Health management data processing method and system based on multi-modal data

Abstract

The invention provides a health management data processing method and system based on multi-modal data, and relates to the technical field of data processing, and the method comprises the steps: extracting time sequence characteristic parameters of health data, calculating correlation coefficients among indexes, constructing an index influence network, carrying out an interference suppression test to determine a causal relationship, and carrying out the fusion of multi-source data to construct a characteristic space. And identifying a state transition path, determining a risk trigger point and an adjustment target value, and generating a dynamically optimized health management scheme. According to the invention, accurate correlation analysis of multi-modal health data can be realized, and the accuracy of health risk prediction and the pertinence and effectiveness of a health management scheme are improved.

Description

Technical Field

[0001] The present invention relates to data processing technology, and in particular to a health management data processing method and system based on multimodal data. Background Art

[0002] Existing health management systems often rely on a single modality of health data for analysis and intervention. For example, they generate regulatory solutions based solely on physiological indicators or behavioral information, ignoring the potential correlations between multiple sources of data and failing to fully reflect the dynamic evolution of an individual's health status. This is especially true in scenarios where individual differences are significant and health issues exhibit multi-dimensional, cross-evolutionary trends. Single-modality data processing methods struggle to provide a basis for precise intervention.

[0003] Furthermore, while some studies have attempted to incorporate diverse health data into management systems, data fusion methods lack the ability to model the causal relationships and dynamic influence patterns between different data modalities. This leads to unclear intervention pathways across various indicators and fragmented regulatory strategies, limiting the scientific nature and effectiveness of personalized health programs. Most existing methods remain at the static indicator analysis stage and are unable to effectively identify the triggering mechanisms and development pathways of health risks.

[0004] Therefore, there is an urgent need for a health management data processing method that can integrate multimodal health data, identify causal relationships between indicators, and build a dynamic adjustment mechanism to achieve accurate characterization of changes in health status and real-time optimization and regulation. Summary of the Invention

[0005] The embodiments of the present invention provide a health management data processing method and system based on multimodal data, which can solve the problems in the prior art.

[0006] A first aspect of an embodiment of the present invention provides a method for processing health management data based on multimodal data, comprising: Receive user health data, extract the time series characteristic parameters of each indicator in the user health data, calculate the correlation coefficient between the time series characteristic parameters of different indicators, and write the indicator combination whose correlation coefficient exceeds the preset correlation threshold into the correlation indicator table; An indicator influence network is constructed based on the correlation indicator table. By introducing a dynamic compensation signal, interference suppression tests are performed on the correlation indicators in the indicator influence network. The attenuation coefficient and distortion degree during signal transmission are calculated. The indicator combinations with stable attenuation coefficients and distortion degrees below a preset threshold are written into the causal relationship table. The signal transmission characteristics and stabilization range are also recorded in the causal relationship table. Based on the causal relationship table, multi-source fusion of physiological, behavioral, and environmental data is performed to generate multimodal health data. Based on the multimodal health data, a feature space of health status is constructed. The state migration path is identified in the feature space, and the risk trigger point is determined by combining the signal transmission characteristics. The adjustment target value is determined by analyzing the coordinated change rules of different modal data within the stabilization action range. By calculating the deviation between the current indicator value and the adjustment target value, a health management plan including the adjustment sequence, adjustment amplitude and adjustment cycle is generated, and the health management plan is dynamically optimized based on real-time feedback during the indicator adjustment process.

[0007] In an optional embodiment, Receive user health data, extract the time series characteristic parameters of each indicator in the user health data, calculate the correlation coefficient between the time series characteristic parameters of different indicators, and write the indicator combination whose correlation coefficient exceeds the preset correlation threshold into the correlation indicator table, including: Receive user health data, calculate the change rate of each indicator in the user health data, determine the fluctuation range of the indicator based on the change rate, set a time series sampling window based on the fluctuation range, automatically shrink the sampling interval of the time series sampling window in the fluctuation-enhanced area, and automatically expand the sampling interval in the fluctuation-weakened area, and obtain a sampled data sequence within the time series sampling window; Extracting time series characteristic parameters from the sampled data sequence, the time series characteristic parameters including fluctuation amplitude ratio, fluctuation duration, and waveform characteristic value, calculating signal entropy values ​​for the time series characteristic parameters to construct a frequency component matrix, generating an eigendecomposition basis based on the frequency component matrix, performing multi-scale decomposition on the time series characteristic parameters using the eigendecomposition basis to obtain a characteristic change sequence, and writing the characteristic change sequence into a time series characteristic table; Based on the time series feature table, the correlation coefficient between the time series feature parameters of different indicators is calculated, the phase of the feature change sequence is calibrated, the sampling deviation is eliminated through dynamic compensation, the frequency domain cross-correlation of the calibrated sequence is calculated, the frequency domain cross-correlation and the time domain correlation coefficient are combined to form the correlation strength, and the indicator combination whose correlation strength exceeds the preset correlation threshold is written into the correlation indicator table.

[0008] In an optional embodiment, Calculating the signal entropy value of the time series characteristic parameters to construct a frequency component matrix, generating an eigendecomposition basis according to the frequency component matrix, and performing multi-scale decomposition on the time series characteristic parameters using the eigendecomposition basis to obtain a characteristic change sequence includes: The time series characteristic parameter sequence is segmented, and the reweighted Shannon entropy value of each segment is calculated. The normalized information entropy sequence is constructed by modifying the sequence using the adaptive weighting coefficient of the signal fluctuation amplitude. The normalized information entropy sequence is arranged according to the time scale to form an entropy distribution matrix. Constructing a frequency component matrix using the entropy distribution matrix, extracting entropy changes at adjacent time scales in the entropy distribution matrix, dividing the distribution intervals of the frequency components based on the entropy changes, calculating the mutual information between the frequency components in different distribution intervals, and writing the frequency component combinations whose mutual information exceeds a preset coupling threshold into a frequency coupling table; Extracting singular values ​​from the frequency component matrix, arranging the singular values ​​in descending order of energy contribution to obtain dominant frequency components, constructing the dominant frequency components as reference basis functions, adjusting the reference basis functions according to the frequency coupling table, minimizing reconstruction errors through iterative optimization, and generating an eigendecomposition basis; The feature decomposition basis is used to perform multi-scale decomposition on the time series feature parameters, the start and end positions of the decomposition interval are determined based on the frequency coupling table, a matching pursuit operation with an adaptive step size is performed in each decomposition interval, and the scale features obtained by decomposition are reorganized in chronological order to generate a feature change sequence.

[0009] In an optional embodiment, An indicator influence network is constructed based on the correlation indicator table. By introducing a dynamic compensation signal, an interference suppression test is performed on the correlation indicators in the indicator influence network. The attenuation coefficient and distortion degree during the signal transmission process are calculated. The indicator combinations with stable attenuation coefficients and distortion degrees below the preset threshold are written into the causal relationship table. The signal transmission characteristics and stable action ranges are recorded in the causal relationship table, including: An indicator influence network is constructed based on the association indicator table, wherein nodes represent health indicators and edges between nodes represent association strengths. The delay time of signal transmission on the edges is calculated, an adjustment period of a time-varying weight is determined based on the delay time, the time-varying weight is updated based on the signal fluctuation frequency within the adjustment period, and the topology of the indicator influence network is optimized based on the time-varying weight; Selecting related indicators in the optimized indicator influence network, injecting a test signal between the related indicators, detecting a transmission distortion of the test signal within a phase calibration interval, and generating a dynamic compensation signal based on the time-varying weight and the transmission distortion; Superimposing the dynamic compensation signal on the test signal to obtain a compensated test signal, extracting the instantaneous frequency characteristics of the compensated test signal, calculating the frequency offset relative to the original test signal, and obtaining the attenuation coefficient and distortion degree of signal transmission according to the frequency offset; The correlation index pair of the attenuation coefficient fluctuation amplitude less than the first threshold and the distortion degree less than the second threshold is written into the causal relationship table, the stable action interval is determined based on the frequency offset, and the signal transmission characteristics within the stable action interval are recorded in the causal relationship table.

[0010] In an optional embodiment, Based on multimodal health data, a feature space of health status is constructed. The state migration path is identified in the feature space. The risk trigger point is determined by combining the signal transmission characteristics. The adjustment target value is determined by analyzing the coordinated change rules of different modal data within the stabilization range. The following are included: Dynamically normalize and preprocess the multimodal health data, calculate the information entropy based on the preprocessed data to obtain the mutual information value between modalities, construct a data association matrix based on the mutual information value, use the data association matrix to design an adaptive weight function, dynamically weight the preprocessed data to obtain fusion features, and generate a feature association table; Segmenting the fused features into sliding windows, extracting statistical features and rate-of-change features within each window to obtain a feature sequence, performing density clustering on the feature sequence in combination with a feature association table to obtain state clusters, analyzing the distribution characteristics of the state clusters to determine state transition regions, and constructing a feature space of health status based on the state transition regions; Extracting a state transition trajectory in the feature space, calculating the geometric features of the state transition trajectory to obtain a state transition point, analyzing the system response curve at the state transition point in combination with signal transfer characteristics, determining a risk trigger point based on the convergence of the system response curve, and generating a risk distribution map; The periodic characteristics and trend characteristics of multimodal health data are extracted within the stabilization interval to obtain a change feature sequence, the phase relationship and amplitude relationship of the change feature sequence are analyzed to identify the coordinated change pattern, the safe activity range is determined based on the coordinated change pattern and the risk distribution map, and the state point with the optimal stability within the safe activity range is selected as the adjustment target value.

