysis. The specific process of anomaly trend correlation analysis is as follows: the system backtracks the comprehensive quantitative index sequence of exercise intensity, the heart rate change rate sequence, and the skin temperature sequence within the time window corresponding to the current prediction sequence (30 minutes prior to the current moment), calculates the cross-correlation coefficient between these sequences and the abnormal portion of the prediction sequence, and determines the cause of the anomaly. For example, if the mean of the comprehensive quantitative index of exercise intensity jumps from 0.65 to 0.88 for five consecutive minutes before the predicted blood glucose value drops rapidly, and the heart rate change rate also increases simultaneously, then the anomaly is determined to be induced by high-intensity exercise. If the skin temperature drops significantly while the exercise intensity remains stable, it may be related to changes in ambient temperature or dehydration. The system matches the correlation analysis results with a pre-defined anomaly type library, which includes, but is not limited to: hypoglycemia risk induced by high-intensity exercise, delayed hypoglycemia caused by prolonged exercise, rebound hyperglycemia after exercise interruption, and immediate hypoglycemia caused by the combined effect of exercise and postprandial insulin spikes. The causal description and risk level corresponding to each type are predefined in the library.

[0087] For example, a type 1 diabetic patient is performing interval running. The predicted sequence generated in step S4 shows that at the 12th minute, the predicted blood glucose level drops to 3.8 mmol / L, below the hypoglycemic threshold of 3.9 mmol / L. Simultaneously, the rate of decline from the 8th to the 12th minute reaches 0.28 mmol / L per minute, approaching the safe limit. Peak-valley analysis shows the trough occurs at the 14th minute, with a value of 3.6 mmol / L, while the user's historical normal trough value is 4.5 mmol / L, a deviation of 2.0 mmol / L. The system triggers an abnormal trend correlation analysis, reviewing the exercise intensity index of the previous 10 minutes and finding that the user performed two sprints between the 4th and 6th minutes, with the index rising from 0.72 to 0.94, and the heart rate change rate increasing synchronously. The correlation analysis results show a high correlation coefficient (0.91), and the system determines the abnormal state type as "risk of acute hypoglycemia induced by high-intensity interval exercise," caused by "a sudden increase in exercise intensity leading to muscle glucose uptake exceeding the liver's glucose output compensatory capacity." This determination result is output to step S6 for adjusting the compensation strategy.

[0088] Step S5 extracts the peak-valley distribution patterns and deviation characteristics of the predicted sequence and correlates them with real-time exercise intensity and physiological parameters for causal analysis. This solves the technical problem in the background technology that existing methods cannot actively identify abnormal blood glucose states and their exercise-related causes. This technical feature, along with the predicted sequence generation in step S4 and the exercise intensity quantification and lag time correction functions in steps S1 to S3, mutually supports and interacts with each other, forming a complete technical means from blood glucose prediction to abnormal identification. Its beneficial effect is that it can provide early warning of abnormal blood glucose events and offer interpretable causal diagnosis, providing a clear basis for the generation of personalized compensation strategies in step S6.

[0089] Step S6: Analyze the correlation between the abnormal blood glucose change and the comprehensive quantitative index of exercise intensity, adjust the lag time estimation range and / or prediction frequency according to the correlation results, and generate a blood glucose lag compensation scheme that includes the adjusted lag time estimation range and / or prediction frequency.

[0090] Specifically, based on the correlation between the abnormal blood glucose levels and the comprehensive quantitative index of exercise intensity, the compensation interval division strategy is adjusted; the personalized correction parameters are updated using the adjusted compensation interval division strategy; the personalized correction parameters include: the weight of the influence of exercise intensity on lag time, the upper limit adjustment coefficient and the lower limit adjustment coefficient of the lag time estimation range; based on the updated personalized correction parameters, the adjusted lag time estimation range and / or prediction frequency are determined, and a blood glucose lag compensation scheme containing the adjusted lag time estimation range and / or prediction frequency is generated.