[0011] In an optional embodiment, Determining the scope of security activities based on the collaborative change pattern and risk distribution map includes: Acquire phase information of multimodal health data, calculate the phase coupling strength between the data, construct a coupling matrix based on the ratio of the maximum to the minimum phase coupling strength, mark the indicator combinations in the coupling matrix whose ratio exceeds a preset coupling threshold as strong coupling pairs, and extract the change trajectory of the strong coupling pairs within the stable working range to construct a collaborative change envelope; Calculating the gradient value of the risk distribution map to obtain a risk gradient, identifying a risk change boundary region based on the risk gradient, and shrinking and adjusting the coordinated change envelope according to the position of the risk change boundary region and the value of the risk gradient to obtain a modified envelope; An optimization objective function is constructed using the modified envelope as a constraint condition, the optimization objective function is solved to obtain a multi-dimensional activity range, and a state transition constraint condition is constructed according to the boundary characteristics of the multi-dimensional activity range; The safe activity area is divided based on the state transition constraint condition, the state stability index is extracted from the safe activity area, and the safe activity area is optimized and adjusted according to the state stability index to obtain the final safe activity range.

[0012] In an optional embodiment, By calculating the deviation between the current indicator value and the adjustment target value, a health management plan including the adjustment sequence, adjustment amplitude and adjustment cycle is generated, including: Obtaining the current indicator value and the adjustment target value, calculating the standardized deviation between the current indicator value and the adjustment target value, constructing an indicator impact matrix based on the standardized deviation, performing eigendecomposition on the indicator impact matrix to obtain eigenvectors, extracting the influence transfer path based on the eigenvectors to determine the adjustment order, and generating a sequence matrix; Acquire historical adjustment data, calculate an indicator response curve in combination with the sequence matrix, determine an adjustment step threshold based on a rate of change of the indicator response curve, adopt an initial adjustment step when the standardized deviation is greater than the adjustment step threshold, and adjust the adjustment step based on real-time feedback data when the standardized deviation is less than the adjustment step threshold, thereby generating an adjustment amplitude table; Collecting real-time indicator data, calculating the rate of change and the trend of change of the real-time indicator data, identifying state change points based on the rate of change and the trend of change, determining the adjustment period according to the time distribution of the state change points, and generating a period distribution table; The adjustment sequence in the sequence matrix is ​​grouped according to the indicator influence matrix, the adjustment step is determined in each group according to the adjustment amplitude table, the adjustment time is set in combination with the cycle distribution table, and a health management plan including the adjustment sequence, adjustment amplitude and adjustment cycle is generated.

[0013] A second aspect of an embodiment of the present invention provides a health management data processing system based on multimodal data, comprising: The first unit is used to receive user health data, extract the time series characteristic parameters of each indicator in the user health data, calculate the correlation coefficient between the time series characteristic parameters of different indicators, and write the indicator combination whose correlation coefficient exceeds a preset correlation threshold into the correlation indicator table; The second unit is used to construct an indicator influence network based on the correlation indicator table, perform interference suppression tests on the correlation indicators in the indicator influence network by introducing dynamic compensation signals, calculate the attenuation coefficient and distortion degree during signal transmission, write indicator combinations with stable attenuation coefficients and distortion degrees below a preset threshold into the causal relationship table, and record the signal transmission characteristics and stabilization range in the causal relationship table; The third unit is used to perform multi-source fusion of physiological, behavioral, and environmental data based on the causal relationship table to generate multimodal health data. Based on the multimodal health data, a feature space of health status is constructed. The state migration path is identified in the feature space. The risk trigger point is determined by combining the signal transmission characteristics. The adjustment target value is determined by analyzing the coordinated change pattern of different modal data within the stabilization interval. The fourth unit is used to generate a health management plan including the adjustment sequence, adjustment amplitude and adjustment cycle by calculating the deviation between the current indicator value and the adjustment target value, and dynamically optimize the health management plan based on real-time feedback during the indicator adjustment process.

[0014] According to a third aspect of an embodiment of the present invention, an electronic device is provided, including: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0015] According to a fourth aspect of an embodiment of the present invention, a computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the method described above is implemented.

[0016] In this embodiment, through multimodal data fusion and dynamic compensation signal testing, accurate identification of causal relationships between health indicators is achieved, effectively improving the pertinence and scientific nature of health management. By constructing an indicator influence network and analyzing signal transmission characteristics, it is possible to accurately locate health risk trigger points and determine the optimal adjustment target value, making health intervention measures more accurate and effective, and avoiding the problems of blind intervention and excessive intervention in traditional methods. The dynamic optimization mechanism based on real-time feedback enables the health management plan to be continuously adjusted according to user responses, realizing personalized and adaptive health management, significantly improving user compliance and health improvement effects, and providing a new technical approach for precision medicine and individualized health management. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 Schematic diagram of a flow chart of a health management data processing method based on multimodal data according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the architecture of the multi-scale decomposition and analysis system for time series feature parameters; Figure 3 This is a schematic diagram for analyzing the frequency offset during signal transmission; Figure 4 Schematic diagram of the comparison of phase coupling intensity ratios of multimodal health data. DETAILED DESCRIPTION

[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0019] The following specific embodiments are used to describe the technical solution of the present invention in detail. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0020] Figure 1 FIG. 1 is a flow chart of a health management data processing method based on multimodal data according to an embodiment of the present invention. Figure 1 As shown, the method includes: Receive user health data, extract the time series characteristic parameters of each indicator in the user health data, calculate the correlation coefficient between the time series characteristic parameters of different indicators, and write the indicator combination whose correlation coefficient exceeds the preset correlation threshold into the correlation indicator table; An indicator influence network is constructed based on the correlation indicator table. By introducing a dynamic compensation signal, interference suppression tests are performed on the correlation indicators in the indicator influence network. The attenuation coefficient and distortion degree during signal transmission are calculated. The indicator combinations with stable attenuation coefficients and distortion degrees below a preset threshold are written into the causal relationship table. The signal transmission characteristics and stabilization range are also recorded in the causal relationship table. Based on the causal relationship table, multi-source fusion of physiological, behavioral, and environmental data is performed to generate multimodal health data. Based on the multimodal health data, a feature space of health status is constructed. The state migration path is identified in the feature space, and the risk trigger point is determined by combining the signal transmission characteristics. The adjustment target value is determined by analyzing the coordinated change rules of different modal data within the stabilization action range. By calculating the deviation between the current indicator value and the adjustment target value, a health management plan including the adjustment sequence, adjustment amplitude and adjustment cycle is generated, and the health management plan is dynamically optimized based on real-time feedback during the indicator adjustment process.

[0021] In an optional embodiment, receiving user health data, extracting time series characteristic parameters of each indicator in the user health data, calculating correlation coefficients between time series characteristic parameters of different indicators, and writing indicator combinations whose correlation coefficients exceed a preset correlation threshold into a correlation indicator table includes: Receive user health data, calculate the change rate of each indicator in the user health data, determine the fluctuation range of the indicator based on the change rate, set a time series sampling window based on the fluctuation range, automatically shrink the sampling interval of the time series sampling window in the fluctuation-enhanced area, and automatically expand the sampling interval in the fluctuation-weakened area, and obtain a sampled data sequence within the time series sampling window; Extracting time series characteristic parameters from the sampled data sequence, the time series characteristic parameters including fluctuation amplitude ratio, fluctuation duration, and waveform characteristic value, calculating signal entropy values ​​for the time series characteristic parameters to construct a frequency component matrix, generating an eigendecomposition basis based on the frequency component matrix, performing multi-scale decomposition on the time series characteristic parameters using the eigendecomposition basis to obtain a characteristic change sequence, and writing the characteristic change sequence into a time series characteristic table; Based on the time series feature table, the correlation coefficient between the time series feature parameters of different indicators is calculated, the phase of the feature change sequence is calibrated, the sampling deviation is eliminated through dynamic compensation, the frequency domain cross-correlation of the calibrated sequence is calculated, the frequency domain cross-correlation and the time domain correlation coefficient are combined to form the correlation strength, and the indicator combination whose correlation strength exceeds the preset correlation threshold is written into the correlation indicator table.

[0022] For example, the system first receives user health data, which may come from a variety of health monitoring devices, such as smartwatches, blood pressure monitors, or blood glucose monitors. The system then calculates the rate of change of each indicator based on the received health data. For example, for blood pressure data, the rate of change can be calculated by dividing the difference between adjacent measurements by the time interval. For heart rate data, the ratio of the change in adjacent heart rate measurements to the time interval can be similarly calculated.

[0023] Based on the calculated rate of change, the fluctuation range of the indicator is determined. For example, if a user's blood sugar changes at a rate of 5mg / dL / hour within two hours after breakfast, but only at a rate of 1mg / dL / hour in other time periods, the post-breakfast period will be identified as an area of ​​increased fluctuation. Based on the identified fluctuation range, a time series sampling window is set to automatically shrink the sampling interval in areas of increased fluctuation and automatically expand the sampling interval in areas of reduced fluctuation. In specific implementation, when the blood sugar change rate exceeds a preset threshold (such as 3mg / dL / hour), the sampling interval is reduced from the standard 30 minutes to 10 minutes; when the change rate is below the threshold, the sampling interval is extended to 60 minutes to optimize data collection efficiency.

[0024] The sampling data sequence is obtained from the adjusted sampling window, and the time series characteristic parameters of the sequence are extracted. The extracted time series characteristic parameters include the fluctuation amplitude ratio, the fluctuation duration, and the waveform characteristic value. The fluctuation amplitude ratio refers to the ratio of the maximum value to the minimum value of the indicator during the fluctuation period; the fluctuation duration refers to the time required for the indicator to return to a stable state from the beginning of the fluctuation; the waveform characteristic value is a parameter describing the waveform shape, such as steepness, smoothness, etc. For example, for blood pressure data, the system may identify that the fluctuation amplitude ratio of systolic blood pressure is 1.3 (highest value 130mmHg, lowest value 100mmHg), the fluctuation duration is 40 minutes, and the waveform characteristic value shows a pattern of rapid rise followed by slow decline.

[0025] Signal entropy is calculated for the extracted time series feature parameters to quantify signal complexity and uncertainty. For each indicator's time series feature parameter, it is decomposed into components of different frequencies, and a frequency component matrix is ​​constructed. For example, heart rate variability can be decomposed into high-frequency components (0.15-0.4 Hz, reflecting parasympathetic nerve activity) and low-frequency components (0.04-0.15 Hz, reflecting sympathetic nerve activity). Based on the frequency component matrix, eigendecomposition bases are generated, which capture the main variation patterns in the data.

[0026] Using the generated feature decomposition basis, we perform multi-scale decomposition on the time series feature parameters to obtain feature change sequences. This decomposition can reveal patterns of change at different time scales, such as short-term (hourly), medium-term (daily), and long-term (weekly) trends in blood glucose. The obtained feature change sequences are written into a time series feature table to provide data support for subsequent correlation analysis.

[0027] Based on the time series feature table, the system calculates the correlation coefficients between the time series feature parameters of different indicators. Before calculating the correlation, the feature change sequence is phase-calibrated to eliminate phase differences caused by asynchronous sampling times. For example, the monitoring times of blood pressure and heart rate may not be completely consistent, so correction is performed through methods such as interpolation or time alignment. Dynamic compensation methods are used to eliminate sampling bias and ensure data comparability. Dynamic compensation methods analyze the patterns of sampling bias based on historical data and apply corresponding correction factors.

[0028] The frequency domain cross-correlation of the calibrated series is calculated, reflecting the degree of correlation between the two indicators at each frequency component. The time domain correlation coefficient is also calculated, reflecting the direct linear relationship between the indicators. Combining the frequency domain cross-correlation and the time domain correlation coefficient forms a correlation strength index, providing a more comprehensive assessment of the correlation. For example, a user's blood pressure and heart rate have a time domain correlation coefficient of 0.72 and a frequency domain cross-correlation of 0.65, resulting in a combined correlation strength of 0.68.

[0029] Indicator combinations with correlation strengths exceeding a preset threshold (e.g., 0.6) are written to the correlation indicator table. The correlation indicator table records highly correlated pairs of health indicators, such as systolic and diastolic blood pressure, heart rate and blood oxygen saturation. This correlation information can be used to assess health status, predict disease risks, or develop personalized health recommendations. For example, identifying a correlation strength of -0.65 between a user's blood sugar level and exercise level indicates that increasing exercise may help lower blood sugar. This information will be recorded in the correlation indicator table and may trigger corresponding health recommendations.

[0030] In this embodiment, by analyzing the rate of change of each indicator in the user's health data and dynamically adjusting the time series sampling window, refined sampling of the data fluctuation area is achieved, effectively improving the ability to capture key change information. At the same time, a multi-scale time series feature parameter system is constructed using the fluctuation amplitude ratio, fluctuation duration and waveform eigenvalues, and multi-scale decomposition is performed through the feature decomposition basis, thereby enhancing the ability to express the changing laws of complex health data. By introducing signal entropy and frequency domain cross-correlation, accurate modeling of the time-frequency joint characteristics between indicators is achieved, effectively reducing the impact of sampling bias and feature drift on the analysis results. The final constructed correlation index table can accurately identify combinations with stable linkage relationships between indicators, providing a more timely and accurate correlation basis for subsequent causal network construction and health intervention strategy formulation, thereby improving the overall health management system's perception and response efficiency to dynamic changes.

[0031] In an optional embodiment, calculating the signal entropy value of the time series characteristic parameters to construct a frequency component matrix, generating an eigendecomposition basis according to the frequency component matrix, and performing multi-scale decomposition on the time series characteristic parameters using the eigendecomposition basis to obtain the characteristic change sequence includes: The time series characteristic parameter sequence is segmented, and the reweighted Shannon entropy value of each segment is calculated. The normalized information entropy sequence is constructed by modifying the sequence using the adaptive weighting coefficient of the signal fluctuation amplitude. The normalized information entropy sequence is arranged according to the time scale to form an entropy distribution matrix. Constructing a frequency component matrix using the entropy distribution matrix, extracting entropy changes at adjacent time scales in the entropy distribution matrix, dividing the distribution intervals of the frequency components based on the entropy changes, calculating the mutual information between the frequency components in different distribution intervals, and writing the frequency component combinations whose mutual information exceeds a preset coupling threshold into a frequency coupling table; Extracting singular values ​​from the frequency component matrix, arranging the singular values ​​in descending order of energy contribution to obtain dominant frequency components, constructing the dominant frequency components as reference basis functions, adjusting the reference basis functions according to the frequency coupling table, minimizing reconstruction errors through iterative optimization, and generating an eigendecomposition basis; The feature decomposition basis is used to perform multi-scale decomposition on the time series feature parameters, the start and end positions of the decomposition interval are determined based on the frequency coupling table, a matching pursuit operation with an adaptive step size is performed in each decomposition interval, and the scale features obtained by decomposition are reorganized in chronological order to generate a feature change sequence.

[0032] Figure 2 This is a schematic diagram of the multi-scale decomposition and analysis system architecture for time series characteristic parameters. In a specific implementation method, a frequency component matrix is ​​constructed by calculating the signal entropy value of the time series characteristic parameters, and a characteristic decomposition basis is generated based on the frequency component matrix. The characteristic decomposition basis is used to perform multi-scale decomposition on the time series characteristic parameters to obtain a characteristic change sequence.

[0033] The time series feature parameter sequence is segmented, and a time window length of 256 data points is selected, with an overlap rate of 50% for adjacent windows. Within each time window, the reweighted Shannon entropy value is calculated. Specifically, the data points within the time window are divided into 8 level intervals according to the amplitude size. The probability of occurrence of the data points within each interval is counted, and the weighting coefficient is set according to the amplitude size. The interval with larger amplitude is given a higher weight. For example, the weight of the interval with the largest amplitude is set to 2.5, the weight of the interval with the smallest amplitude is 0.5, and the intermediate interval changes linearly. The original Shannon entropy calculation result is multiplied by the corresponding weight coefficient to obtain the reweighted Shannon entropy value. To achieve normalization, the entropy value of all windows is subtracted from the minimum entropy value and then divided by the entropy value range to obtain a normalized information entropy sequence with a value range of 0 to 1. The normalized information entropy sequence is arranged according to the time scale to form an entropy distribution matrix. The matrix dimension is the number of time windows multiplied by the number of scales. In this embodiment, the number of scales is set to 10.

[0034] The frequency component matrix is ​​constructed using the entropy distribution matrix. The entropy change is extracted by calculating the entropy difference between adjacent time scales. For example, for scale j and scale j+1, the entropy difference at each time point is calculated to form an entropy change sequence. Based on the distribution characteristics of the entropy change, the entropy change is divided into three frequency component intervals: high, medium, and low. The high-frequency interval corresponds to the portion with an absolute entropy change greater than 0.15, the low-frequency interval corresponds to the portion with an absolute entropy change less than 0.05, and the medium-frequency interval corresponds to the remainder. The mutual information between different frequency component intervals is calculated using an entropy-based mutual information calculation method. Each frequency component interval is treated as a random variable, and its joint distribution and marginal distribution are statistically analyzed to calculate the mutual information value. A mutual information threshold is set to 0.3. When the mutual information between two frequency component intervals exceeds this threshold, they are considered to be coupled, and their combined information is recorded in a frequency coupling table. The frequency coupling table stores data in a key-value pair format, where the key is the frequency component identifier and the value is a list of coupled frequency components.