[0091] More specifically, step S6 analyzes the correlation between the abnormal blood glucose state determined in step S5 and the comprehensive quantitative index of exercise intensity generated in step S1. Based on the correlation results, it adjusts the lag time estimation range and / or prediction frequency to generate a blood glucose lag compensation scheme containing the adjusted parameters. The system first obtains the abnormal state type and its cause output in step S5, and simultaneously obtains the comprehensive quantitative index sequence of exercise intensity within the time window corresponding to the abnormal state (e.g., 10 minutes before the abnormality occurs to 5 minutes after). The correlation analysis is performed by calculating the Pearson correlation coefficient between the severity score of the abnormal state and the exercise intensity index sequence, where the severity score of the abnormal state is comprehensively quantified based on the magnitude and duration of deviation from the safety limit in the predicted sequence. An absolute value of the correlation coefficient greater than 0.7 is considered a strong correlation, between 0.4 and 0.7 is considered a moderate correlation, and less than 0.4 is considered a weak correlation. A strong correlation indicates that the abnormal state is mainly caused by changes in exercise intensity, a moderate correlation suggests the presence of other confounding factors (such as diet or medication), and a weak correlation indicates that exercise intensity is not the main trigger.

[0092] Based on the correlation results, the system adjusts the compensation interval division strategy. The compensation interval division strategy divides the value range of the exercise intensity index into several partitions, each partition corresponding to a set of lag time estimation parameters and prediction frequency parameters. Initially, the strategy is set based on population average data. For example, an exercise intensity index of 0 to 0.3 corresponds to the low-intensity zone, with a lag time range of [15, 30] minutes and a prediction frequency of once every 2 minutes; 0.3 to 0.6 corresponds to the medium-intensity zone, with a lag time range of [10, 20] minutes and a prediction frequency of once per minute; and 0.6 to 1.0 corresponds to the high-intensity zone, with a lag time range of [5, 12] minutes and a prediction frequency of once every 30 seconds. When correlation analysis shows a strong correlation between abnormal states and exercise intensity indicators within a specific intensity range, the system subdivides and adjusts that range: if the abnormality type is hypoglycemia risk and the correlation is positive, indicating that the lag time within that intensity range is too large, leading to prediction lag, the lower and upper limits of the estimated lag time range for the corresponding range are each shortened by a certain proportion (e.g., 10% of the original range length); if the abnormality type is rebound hyperglycemia and the correlation is negative, indicating that the lag time is too small, leading to overcompensation, the range is extended by the same proportion. This adjusted compensation range segmentation strategy makes the lag time parameters under different exercise intensities more closely match the user's individual response patterns.

[0093] By adjusting the compensation interval division strategy, the system updates three personalized correction parameters: the weight of the impact of exercise intensity on lag time, the upper limit adjustment coefficient of the lag time estimation range, and the lower limit adjustment coefficient. The initial weight of the impact of exercise intensity on lag time is set to 0.5, meaning that for every 0.1 increase in intensity, the lag time center value shortens by 5%. In cases of strong correlation and recurring hypoglycemia abnormalities, this weight is increased to 0.7, making the system more sensitive to changes in exercise intensity. The upper limit adjustment coefficient and the lower limit adjustment coefficient control the degree of asymmetry between the upper and lower boundaries of the lag time interval and the center value, respectively, with an initial value of 1.0 for both. If the abnormal state is manifested as an excessively deep drop in blood glucose (the trough is 30% lower than the historical normal value), the system increases the lower limit adjustment coefficient to 1.2, further lowering the lower boundary of the interval and triggering the predicted decline earlier; if the abnormal state is manifested as an excessively slow recovery of blood glucose (prolonged recovery time), the upper limit adjustment coefficient is increased to 1.3, delaying the predicted decline endpoint. Based on the updated personalized correction parameters, the system recalculates the lag time estimation range and prediction frequency corresponding to each exercise intensity interval. The central value of the estimated lag time range is determined by the product of the motion intensity index and the influence weight, while the upper and lower limits are obtained by multiplying the central value by the upper and lower limit adjustment coefficients, respectively. The prediction frequency is also set differently for each interval: in strongly correlated intervals, the prediction frequency is increased to once every 30 seconds to increase the risk monitoring density; in weakly correlated intervals, the frequency is restored to once every 2 minutes to reduce the computational load.