[0035] Singular values ​​are extracted from the frequency component matrix and processed using the singular value decomposition method. The energy contribution of each singular value is calculated as the square of the singular value divided by the sum of the squares of all singular values. The values ​​are sorted in descending order of energy contribution, and the singular vectors corresponding to the top K singular values ​​with a cumulative contribution of 95% are selected as the dominant frequency components. In this embodiment, the value of K is 5. These dominant frequency components are constructed into a set of benchmark basis functions. The benchmark basis functions are adjusted according to the frequency coupling table. For frequency components with coupled relationships, the corresponding basis functions are weighted and combined. The weight coefficients are determined based on the mutual information value; higher mutual information values ​​give greater weights. The reconstruction error is minimized through iterative optimization using gradient descent, with a learning rate of 0.01 and a maximum number of iterations of 100. Iterations are terminated early when the rate of change of the reconstruction error is less than 0.001. The basis function parameters are continuously adjusted during the optimization process until the reconstruction error is minimized. The resulting eigendecomposition basis consists of the optimized set of basis functions, each representing a specific frequency component.

[0036] The time series feature parameters are decomposed at multiple scales using an eigendecomposition basis. The start and end positions of the decomposition intervals are determined based on a frequency coupling table, and the time series data is divided into multiple decomposition intervals. For frequency combinations with high mutual information, the same decomposition interval is used. Within each decomposition interval, a matching pursuit operation with an adaptive step size is performed. The initial step size is set to 1 / 10 of the decomposition interval length. The step size is dynamically adjusted based on the reconstruction error, decreasing when the error is large and increasing when the error is small. The step size range is limited to 0.5 to 2 times the initial step size. During the matching pursuit process, the basis function with the largest inner product with the residual signal is selected for projection. The projection coefficients are calculated, and the projected component is subtracted from the residual signal to generate a new residual signal. This process is repeated until the residual energy is less than 5% of the original signal energy or the maximum number of iterations, 100, is reached. The decomposed features at each scale are reassembled in chronological order to generate a feature change sequence. For overlapping regions, a linear weighting method is used for smoothing the transition, with data points near the center of the interval receiving higher weights and data points near the edges receiving lower weights, ensuring the temporal continuity of the feature change sequence.

[0037] In this embodiment, by introducing the reweighted Shannon entropy to construct a normalized entropy distribution matrix, the joint expression of information complexity and fluctuation intensity in the time series characteristics of health indicators is achieved, and the adaptability of feature modeling to dynamic non-stationary signals is enhanced. A frequency component matrix is ​​constructed based on the entropy value change, and high-coupling frequency components are identified through mutual information analysis, effectively revealing the potential collaborative variation pattern under multiple time scales. Further combining singular value extraction and energy sorting, a representative dominant frequency basis function is constructed, and the reconstruction error is iteratively optimized through the coupling relationship to generate a decomposition basis that can accurately characterize multi-scale dynamic characteristics. Finally, matching pursuit decomposition is performed under the guidance of this basis, which not only improves the resolution and stability of feature extraction, but also ensures the complete retention of key changes. Overall, this scheme significantly improves the resolution of multi-scale feature changes in health indicators, providing more accurate input data for subsequent state identification and risk warning.

[0038] In an optional embodiment, an indicator influence network is constructed based on the correlation indicator table. By introducing a dynamic compensation signal, an interference suppression test is performed on the correlation indicators in the indicator influence network. The attenuation coefficient and distortion degree during the signal transmission process are calculated. The indicator combination with a stable attenuation coefficient and a distortion degree below a preset threshold is written into the causal relationship table. The signal transmission characteristics and the stabilization action range are recorded in the causal relationship table, including: An indicator influence network is constructed based on the association indicator table, wherein nodes represent health indicators and edges between nodes represent association strengths. The delay time of signal transmission on the edges is calculated, an adjustment period of a time-varying weight is determined based on the delay time, the time-varying weight is updated based on the signal fluctuation frequency within the adjustment period, and the topology of the indicator influence network is optimized based on the time-varying weight; Selecting related indicators in the optimized indicator influence network, injecting a test signal between the related indicators, detecting a transmission distortion of the test signal within a phase calibration interval, and generating a dynamic compensation signal based on the time-varying weight and the transmission distortion; Superimposing the dynamic compensation signal on the test signal to obtain a compensated test signal, extracting the instantaneous frequency characteristics of the compensated test signal, calculating the frequency offset relative to the original test signal, and obtaining the attenuation coefficient and distortion degree of signal transmission according to the frequency offset; The correlation index pair of the attenuation coefficient fluctuation amplitude less than the first threshold and the distortion degree less than the second threshold is written into the causal relationship table, the stable action interval is determined based on the frequency offset, and the signal transmission characteristics within the stable action interval are recorded in the causal relationship table.

[0039] In the implementation method of constructing an indicator influence network based on an associated indicator table, first, historical health indicator data is obtained from the database and an associated indicator table is extracted. The associated indicator table contains multiple health indicator pairs and their correlation coefficients, such as blood sugar and insulin, blood pressure and heart rate, and other indicator pairs. These health indicators are used as network nodes, and the correlation coefficients are used as the weight values ​​of the connecting edges between nodes to construct the initial indicator influence network. In order to calculate the delay time of signal transmission, the time series analysis method is used to obtain the average delay time by comparing the time difference between the changes in the values ​​of the two associated indicators. For example, the delay time for increased insulin secretion after blood sugar rises is an average of 15 minutes. Based on this delay time, the system sets the adjustment period of the time-varying weight to twice the delay time, that is, 30 minutes.

[0040] During each adjustment cycle, the fluctuation frequency of the indicator is monitored, and changes in the indicator value are recorded at a frequency of 10 samples per minute. When the fluctuation frequency of indicator A is 3 times per hour and the fluctuation frequency of indicator B is 2 times per hour, the system calculates the relative fluctuation ratio between the two as 1.5 and updates the time-varying weight of the connection edge accordingly. For example, if the initial weight is 0.75, the updated time-varying weight becomes 0.75×(1+0.1×1.5)=0.8625. In this way, the system optimizes the network topology based on the time-varying weight, removes connections with weights below 0.3, and retains strongly associated indicator pairs.

[0041] In the optimized indicator influence network, pairs of related indicators were selected for interference rejection testing. For the blood glucose and insulin pair, a sinusoidal test signal with an amplitude of 2.0 and a frequency of 0.05 Hz was injected. Within a 60-second phase calibration interval, the test signal's transmission distortion, including phase offset and amplitude attenuation, was detected. When a phase offset of 12 degrees and an amplitude attenuation of 15% were detected, a dynamic compensation signal was generated based on a time-varying weight of 0.8625 and transmission distortion parameters. The compensation signal calculation takes into account a phase correction factor of 0.85 and an amplitude gain factor of 1.18, generating a compensation waveform with a phase advance of 12 degrees and an amplitude gain of 15%.

[0042] The dynamic compensation signal is superimposed on the original test signal to generate a compensated test signal. Using short-time Fourier transform technology, the instantaneous frequency characteristics of the compensated test signal are extracted, and the spectrum is calculated every 10 milliseconds. Compared to the nominal frequency of the original test signal of 0.05 Hz, the system calculates a frequency offset ranging from -0.002 Hz to +0.003 Hz. Based on this frequency offset, the signal transmission attenuation coefficient is calculated to be 0.92, and the degree of distortion is 0.08. The system repeats the test 20 times, and the mean and standard deviation of the attenuation coefficient are recorded, which are 0.93 and 0.02, respectively.

[0043] When the fluctuation amplitude of the attenuation coefficient 0.02 is less than the first threshold 0.05, and the distortion degree 0.08 is less than the second threshold 0.1, the system writes the associated indicator pair into the causal relationship table. Based on the distribution of frequency offset, the system determines that the stable action range is 4.0-7.0mmol / L for blood glucose and 10-25μIU / mL for insulin. Signal transmission characteristics are recorded within this range, including an average delay time of 15 minutes, an attenuation coefficient of 0.93, a distortion degree of 0.08, and compensation parameters of a phase correction factor of 0.85 and an amplitude gain factor of 1.18. These data are fully recorded in the causal relationship table to provide data support for the prediction and intervention of subsequent health indicators.

[0044] The above process was repeated for more pairs of indicators, such as blood pressure and heart rate, and weight and metabolic rate. For blood pressure and heart rate, the system injected a test signal with a frequency of 0.03Hz, resulting in an attenuation coefficient of 0.88 and a distortion degree of 0.12. Because the distortion degree exceeded the threshold of 0.1, this indicator pair was not entered into the causal relationship table. In this way, health indicator pairs with stable causal relationships were screened and their action characteristics were accurately recorded, providing a reliable predictive model foundation for the health management system. Through dynamic compensation and rigorous testing, this method ensures the accuracy and stability of causal relationship judgments and avoids the spurious correlation problem in traditional correlation analysis.

[0045] By constructing an indicator influence network with health indicators as nodes and correlation strength as edges, and dynamically adjusting time-varying weights based on signal transmission delays, adaptive modeling of the dynamic correlation structure between indicators is achieved, enhancing the network's ability to express non-stationary health data changes. Introducing test signals and identifying transmission distortion within the phase calibration interval enables the system to proactively identify interference paths and signal distortion characteristics, improving the credibility of causal inference. Through the construction of compensated test signals and analysis of frequency characteristics, the true signal transmission relationship between indicators is effectively quantified, avoiding false associations caused by statistical correlation alone. Further, using attenuation coefficient stability and distortion level as screening criteria, pairs of indicators with stable transmission characteristics and causal effects are accurately selected, providing a solid foundation for constructing a causal relationship table. Overall, this solution improves the accuracy and stability of causal identification between health indicators, providing a more reliable logical basis for personalized health regulation strategies.

[0046] Figure 3This figure illustrates the analysis of frequency offset during signal transmission. The solid gray line in the figure represents the real-time frequency offset curve, showing the deviation from the nominal frequency (0.05Hz) during signal transmission. This curve exhibits a distinct fluctuation, with multiple significant changes throughout the test period. In particular, a significant downward offset of approximately -0.0025Hz occurs around 30 seconds, while a significant upward offset peaks near +0.002Hz around 80 seconds. The light gray band in the figure represents the stable operating range, which is the frequency offset range within which the system determines that signal transmission characteristics remain stable. This range is roughly between -0.001Hz and +0.001Hz. According to the technical description, this range corresponds to the physiological ranges of blood glucose 4.0-7.0mmol / L and insulin 10-25μIU / mL.

[0047] By observing the relationship between the offset curve and the stable action range, we found that the offset remained within the stable range for some time periods (such as 10-20 seconds and 50-60 seconds), but exceeded this range in other time periods. This fluctuation provides basic data for the system to calculate the attenuation coefficient (approximately 0.93) and distortion level (approximately 0.08) of signal transmission, and further used to determine whether a stable causal relationship exists between indicators.

[0048] In an optional embodiment, constructing a feature space of health status based on multimodal health data, identifying state migration paths in the feature space, determining risk trigger points in combination with signal transmission characteristics, and determining adjustment target values ​​by analyzing the coordinated change patterns of different modal data within the stabilization interval include: Dynamically normalize and preprocess the multimodal health data, calculate the information entropy based on the preprocessed data to obtain the mutual information value between modalities, construct a data association matrix based on the mutual information value, use the data association matrix to design an adaptive weight function, dynamically weight the preprocessed data to obtain fusion features, and generate a feature association table; Segmenting the fused features into sliding windows, extracting statistical features and rate-of-change features within each window to obtain a feature sequence, performing density clustering on the feature sequence in combination with a feature association table to obtain state clusters, analyzing the distribution characteristics of the state clusters to determine state transition regions, and constructing a feature space of health status based on the state transition regions; Extracting a state transition trajectory in the feature space, calculating the geometric features of the state transition trajectory to obtain a state transition point, analyzing the system response curve at the state transition point in combination with signal transfer characteristics, determining a risk trigger point based on the convergence of the system response curve, and generating a risk distribution map; The periodic characteristics and trend characteristics of multimodal health data are extracted within the stabilization interval to obtain a change feature sequence, the phase relationship and amplitude relationship of the change feature sequence are analyzed to identify the coordinated change pattern, the safe activity range is determined based on the coordinated change pattern and the risk distribution map, and the state point with the optimal stability within the safe activity range is selected as the adjustment target value.

[0049] Exemplarily, a causal relationship table is used for multi-source fusion to generate multimodal health data. The known causal relationship table records the dependencies and influences between physiological, behavioral and environmental data, such as the positive correlation between heart rate and activity intensity (for every 10% increase in activity intensity, the heart rate increases by an average of 12 beats / minute), the relationship between blood pressure and ambient temperature (for every 5°C increase in temperature, the systolic blood pressure decreases by an average of 3 mmHg), etc. Based on these known relationships, physiological data (heart rate, blood pressure, blood sugar, etc.), behavioral data (number of steps, activity duration, sleep quality, etc.) and environmental data (temperature, humidity, air pressure, etc.) are collected, and the correlation of the original data is adjusted according to the causal relationship table to generate a more complete multimodal health data set.

[0050] The collected multimodal data is pre-processed with dynamic normalization. The dimensions and numerical ranges of health data of different modalities vary significantly. For example, heart rate is usually in the range of 60-180 beats / minute, while blood sugar may be in the range of 3.9-11.1mmol / L. A dynamic normalization method is used to determine the reference range based on individual historical data, and map each modality data to the 0-1 interval. The dynamic normalization reference value will be dynamically adjusted according to the user status. For example, the upper limit of the heart rate reference can be adjusted to 160 beats / minute in exercise and 100 beats / minute in resting state. In actual processing, the heart rate data is normalized using 68-150 beats / minute as the reference range; the blood pressure data is normalized using 90 / 60-140 / 90mmHg as the reference range; and the ambient temperature is normalized using 15-35℃ as the reference range.

[0051] After normalization, the mutual information between different modalities is calculated. The information entropy is calculated by statistically analyzing the distribution characteristics of the data for each modality. For example, the information entropy for heart rate data might be 2.45, while the information entropy for blood pressure data might be 2.78. The joint distribution between modal pairs is then calculated to obtain the inter-modal mutual information. In a real-world example, the mutual information between heart rate and activity level might be 0.82, indicating a high correlation; the mutual information between heart rate and ambient temperature might be 0.35, indicating a moderate correlation; and the mutual information between blood pressure and sleep quality might be 0.58, indicating a significant correlation.

[0052] A data association matrix is ​​constructed based on the mutual information values. This matrix is ​​symmetric, with diagonal elements equal to 1 and off-diagonal elements representing the mutual information values ​​between corresponding modalities. For example, in an association matrix containing four modalities: heart rate, blood pressure, activity level, and ambient temperature, the correlation strength between heart rate and activity level is 0.82, and the correlation strength between heart rate and ambient temperature is 0.35.

[0053] An adaptive weighting function is designed using the data association matrix. This weighting function dynamically adjusts the weights of each modality based on the association matrix and the current user state. In the resting state, physiological data may be given a higher weight (heart rate weight 0.4, blood pressure weight 0.3); while in the exercise state, behavioral data may be given a higher weight (activity weight 0.5, heart rate weight 0.3). These weights are applied to the preprocessed data for weighted fusion, resulting in a fused feature vector. For example, a fused feature in the resting state may combine 70% physiological data and 30% environmental data; a fused feature in the exercise state may combine 60% behavioral data and 40% physiological data. Based on the fusion process, a feature association table is generated, recording the mapping relationship and weight distribution between the original and fused features.

[0054] Segment the fused features using a sliding window. Set the window size to 5 minutes and the step size to 1 minute to segment the continuous fused feature data. Extract statistical features within each window, including mean (e.g., average heart rate within 5 minutes: 82 beats / minute), standard deviation (e.g., standard deviation of heart rate: 4.3 beats / minute), peak value (e.g., maximum heart rate: 89 beats / minute), valley value (e.g., minimum heart rate: 76 beats / minute), etc. At the same time, extract the rate of change features, including rise rate (e.g., maximum heart rate rise rate: 2.5 beats / minute / second), fall rate (e.g., maximum heart rate fall rate: 1.8 beats / minute / second), and fluctuation frequency (e.g., heart rate fluctuation frequency: 0.12Hz). This process results in a feature sequence, with each time window corresponding to a set of feature vectors.

[0055] Density clustering is performed on the feature sequences using the feature association table. A density clustering algorithm is used, with core parameters including a neighborhood radius set to 0.15 and a minimum sample size set to 4. Through density clustering, similar health status points in the feature space are clustered into clusters, such as "resting state cluster," "light activity cluster," "moderate exercise cluster," and "high-intensity exercise cluster." These clusters reflect the distribution of users' health status at different activity intensities.

[0056] Analyze the distribution characteristics of state clusters and identify state transition regions. Low-density areas between clusters typically represent transition zones between states. For example, the transition zone between "resting state" and "light activity" includes a heart rate of 70-85 beats per minute and an activity level of 20-50 steps per minute. These transition zones are key areas of health state change and require focused monitoring. Based on the identified state transition regions, a feature space for health states is constructed. This multidimensional space encompasses the characteristic distributions of various health states and the state transition paths.

[0057] State transition trajectories are extracted from the constructed feature space. State transition trajectories are formed by connecting adjacent state points in the time series. Geometric features of the trajectories are calculated, including curvature (e.g., a curvature of 0.75 at the inflection point of the trajectory), rate of change of direction (e.g., a rate of change of direction of 30 degrees per minute), and speed (e.g., a state change speed of 0.15 units per minute). Points where the geometric features of the trajectory change significantly are typically state transition points, such as the transition from rest to movement or from light activity to moderate activity.