[0094] For example, a type 1 diabetic patient experienced hypoglycemia (blood glucose levels below 3.5 mmol / L) 8 to 12 minutes after exercise during three consecutive high-intensity running sessions. Step S5 determined the abnormality type as "risk of acute hypoglycemia induced by high-intensity exercise." Step S6 extracted the exercise intensity index sequence and found that the mean index values ​​in the 5 minutes before the abnormality occurred were between 0.82 and 0.91, with a correlation coefficient of 0.88 with the severity score, indicating a strong association. The system further subdivided the high-intensity zone (0.6 to 1.0) into two sub-intervals: 0.6 to 0.8 and 0.8 to 1.0. For the 0.8 to 1.0 sub-interval, the lower limit of the original lag time range [5, 12] minutes was adjusted from 5 minutes to 4 minutes, and the upper limit was adjusted from 12 minutes to 10 minutes, shortening the interval length to respond more quickly to the decrease in blood glucose; the prediction frequency was increased from once every 30 seconds to once every 15 seconds. Simultaneously, the personalized correction parameters are updated: the weight of exercise intensity's influence is increased from 0.5 to 0.75, the lower limit adjustment coefficient is increased from 1.0 to 1.3, and the upper limit adjustment coefficient remains unchanged at 1.0. Based on the new parameters, the lag time range corresponding to the 0.8 to 1.0 interval is recalculated to [4, 8] minutes, and the prediction frequency is locked at once every 15 seconds. The generated blood glucose lag compensation scheme includes the above-mentioned adjusted lag time estimation range and prediction frequency, and is output to step S7 for subsequent prediction processing.

[0095] Step S6 establishes a quantitative link between abnormal states and exercise intensity indicators through correlation analysis, and dynamically adjusts the compensation interval division strategy and personalized correction parameters accordingly. This solves the problem in the background technology where the compensation strategy is fixed and cannot be adaptively optimized based on the user's actual abnormal response. This technical feature, together with the abnormal state judgment in step S5 and the exercise intensity quantification and lag time correction functions in steps S1 to S3, supports and interacts with each other, forming a closed-loop adaptive mechanism from anomaly analysis to strategy adjustment. Its beneficial effect is that it enables the prediction system to continuously learn the user's individual response characteristics in multiple exercise events, successively optimize the lag time estimation and prediction frequency, and significantly reduce the false negative and false positive rates of abnormal blood glucose events.

[0096] Furthermore, it also includes step S7: using the adjusted lag time estimation range and prediction frequency determined in the blood glucose lag compensation scheme, the subsequently collected acceleration data and heart rate signals are reprocessed to predict blood glucose changes, and the actual blood glucose values ​​of the same period are obtained, the prediction accuracy is calibrated, and the final blood glucose change lag time adjustment value is determined.

[0097] Specifically, the following steps are taken: acquiring subsequent acceleration data and heart rate signals; using the lag time estimation range and prediction frequency in the blood glucose lag compensation scheme, performing dynamic prediction of blood glucose concentration on the subsequently acquired acceleration data and heart rate signals; comparing the prediction results with the actual blood glucose values ​​acquired during the same period to calculate the prediction deviation; and dynamically adjusting the prediction parameters based on the prediction deviation using real-time update frequency technology until the prediction deviation is lower than a preset threshold, at which point the lag time is determined as the final blood glucose change lag time adjustment value.

[0098] Step S7 uses the blood glucose lag compensation scheme generated in step S6 to reprocess the subsequently collected acceleration data and heart rate signals for blood glucose change prediction, and obtains the actual blood glucose values ​​for the same period, calibrates the prediction accuracy, and determines the final blood glucose change lag time adjustment value. The user continues to wear the wearable device and the continuous glucose monitor. The system continuously collects acceleration data and heart rate signals within a set time window, such as the next 30 minutes, while simultaneously recording the actual blood glucose values ​​output by the continuous glucose monitor at the same time resolution. The system first reads the adjusted lag time estimation range and prediction frequency from the blood glucose lag compensation scheme, and uses these parameters to perform the blood glucose concentration dynamic prediction processing described in steps S1 to S4 on the subsequently collected acceleration data and heart rate signals, generating a new prediction sequence. This prediction sequence contains multiple future time points, such as blood glucose concentration values ​​at a prediction point every 30 seconds or every minute. The system calculates the difference point by point between each prediction point in the prediction sequence and the actual blood glucose values ​​collected at the same period, aligning them by time, to obtain the prediction deviation sequence. The statistics of the prediction deviation sequence include the mean absolute deviation, the maximum absolute deviation, and the directional distribution of the deviation.

[0099] When the average absolute deviation of the prediction error exceeds a preset threshold, such as 0.5 mmol / L, or the maximum absolute deviation exceeds 1.5 mmol / L, the system dynamically adjusts the prediction parameters using real-time update frequency technology. Real-time update frequency technology refers to the process of iteratively correcting the lag time estimation range and prediction frequency: the system moves the lower and upper limits of the lag time estimation range in small steps, such as 0.5 minutes each time, in the direction of reducing the deviation, and re-executes the prediction and deviation calculation; if the predicted values ​​are generally higher than the actual values ​​(i.e., the predicted blood glucose decrease lags behind the actual decrease), the lag time estimation range is shortened overall; if the predicted values ​​are generally lower than the actual values ​​(the predicted decrease precedes the actual decrease), the range is extended overall. After each adjustment, the system recalculates the prediction deviation and compares the changes in the two deviations. When the absolute value of the prediction deviation continuously decreases with adjustment and converges below the preset threshold, the system stops adjusting, and the lag time used at this point is determined as the final adjusted lag time value for blood glucose changes. The prediction frequency adjustment and the lag time interval adjustment are carried out in parallel: if the prediction deviation increases significantly during the rapid blood glucose change phase (the rate of decline exceeds 0.2 mmol / L per minute), the system will increase the prediction frequency by one level, for example, from once per minute to once every 30 seconds; if the deviation does not improve during the stable phase, it will be restored to the original frequency to reduce the computational load.

[0100] For example, a type 1 diabetic patient has completed steps S1 to S6. The adjusted lag time estimate range in the blood glucose lag compensation scheme generated in step S6 is [6, 10] minutes, and the prediction frequency is once every 30 seconds. In a subsequent exercise session, the system collects new acceleration and heart rate data, uses the parameters in the compensation scheme to perform prediction, and obtains a blood glucose prediction sequence for the next 30 minutes. Simultaneously, the actual blood glucose value recorded by the continuous glucose monitor shows that the actual blood glucose begins to decrease at the 6th minute and drops to 4.0 mmol / L at the 9th minute, while the predicted sequence only begins to decrease at the 7th minute and drops to 4.0 mmol / L at the 10th minute, with the prediction lagging behind the actual value by approximately 1 minute. The average absolute deviation of the prediction is 0.65 mmol / L, exceeding the threshold of 0.5 mmol / L. The system activates real-time update frequency technology, shortening both the lower and upper limits of the lag time estimate range by 0.5 minutes, adjusting it to [5.5, 9.5] minutes, and re-executes the prediction. The average absolute deviation of the second prediction decreases to 0.4 mmol / L, still slightly above the threshold but with significant improvement. The system further shortened the range by 0.5 minutes to [5,9] minutes, and the deviation of the third prediction decreased to 0.3 mmol / L, below the threshold. The system stopped adjusting, and the lag time of 5 to 9 minutes was determined as the final lag time adjustment value for blood glucose changes. This value was stored in the user's personalized parameter library as a supplement to the historical blood glucose curve reference data when the next step S2 is executed, and was used to update the calculation model of the initial lag time estimation range.

[0101] Step S7 introduces actual blood glucose values ​​as the gold standard feedback and uses an iterative fine-tuning method to perform closed-loop calibration of the lag time estimation range in the compensation scheme. This solves the technical problem in existing technologies where the prediction model cannot continuously optimize parameters based on real blood glucose responses, leading to decreased accuracy after long-term use. This technical feature, along with the compensation scheme generated in step S6 and the prediction processing functions in steps S1 to S4, mutually supports and interacts with each other, forming a complete closed-loop adaptive system from parameter generation to effect verification and parameter optimization. Its beneficial effect is that it enables the prediction method to have online learning capabilities, adapting to the slow drift of the user's physiological state over time, such as changes in insulin sensitivity and improved physical fitness, thereby maintaining high prediction accuracy in the long term. By feeding the final lag time adjustment value back to the historical database in step S2, the system achieves self-maintenance and self-optimization of the individualized lag time model, significantly improving the robustness and clinical applicability of the method.