[0058] Analyze the system response curve at the state transition point by combining signal transfer characteristics. The system response curve reflects the response of health indicators to changes in activity. For example, the response curve for a rise in heart rate after starting exercise might show an initial heart rate of 65 beats / minute, rising to 95 beats / minute after 1 minute, and stabilizing at 115 beats / minute after 3 minutes. Analyze the convergence of the response curve, including the rise time (the time from the start of the change to reaching 90% of the target value, such as 2.3 minutes), the stabilization time (the time required to reach a stable state, such as 3.5 minutes), and the overshoot (the maximum amplitude exceeding the stable value, such as 8%).

[0059] Risk trigger points are determined based on the convergence of the system response curve. A normal response curve should converge to a stable value within a reasonable time. For example, after starting moderate-intensity exercise, the heart rate should stabilize within 2-3 minutes. If the response curve exhibits abnormal characteristics, such as a continuous increase in heart rate without convergence, an abnormal rate of increase (such as more than 40 beats per minute), or an excessive overshoot (such as more than 20%), it is identified as a potential risk trigger point. Based on these risk trigger points, a risk distribution map is generated to show the distribution of risk levels under different activity types, intensities, and environmental conditions.

[0060] Within the stabilization interval, we extract cyclical and trend features from multimodal health data. Cyclic features include daily variations (e.g., heart rate averages 62 beats / minute between 6:00 AM and 8:00 AM, 78 beats / minute between 2:00 PM and 4:00 PM) and weekly variations (e.g., average activity of 8,500 steps / day on weekdays and 12,000 steps / day on weekends). Trend features include long-term trends, such as a decrease in resting heart rate from an average of 68 beats / minute to 65 beats / minute over a month, indicating possible improvement in cardiopulmonary function. These features constitute a sequence of variation features.

[0061] Analyze the phase and amplitude relationships of the characteristic sequence of changes to identify synergistic change patterns. These patterns reflect the changing relationships between multiple health indicators, such as the synergistic relationship between increased activity and heart rate (with a phase difference of approximately 30 seconds and an amplitude ratio of approximately 1:0.4, meaning a 100% increase in activity results in a approximately 40% increase in heart rate), and the synergistic relationship between increased ambient temperature and skin temperature (with a phase difference of approximately 15 minutes and an amplitude ratio of approximately 1:0.3).

[0062] Determine the safe activity range based on the coordinated change pattern and risk distribution map. The safe activity range refers to the range of activity intensity and duration that users can perform without triggering health risks. For example, for a middle-aged user, the safe heart rate range may be 60-135 beats / minute, the safe activity intensity range may be 3500-7500 steps per hour, and the safe continuous exercise time may be 30-60 minutes. Within the determined safe activity range, the state point with optimal stability is selected as the adjustment target value. Optimal stability refers to the state point with the smallest system response fluctuation and the fastest recovery speed. For example, a moderate-intensity exercise state with a heart rate maintained at 110-115 beats / minute may be the optimal stability point for a certain user.

[0063] Existing technologies for health state identification and risk prediction typically rely on static feature extraction and single-modal data analysis, making it difficult to accurately reflect the coupling relationships between multidimensional health data and their impact on state evolution. This results in coarse state segmentation, delayed risk trigger identification, and inability to dynamically perceive and precisely regulate individual health states. To address this issue, the proposed technical solution constructs a data association matrix and introduces an adaptive weighting function to achieve dynamic fusion of multimodal health data, effectively improving the integrity and expressiveness of intermodal information transfer. Furthermore, by combining sliding window feature extraction with density clustering methods, it accurately identifies transition regions within health states and constructs a more continuous and interpretable health state feature space. Furthermore, by analyzing the geometric characteristics of state transition trajectories and combining them with the convergence of system response curves, it accurately identifies risk trigger points. Compared to existing technologies that rely solely on critical value judgment, this application significantly improves the foresight and rationality of risk prediction. Furthermore, by analyzing the coordinated change patterns of multimodal data within the stable action range, it proposes a safe activity range identification and stable state point screening mechanism, overcoming the limitations of traditional methods, which often involve arbitrary selection of regulatory targets and poor generalization. This improvement takes improving the scientificity and stability of health status regulation as its starting point, realizes the transformation of health management strategy generation from static response to dynamic coordination, and enhances the system's adaptability and guidance capabilities to individual health changes.

[0064] In an optional embodiment, determining the scope of safety activities based on the coordinated change pattern and the risk distribution map includes: Acquire phase information of multimodal health data, calculate the phase coupling strength between the data, construct a coupling matrix based on the ratio of the maximum to the minimum phase coupling strength, mark the indicator combinations in the coupling matrix whose ratio exceeds a preset coupling threshold as strong coupling pairs, and extract the change trajectory of the strong coupling pairs within the stable working range to construct a collaborative change envelope; Calculating the gradient value of the risk distribution map to obtain a risk gradient, identifying a risk change boundary region based on the risk gradient, and shrinking and adjusting the coordinated change envelope according to the position of the risk change boundary region and the value of the risk gradient to obtain a modified envelope; An optimization objective function is constructed using the modified envelope as a constraint condition, the optimization objective function is solved to obtain a multi-dimensional activity range, and a state transition constraint condition is constructed according to the boundary characteristics of the multi-dimensional activity range; The safe activity area is divided based on the state transition constraint condition, the state stability index is extracted from the safe activity area, and the safe activity area is optimized and adjusted according to the state stability index to obtain the final safe activity range.

[0065] This embodiment provides a method for determining safe activity ranges based on coordinated change patterns and risk distribution maps. This method uses multimodal health data analysis, phase coupling calculation, risk boundary identification, and safe zone optimization to accurately define safe ranges for human activity.

[0066] Among them, obtaining phase information of multimodal health data is the first step in determining the safe range of activity. Phase information reflects the temporal relationship and change pattern between different data. Signal processing methods such as Hilbert transform are used to extract the instantaneous phase of each data, such as the phase change of the heart rate signal over time and the phase change trend of the activity signal. For heart rate data, the phase may be expressed as a fluctuation cycle throughout the day, with the phase being 0° at 6 am, 90° at noon, 180° at 6 pm, and 270° at midnight. For activity data, the phase may be expressed as an alternating cycle of activity and rest, with the phase being 0° at the beginning of activity, 90° at peak activity, 180° when activity decreases, and 270° at rest.

[0067] Calculating the phase coupling strength between data is the key to understanding the synergistic relationship of multimodal data. Phase coupling strength is measured by calculating the distribution consistency of the phase difference between two data sequences. For example, the distribution of the phase difference between heart rate and activity level during the day is calculated. If the phase difference between the two is stable at about 45° (that is, the change in activity level leads the change in heart rate by about 45°), it indicates that the two have a strong phase coupling relationship. In actual calculations, the phase difference between heart rate and activity level can be counted every hour within 24 hours. If the standard deviation of the phase difference is less than 20°, it is considered that the two have a stable phase coupling relationship. The phase coupling strength is calculated for all modal data pairs to obtain a coupling strength matrix.

[0068] A coupling matrix is ​​constructed based on the ratio of the maximum to minimum phase coupling strength. The coupling matrix is ​​a symmetric matrix in which each element represents the phase coupling ratio between two corresponding data points. For example, the maximum phase coupling strength between heart rate and activity level may be 0.85 (when the two change in highly synchronized manner) and the minimum may be 0.15 (when the relationship between the two changes is weakest), with a ratio of 5.67. The maximum phase coupling strength between heart rate and ambient temperature may be 0.45 and the minimum may be 0.25, with a ratio of 1.8. These ratios constitute the elements of the coupling matrix. In actual cases, a user's coupling matrix may include the following coupling ratios: heart rate-activity level (5.67), heart rate-blood pressure (4.23), heart rate-ambient temperature (1.8), activity level-blood pressure (3.15), activity level-ambient temperature (1.35), and blood pressure-ambient temperature (1.62).

[0069] Indicator combinations in the coupling matrix whose ratios exceed a preset coupling threshold are labeled as strongly coupled pairs. If the coupling threshold is set to 3.0, then in the above example, the three pairs of indicators—heart rate-activity level (5.67), heart rate-blood pressure (4.23), and activity level-blood pressure (3.15)—are labeled as strongly coupled. These strongly coupled pairs reflect the most critical synergistic relationships in health status changes and require special attention.