[0102] As attached Figure 2 This figure shows the effect of the method for predicting and compensating for blood glucose changes in exercise intensity and physiological parameters according to an embodiment of the present invention. The solid black line in the figure represents the measured curve of the continuous glucose monitoring system (CGM), and the dashed black line represents the predicted curve of the present invention. During the exercise monitoring period from 0 to 35 minutes, the measured values ​​of the CGM and the predicted values ​​of the present invention are highly consistent at each time point: 6.70 mmol / L at 0 minutes, 5.80 mmol / L at 5 minutes, 3.40 mmol / L at 10 minutes, and the deviation between the measured 3.50 mmol / L and the predicted 3.65 mmol / L at 12 minutes is only 0.15 mmol / L. The measured values ​​at 15 minutes, 17 minutes, 20 minutes, 25 minutes, 30 minutes, and 35 minutes are almost identical to the predicted values. The above data shows that the dynamic blood glucose concentration prediction sequence generated by the method of the present invention can track the real blood glucose change trend in real time, with minimal prediction deviation throughout the exercise process. This verifies that the present invention, through dynamic correction of the lag time interval, fusion of multi-source physiological parameters, and closed-loop adaptive compensation, has excellent prediction accuracy and reliability in complex exercise scenarios.

[0103] Secondly, such as Figure 3 As shown, this embodiment provides an intelligent compensation system for blood glucose detection, including:

[0104] The data acquisition module collects human acceleration data and heart rate signals; the intensity quantification module fuses and generates a comprehensive quantitative index of exercise intensity; the state determination module compares the comprehensive quantitative index of exercise intensity with a preset threshold to determine whether the exercise is in a high-intensity or low-intensity state; based on the spectral energy distribution characteristics of real-time acceleration data, it identifies the current exercise metabolic mode; for a high-intensity exercise state, it extracts reference data from the corresponding historical blood glucose curve sub-database to determine the preliminary lag time estimation range of blood glucose changes; the lag correction module uses the frequency fluctuations of the real-time acceleration data as correction data to dynamically adjust the preliminary lag time estimation range, wherein the dynamic offset adjustment includes: when the amplitude of the determined frequency fluctuation shows a sharp change, adjusting the lag time estimation range according to the direction and amplitude of the change. The system selectively shortens or extends the lower and upper limits of the initial lag time estimation range to generate a corrected blood glucose change lag time interval; a prediction sequence generation module is used to generate a dynamic blood glucose concentration prediction sequence through time series analysis, using the corrected blood glucose change lag time interval, combined with real-time collected heart rate change rate and skin temperature as current physiological parameters, and historical blood glucose periodic change trend data; an anomaly judgment module is used to judge abnormal blood glucose change states based on the distribution pattern and deviation characteristics of the dynamic blood glucose concentration prediction sequence; and a compensation scheme generation module is used to analyze the correlation between the abnormal blood glucose change states and the comprehensive quantitative index of exercise intensity, adjust the lag time estimation range and / or prediction frequency according to the correlation results, and generate a blood glucose lag compensation scheme that includes the adjusted lag time estimation range and / or prediction frequency.

[0105] Specifically, the intelligent compensation system for blood glucose monitoring is integrated into a wearable device, which includes a hardware carrier and embedded software modules. The hardware carrier is equipped with a triaxial accelerometer, a photoelectric heart rate sensor, a skin temperature sensor, and a Bluetooth module for wireless communication with the continuous glucose monitor. The system includes a data acquisition module, an intensity quantification module, a status determination module, a hysteresis correction module, a prediction sequence generation module, an anomaly detection module, and a compensation scheme generation module. Each module is burned into the device's ROM as firmware, eliminating the need for an external server during operation and ensuring low latency and offline availability.

[0106] The data acquisition module consists of sensors and an analog front-end, acquiring raw signals of acceleration, heart rate, and skin temperature in real time at a sampling rate of 10Hz. It also obtains the actual blood glucose value output by the dynamic blood glucose monitor via Bluetooth for subsequent calibration. The acquired raw data is temporarily stored in the device's built-in SRAM and is timestamped to ensure temporal alignment of the multimodal data. The intensity quantification module uses an embedded microprocessor to perform frequency domain analysis and multidimensional mapping algorithms: bandpass filtering and fast Fourier transform are applied to the acceleration data to extract temporal fluctuation characteristics and frequency amplitude changes; the average heart rate, heart rate variability, and heart rate change rate are calculated for the heart rate signal; after normalizing the above five feature values, a weighted summation is used to generate a comprehensive quantitative index of exercise intensity (value 0~1), which is then passed to the state determination module.