[0070] The trajectories of strongly coupled pairs within the stable operating range are extracted to construct a coordinated change envelope. The stable operating range refers to the data range in which the user is in a healthy state. Within this stable operating range, the trajectories of strongly coupled pairs are extracted, such as the heart rate-activity relationship at different activity intensities: resting state (heart rate 65 beats / minute, activity 0 steps / minute), light activity (heart rate 85 beats / minute, activity 60 steps / minute), moderate activity (heart rate 110 beats / minute, activity 120 steps / minute), and high-intensity activity (heart rate 140 beats / minute, activity 180 steps / minute). These data points constitute the heart rate-activity relationship trajectory. Similarly, the heart rate-blood pressure and activity level-blood pressure relationship trajectories are extracted. Based on these trajectory points, a coordinated change envelope is constructed. This envelope is a multidimensional surface that encompasses the range of variation of all strongly coupled pairs within the stable operating range.

[0071] The risk gradient is calculated by calculating the gradient of the risk distribution map. The risk distribution map reflects the health risk level under different state combinations. The gradient of each dimension of the risk distribution map is calculated to represent the rate of change of risk level with state changes. For example, when the heart rate increases from 120 beats / minute to 140 beats / minute, the risk level increases from 0.3 to 0.7, with a gradient of 0.02 / beat / minute; when the activity level increases from 150 steps / minute to 180 steps / minute, the risk level increases from 0.4 to 0.8, with a gradient of 0.013 / step / minute.

[0072] Identify risk change boundary areas based on the risk gradient. Areas with larger risk gradients are typically boundary areas where risk levels change significantly. Set the gradient threshold to 0.01 / unit. When the risk gradient exceeds the threshold, the area is marked as a risk change boundary area. In the above example, heart rates in the 120-140 beats / minute range and activity levels in the 150-180 steps / minute range are identified as risk change boundary areas. These boundary areas are often critical areas where health status transitions from safe to risky and require special attention.

[0073] The covariation envelope is contracted and adjusted based on the location of the risk change boundary and the value of the risk gradient to obtain a revised envelope. The principle of contraction adjustment is to apply a greater degree of contraction in areas with large risk gradients, ensuring that the revised envelope maintains a safe distance from high-risk areas. For example, for the heart rate-activity covariation envelope, contraction is performed in the heart rate range of 120-140 beats / minute and the activity range of 150-180 steps / minute. The adjusted envelope reduces the upper limit of heart rate from 140 beats / minute to 130 beats / minute and the upper limit of activity from 180 steps / minute to 165 steps / minute. Similarly, the covariation envelopes of other strongly coupled pairs are adjusted to obtain the revised envelopes.

[0074] The modified envelope is used as a constraint to construct an optimization objective function. The optimization goal is to maximize the user's activity space while ensuring safety. The objective function includes safety constraints (staying within the modified envelope) and mobility requirements (maximizing the range of activity). Described mathematically, the goal is to maximize the volume of the range of activity while ensuring that all state points lie within the modified envelope and that the risk level does not exceed a safety threshold (e.g., 0.5).

[0075] Solve the optimization objective function to obtain a multidimensional activity range. Using numerical optimization methods (such as gradient descent and genetic algorithms), the objective function is solved to determine the maximum activity range that satisfies safety constraints. For the example above, the optimization results might be: a heart rate range of 60-125 beats / minute, an activity range of 0-160 steps / minute, and a blood pressure range of 90 / 60-135 / 85 mmHg. This multidimensional activity range constitutes the boundary of the user's safe activity.

[0076] State transition constraints are constructed based on the boundary characteristics of the multi-dimensional activity range. These constraints describe the rules that users should follow when transitioning between different activity states to ensure the safety of the transition process. For example, when transitioning from a resting state to a moderately active state, the heart rate should not increase by more than 20 beats per minute, and the increase in activity level should not exceed 40 steps per minute. When returning from a high-intensity activity state to a resting state, the heart rate should decrease for at least 10 minutes before returning to the resting level. These constraints ensure a smooth transition process and avoid health risks caused by drastic changes.

[0077] Divide the safe activity area based on state transition constraints. The multidimensional activity range is further subdivided into different safe activity areas based on the state transition constraints. Each area corresponds to a specific activity type and intensity level. For example, it can be divided into a resting area (heart rate 60-75 beats / minute, activity volume 0-20 steps / minute), a light activity area (heart rate 75-95 beats / minute, activity volume 20-80 steps / minute), a moderate activity area (heart rate 95-115 beats / minute, activity volume 80-130 steps / minute) and a high-intensity activity area (heart rate 115-125 beats / minute, activity volume 130-160 steps / minute).

[0078] Extract state stability indicators within the safe activity zone. State stability indicators reflect the user's adaptability and stability within a specific activity zone. By analyzing historical user data, indicators such as state volatility (such as heart rate standard deviation), recovery speed (such as the time required for heart rate to return to resting levels after activity), and sustained tolerance (such as the maximum time a specific activity intensity can be maintained) are calculated within each activity zone. For example, in the moderate activity zone (heart rate 95-115 beats / minute), a user's heart rate standard deviation may be 3.2 beats / minute, heart rate recovery time may be 4.5 minutes, and sustained tolerance time may be 45 minutes.

[0079] The safe activity area is optimized and adjusted according to the state stability index to obtain the final safe activity range. The principle of optimization and adjustment is to expand the area with high stability and shrink the area with low stability. If the user shows high stability in the moderate activity area (small heart rate fluctuations, fast recovery, and long tolerance time), the range of this area can be appropriately expanded, for example, the upper limit of the heart rate can be increased from 115 beats / minute to 118 beats / minute; if the user has poor stability in the high-intensity activity area (large heart rate fluctuations and slow recovery), the range of this area needs to be shrunk, for example, the upper limit of the heart rate can be reduced from 125 beats / minute to 120 beats / minute. After these adjustments, a personalized safe activity range is finally obtained, which fully takes into account the user's physiological characteristics, activity patterns and environmental influences, and provides users with precise health management guidance.

[0080] In this embodiment, by incorporating phase coupling analysis of multimodal health data, constructing a coupling matrix, and identifying strongly coupled pairs, high-precision modeling of the synergistic relationships between different health indicators is achieved. This overcomes the limitations of existing technologies that rely on static correlations and ignore temporal coupling. A synergistic change envelope is constructed based on the trajectory of changes in strongly coupled pairs, which can truly reflect the linkage characteristics of multiple indicators in an individual's stable health state. Furthermore, by combining the gradient information of the risk distribution map, the risk boundary is dynamically identified and the envelope is adjusted. The resulting modified envelope retains the stable characteristics of synergistic change while avoiding potential high-risk areas. Compared with traditional methods that use a single risk indicator for demarcation, it is more holistic and accurate. Based on this, an optimization objective function is constructed to generate a multidimensional activity range that satisfies the synergistic constraints. The safe area is iteratively optimized using a state stability indicator, ensuring that the final safe activity range is highly feasible and stable. Overall, this solution improves the dynamic adaptability and individual responsiveness of safe zone demarcation, providing a more effective intervention boundary and behavioral guidance foundation for precision health management.

[0081] Figure 4 This is a schematic diagram comparing the phase coupling strength ratios of multimodal health data. The horizontal dashed line in the chart marks the preset coupling threshold of 3.0. Indicator combinations exceeding this threshold are considered strongly coupled. As can be seen from the data, the current user exhibited the highest phase coupling strength ratio (5.7) for the heart rate-activity combination, significantly higher than the reference group (4.2) and the healthy control group (3.9). The coupling ratio for the heart rate-blood pressure combination was also high, reaching 4.2 for the current user, 3.8 for the reference group, and 3.5 for the healthy control group, all exceeding the coupling threshold. In the activity-blood pressure combination, the current user's coupling ratio was 3.2, slightly above the threshold, while the reference group (2.9) and the healthy control group (2.6) were below the threshold. The heart rate-ambient temperature combination exhibited lower coupling strength, with the three data sets showing values ​​of 1.8, 2.1, and 2.4, respectively, falling short of the strong coupling standard.

[0082] These data clearly show that the current users' strong coupling pairs are mainly concentrated in three groups of indicators: heart rate-activity, heart rate-blood pressure, and activity-blood pressure. These strong coupling pairs will be used to construct a coordinated change envelope, and further combined with the gradient information of the risk distribution map to determine the safe activity range.

[0083] In an optional embodiment, generating a health management plan including an adjustment sequence, an adjustment amplitude, and an adjustment period by calculating the deviation between the current indicator value and the adjustment target value includes: Obtaining the current indicator value and the adjustment target value, calculating the standardized deviation between the current indicator value and the adjustment target value, constructing an indicator impact matrix based on the standardized deviation, performing eigendecomposition on the indicator impact matrix to obtain eigenvectors, extracting the influence transfer path based on the eigenvectors to determine the adjustment order, and generating a sequence matrix; Acquire historical adjustment data, calculate an indicator response curve in combination with the sequence matrix, determine an adjustment step threshold based on a rate of change of the indicator response curve, adopt an initial adjustment step when the standardized deviation is greater than the adjustment step threshold, and adjust the adjustment step based on real-time feedback data when the standardized deviation is less than the adjustment step threshold, thereby generating an adjustment amplitude table; Collecting real-time indicator data, calculating the rate of change and the trend of change of the real-time indicator data, identifying state change points based on the rate of change and the trend of change, determining the adjustment period according to the time distribution of the state change points, and generating a period distribution table; The adjustment sequence in the sequence matrix is ​​grouped according to the indicator influence matrix, the adjustment step is determined in each group according to the adjustment amplitude table, the adjustment time is set in combination with the cycle distribution table, and a health management plan including the adjustment sequence, adjustment amplitude and adjustment cycle is generated.