[0107] The state determination module first compares the comprehensive quantitative index of exercise intensity with a preset threshold of 0.6. If the index is greater than 0.6, it is determined to be a high-intensity exercise state; otherwise, it is a low-intensity exercise state. For the high-intensity exercise state, the module further identifies the exercise metabolic mode based on the spectral energy distribution characteristics of real-time acceleration data: calculating the proportion E of the dominant frequency band (1.5-2.5Hz) energy to the total energy and the ratio R of the high-frequency band (2.5-4Hz) to the low-frequency band (0.5-1.5Hz). If E≥0.7 and the proportion of high-frequency energy is less than 0.15, it is determined to be a continuous aerobic mode; if R>1.2 and the proportion of dominant frequency band energy is less than 0.5, it is determined to be an intermittent anaerobic mode. According to the identified mode, the module extracts reference data of blood glucose curves for at least five previous exercises of the same or similar intensity from the corresponding historical blood glucose curve sub-library (aerobic mode sub-library or anaerobic mode sub-library) in the device's built-in Flash or cloud database, analyzes the response delay interval (taking the 25th percentile to the 75th percentile), and determines the preliminary lag time estimation range. For low-intensity exercise, the module directly uses a preset default lag time interval (e.g., [10, 20] minutes for type 1 diabetes users) as the initial estimation range.

[0108] The lag correction module monitors the acceleration frequency amplitude sequence in real time and calculates the average difference of three consecutive windows as a measure of the change amplitude. When the change amplitude exceeds the abrupt change threshold of 0.15, it is determined that the frequency fluctuation amplitude has changed abruptly, and the lower and upper limits of the initial lag time estimation range are dynamically offset according to the direction of change (positive for increasing exercise intensity, negative for decreasing intensity). The offset calculation rule is: offset = base value × amplitude change rate, where the base value is one-quarter of the total length of the initial lag time estimation range, the amplitude change rate is the ratio of the current change amplitude to the threshold, and the offset does not exceed half of the total range length. The adjusted interval is rounded to generate the corrected blood glucose change lag time interval.

[0109] The prediction sequence generation module divides the prediction time axis into a metabolic response segment [L, U] and a steady-state trend segment (U, U+ΔT] (ΔT=30 minutes) based on the corrected lag time interval [L, U]. In the metabolic response segment, the module calculates the Pearson correlation coefficient r between the heart rate change rate, skin temperature, and historical periodic change trend data, and combines it with the comprehensive quantitative index M of exercise intensity to generate the first subsequence point by point according to the weighted average formula in the above embodiment. In the steady-state trend segment, the module performs linear extrapolation based on the low-frequency characteristics (average change rate) of historical blood glucose periodic change trend data to generate the second subsequence. The two subsequences are weighted averaged and smoothed at point U to generate a complete dynamic prediction sequence containing multiple future time points.

[0110] The anomaly detection module extracts the peak and trough distribution patterns and the abnormal deviation of the predicted value from historical normal values ​​from the predicted sequence. If the duration of the decline or the recovery time deviates from the preset normal range, or if the predicted value is lower than the hypoglycemia threshold (3.9 mmol / L) and the rate of decline exceeds the safety limit, an abnormal trend correlation analysis is triggered. By retrospectively analyzing exercise intensity indicators, heart rate change rate, and skin temperature sequences, the cross-correlation coefficient is calculated to determine the type and cause of the abnormal state (e.g., "risk of acute hypoglycemia induced by high-intensity intermittent exercise").

[0111] The compensation scheme generation module analyzes the correlation between abnormal states and comprehensive quantitative indicators of exercise intensity (correlation coefficient > 0.7 indicates a strong correlation). Based on the correlation results, the compensation interval division strategy is adjusted: the exercise intensity index value range is further subdivided, and personalized correction parameters such as the influence weight of exercise intensity on lag time, the upper limit adjustment coefficient and lower limit adjustment coefficient of the lag time estimation range are updated. The lag time estimation range and prediction frequency corresponding to each intensity interval are recalculated, generating an adaptive blood glucose lag compensation scheme that includes the adjusted parameters. All modules work collaboratively, enabling the system to complete the entire process from data acquisition to compensation scheme generation in real time under exercise scenarios.

[0112] Thirdly, this embodiment provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the intelligent compensation method for blood glucose detection as provided in this embodiment.

[0113] The above description is merely a specific embodiment of this specification. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, modules, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here. It should be understood that the scope of protection of this specification is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this specification, and these modifications or substitutions should all be covered within the scope of protection of this specification.

Claims

1. A smart compensation method for blood glucose detection, characterized in that, include: Step S1: Collect human acceleration data and heart rate signals, and fuse them to generate a comprehensive quantitative index of exercise intensity; Step S2: Compare the comprehensive quantitative index of exercise intensity with the preset threshold to determine whether it is a high-intensity exercise state or a low-intensity exercise state, and extract historical blood glucose curve reference data for the high-intensity exercise state to determine the preliminary lag time estimation range of blood glucose change. Step S3: Combining the frequency fluctuations of the real-time collected acceleration data as correction data, dynamically offset the range of the preliminary lag time estimation. The dynamic offset adjustment includes: when the amplitude of the frequency fluctuation shows a sharp change, selectively shortening or extending the lower limit and upper limit of the preliminary lag time estimation range according to the direction and amplitude of the change, to generate a corrected blood glucose change lag time interval. Step S4: Using the corrected blood glucose change lag time interval, combined with the real-time collected heart rate change rate and skin temperature as current physiological parameters, and historical blood glucose periodic change trend data, a dynamic blood glucose concentration prediction sequence is generated through time series analysis. Step S5: Determine the abnormal state of blood glucose changes based on the distribution pattern and deviation characteristics of the dynamic blood glucose concentration prediction sequence; Step S6: Analyze the correlation between the abnormal blood glucose change and the comprehensive quantitative index of exercise intensity, adjust the lag time estimation range and / or prediction frequency according to the correlation results, and generate a blood glucose lag compensation scheme that includes the adjusted lag time estimation range and / or prediction frequency.

2. The intelligent compensation method for blood glucose detection according to claim 1, characterized in that, Step S1 specifically includes: Human acceleration data and heart rate signals are collected by wearable devices, and the acceleration data is processed by frequency domain analysis to extract the temporal fluctuation characteristics and frequency amplitude changes of the motion. The heart rate signal is analyzed to extract the average heart rate, heart rate variability, and heart rate rise or fall rate as physiological parameter change characteristics of the heart rate signal. By using a multidimensional mapping method, the temporal fluctuation characteristics, the frequency amplitude changes, and the physiological parameter changes of the heart rate signal are fused together to calculate the comprehensive quantitative index of exercise intensity.

3. The intelligent compensation method for blood glucose detection according to claim 1, characterized in that, Step S2 specifically includes: Obtain the comprehensive quantitative index of exercise intensity and compare it with the preset exercise intensity threshold; If the comprehensive quantitative index of exercise intensity exceeds the preset exercise intensity threshold, it is determined to be a high-intensity exercise state; otherwise, it is a low-intensity exercise state. Based on the spectral energy distribution characteristics of real-time acceleration data, the current exercise metabolic mode is identified as either a continuous aerobic mode or an intermittent anaerobic mode. For the high-intensity exercise state, based on the identified exercise metabolic pattern, reference data is extracted from the corresponding historical blood glucose curve sub-database, the response delay interval is analyzed, and the preliminary lag time estimation range of blood glucose changes is determined. For low-intensity motion states, a preset default lag time interval is used as the initial lag time estimation range.

4. The intelligent compensation method for blood glucose detection according to claim 1, characterized in that, In step S3, the preliminary lag time estimation range is dynamically offset and adjusted, specifically including: The frequency fluctuations of the acceleration data collected in real time are used as correction data; If the amplitude of the frequency fluctuation shows a sharp change, the lower and upper limits of the preliminary lag time estimation range are shortened or extended according to the direction and amplitude of the change, generating a corrected blood glucose change lag time interval; wherein, the formula for calculating the offset is: offset = base value × amplitude change rate, the base value is set to one-quarter of the total length of the preliminary lag time estimation range, the amplitude change rate is the ratio of the current frequency amplitude change to the preset sharp change threshold, and the offset does not exceed one-half of the total length of the preliminary lag time estimation range.

5. The intelligent compensation method for blood glucose detection according to claim 1, characterized in that, In step S4, the generation of a dynamic blood glucose concentration prediction sequence through time-series analysis specifically includes: Based on the corrected blood glucose change lag time interval [L, U], the prediction time axis is divided into a metabolic response segment and a steady-state trend segment. The metabolic response segment corresponds to the time points within the interval [L, U], and the steady-state trend segment corresponds to the time points within the interval (U, U+ΔT], where ΔT is the preset prediction duration. In the metabolic response phase, a weighted average method is used to predict short-period fluctuations in the heart rate change rate, skin temperature, and the periodic change trend data at continuous time points, generating the first subsequence; In the steady-state trend segment, long-term baseline prediction is performed based on the low-frequency characteristics of historical blood glucose cyclical change trend data to generate a second subsequence; The first subsequence and the second subsequence are spliced ​​and coupled along the time axis to generate a dynamic prediction sequence containing predicted blood glucose concentration values ​​for multiple future time points.

6. The intelligent compensation method for blood glucose detection according to claim 1, characterized in that, Step S5 specifically includes: The distribution patterns of peaks and troughs, as well as the abnormal deviations between predicted values ​​and historical normal values, are extracted from the dynamic blood glucose concentration prediction sequence. If the distribution pattern of the peaks and troughs deviates from the preset normal range, or if the abnormal deviation exceeds the preset safety limit, then an abnormal trend correlation analysis is triggered. Based on the results of the abnormal trend correlation analysis, the type and cause of abnormal blood glucose changes are determined.

7. The intelligent compensation method for blood glucose detection according to claim 1, characterized in that, In step S6, the blood glucose lag compensation scheme specifically includes: Based on the correlation between the abnormal blood glucose levels and the comprehensive quantitative index of exercise intensity, the compensation interval division strategy is adjusted. The personalized correction parameters are updated by adjusting the compensation interval division strategy; the personalized correction parameters include: the weight of the influence of exercise intensity on lag time, the upper limit adjustment coefficient and the lower limit adjustment coefficient of the lag time estimation range; Based on the updated personalized correction parameters, the adjusted lag time estimation range and / or prediction frequency are determined, and a blood glucose lag compensation scheme containing the adjusted lag time estimation range and / or prediction frequency is generated.

8. The intelligent compensation method for blood glucose detection according to claim 1, characterized in that, It also includes step S7: using the adjusted lag time estimation range and prediction frequency determined in the blood glucose lag compensation scheme, the subsequently collected acceleration data and heart rate signals are reprocessed to predict blood glucose changes, and the actual blood glucose values ​​of the same period are obtained, the prediction accuracy is calibrated, and the final blood glucose change lag time adjustment value is determined.

9. An intelligent compensation system for blood glucose detection, characterized in that, include: The data acquisition module is used to collect human acceleration data and heart rate signals; The intensity quantification module is used to integrate and generate a comprehensive quantitative index of exercise intensity. The state determination module is used to compare the comprehensive quantitative index of exercise intensity with a preset threshold to determine whether it is a high-intensity exercise state or a low-intensity exercise state. Based on the spectral energy distribution characteristics of real-time acceleration data, the current exercise metabolic mode is identified. For those identified as high-intensity exercise, reference data is extracted from the corresponding historical blood glucose curve sub-database according to the identified exercise metabolic mode to determine the preliminary lag time estimation range of blood glucose changes. The hysteresis correction module is used to combine the frequency fluctuations of the real-time collected acceleration data as correction data to dynamically offset the range of the preliminary hysteresis time estimation. The dynamic offset adjustment includes: when the amplitude of the frequency fluctuation is determined to change drastically, selectively shortening or extending the lower limit and upper limit of the preliminary hysteresis time estimation range according to the direction and amplitude of the change, so as to generate a corrected hysteresis time interval for blood glucose changes. The prediction sequence generation module is used to generate a dynamic prediction sequence of blood glucose concentration by using the corrected blood glucose change lag time interval, combined with the real-time collected heart rate change rate and skin temperature as current physiological parameters, and historical blood glucose periodic change trend data. An anomaly detection module is used to determine abnormal blood glucose changes based on the distribution pattern and deviation characteristics of the dynamic blood glucose concentration prediction sequence. The compensation scheme generation module is used to analyze the correlation between the abnormal blood glucose change state and the comprehensive quantitative index of exercise intensity, adjust the lag time estimation range and / or prediction frequency according to the correlation results, and generate a blood glucose lag compensation scheme that includes the adjusted lag time estimation range and / or prediction frequency.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1 to 8.

Images