[0084] The present invention provides a health management program generation method, which generates a health management program including an adjustment sequence, an adjustment amplitude and an adjustment cycle by calculating the deviation between a current indicator value and an adjustment target value.

[0085] First, the current indicator values ​​are obtained. These values ​​are collected in real time by health monitoring equipment, including physiological indicators (such as heart rate 78 beats / minute, blood pressure 135 / 85 mmHg, and blood glucose 6.2 mmol / L), behavioral indicators (such as an average daily step count of 8,500, sleep duration of 6.5 hours, and sedentary time of 4.2 hours), and environmental indicators (such as indoor temperature 24°C, humidity 65%, and air quality index 85). The target value is adjusted by selecting the optimal stability point within the safe activity range, such as a target heart rate of 70 beats / minute, a target blood pressure of 120 / 80 mmHg, and a target daily step count of 10,000 steps.

[0086] Calculating the standardized deviation between the current indicator value and the adjusted target value makes indicators of different dimensions comparable. The standardized deviation is calculated by dividing the original deviation by the standard deviation or normal fluctuation range of the corresponding indicator. For example, if the current heart rate is 78 beats / minute, the target value is 70 beats / minute, and the normal fluctuation range of heart rate is 10 beats / minute, then the standardized deviation is 0.8; if the current blood pressure is 135 / 85 mmHg and the target value is 120 / 80 mmHg, the normal fluctuation range of systolic pressure is 20 mmHg, and the normal fluctuation range of diastolic pressure is 10 mmHg, then the standardized deviation of systolic pressure is 0.75, and the standardized deviation of diastolic pressure is 0.5; if the current average daily step count is 8,500 steps and the target value is 10,000 steps, and the normal fluctuation range of step count is 2,000 steps, then the standardized deviation is -0.75.

[0087] The indicator influence matrix constructed based on standardized deviation reflects the mutual influence relationship between various indicators. The indicator influence matrix is ​​a square matrix in which each element represents the degree of influence of the row indicator on the column indicator. The degree of influence is determined by analyzing the correlation and causal relationship of indicator changes in historical data. For example, in an influence matrix containing four indicators: heart rate, blood pressure, number of steps, and sleep duration: the degree of influence of heart rate on blood pressure is 0.7 (for every 1 standard unit change in heart rate, the average change in blood pressure is 0.7 standard units); the degree of influence of number of steps on heart rate is 0.6; the degree of influence of sleep duration on heart rate is -0.5 (for every 1 standard unit increase in sleep duration, the heart rate decreases by 0.5 standard units on average); the degree of influence of sleep duration on blood pressure is -0.4, etc. Construct a complete impact matrix, such as: heart rate impact row [1, 0.7, 0.3, 0.2], blood pressure impact row [0.4, 1, 0.1, 0.1], step count impact row [0.6, 0.5, 1, 0.3], sleep duration impact row [-0.5, -0.4, 0.2, 1].

[0088] Perform eigendecomposition on the indicator impact matrix to obtain eigenvectors. Eigendecomposition is a method for extracting key influencing factors from matrix analysis. By calculating the eigenvalues ​​and eigenvectors of the indicator impact matrix, the most significant combinations of factors influencing the system can be identified. For example, performing eigendecomposition on the above impact matrix may yield four eigenvectors: the first eigenvector [0.6, 0.5, 0.5, 0.4] corresponds to an eigenvalue of 2.3, indicating the combined influence of all indicators; the second eigenvector [0.5, 0.4, -0.3, -0.7] corresponds to an eigenvalue of 1.5, indicating the opposing relationship between physiological and behavioral indicators; the third eigenvector [-0.2, 0.5, 0.7, -0.5] corresponds to an eigenvalue of 0.8; and the fourth eigenvector [0.6, -0.6, 0.3, -0.5] corresponds to an eigenvalue of 0.4. The larger the eigenvalue, the more significant the corresponding eigenvector's influence.

[0089] The influence transfer paths extracted from the eigenvectors determine the adjustment order. The size of the eigenvector components reflects the importance of each indicator in the impact pattern. By analyzing the structure of the eigenvectors, the influence transfer paths between indicators can be identified. For example, the first eigenvector [0.6, 0.5, 0.5, 0.4] shows that heart rate (0.6) has the greatest impact on the system, followed by blood pressure (0.5) and step count (0.5), and finally sleep duration (0.4). Combining the direct influence relationships in the indicator influence matrix, we can construct the influence transfer paths: step count → heart rate → blood pressure, and sleep duration → heart rate → blood pressure. Based on these paths, the adjustment order is determined as: step count and sleep duration (basic adjustment indicators) → heart rate (intermediate adjustment indicator) → blood pressure (final adjustment indicator). This adjustment order starts with the fundamental influencing factors and gradually adjusts to the final target indicator, resulting in better adjustment results. This adjustment order information is organized into a sequence matrix, where each row represents a adjustment path and each column represents a stage of adjustment.

[0090] Obtain historical adjustment data and combine it with the sequential matrix to calculate the indicator response curve. This historical adjustment data records the implementation and effectiveness of past adjustment measures. For example, increasing the average daily step count by 2,000 reduces the heart rate by an average of 5 beats per minute; increasing sleep duration by one hour reduces the heart rate by an average of 3 beats per minute. Combined with the adjustment paths in the sequential matrix, calculate the indicator response curve for each path. The indicator response curve describes the response of the target indicator to the adjustment measure, including characteristics such as response delay, response rate, and stability level. For example, the response curve for increasing step count to heart rate might show: one week after the step increase, the heart rate decreases by an average of 2 beats per minute; two weeks later, the heart rate decreases by an average of 4 beats per minute; and after four weeks, the heart rate decreases by an average of 5 beats per minute and stabilizes. Similarly, analyze the response curves of other adjustment paths.

[0091] The adjustment step threshold is determined based on the rate of change of the indicator's response curve. The rate of change reflects the indicator's sensitivity to the adjustment measure and is an important factor in determining the adjustment step size. By analyzing the slope of the response curve, the range and critical point where the response is most significant are determined. For example, the heart rate response curve for step count shows that when the step count increases between 1000 and 3000 steps, the heart rate variability decreases by approximately 2 beats / minute per 1000 steps. When the step count increases beyond 3000 steps, the heart rate variability decreases to 1 beat / minute per 1000 steps, indicating that the response curve flattens. Based on these characteristics, the adjustment step threshold is determined. For heart rate, the standardized deviation adjustment step threshold might be set at 0.5; for blood pressure, the adjustment step threshold might be set at 0.6; and for step count, the adjustment step threshold might be set at 0.7.

[0092] When the standardized deviation is greater than the adjustment step threshold, the initial adjustment step is adopted. The initial adjustment step is an adjustment starting point set based on experience, usually set to a range that can produce significant effects but does not cause over-adjustment. For example, when the standardized deviation of the heart rate is 0.8 (greater than the threshold of 0.5), the initial adjustment step is adopted: increase the average daily step count by 2000 steps and increase the sleep time by 0.5 hours. These initial adjustment steps are set based on the average response effect of historical data and can significantly adjust the heart rate towards the target value.

[0093] When the standardized deviation is less than the adjustment step threshold, the adjustment step is adjusted based on real-time feedback data. When the indicator is close to the target value, more precise adjustment is required to avoid over-adjustment and oscillation. By analyzing the user's actual response to the previous adjustment measures, the adjustment step is dynamically adjusted. For example, if the user increases the number of steps by 2000 and the heart rate drops by 6 times / minute (more significant than the expected 5 times / minute), it means that the user has a high responsiveness to the step adjustment, and the adjustment step can be reduced to 1500 steps later; if the user increases the sleep time by 0.5 hours and the heart rate only drops by 1 time / minute (lower than the expected 3 times / minute), it means that the responsiveness is low, and the adjustment step can be increased to 0.8 hours later. This adjustment of the adjustment step based on real-time feedback makes the health management plan more personalized and accurate. According to the determination result of the adjustment step, an adjustment amplitude table is generated to record the adjustment amplitude of different indicators under different deviation conditions.

[0094] Collect real-time indicator data and calculate the rate of change and trend of this real-time indicator data. Real-time data is continuously collected by health monitoring devices, such as heart rate, which is recorded once a minute, blood pressure, and step count, which is accumulated once an hour. Calculate the rate of change (the difference between the values ​​at adjacent time points divided by the time interval) and the trend of change (determined by the slope of the linear regression within the sliding window) for this time series data. For example, a user's heart rate data shows an average rate of change of +3 beats / minute per hour between 6 and 8 in the morning (an upward trend); an average rate of change of -2 beats / minute per hour between 2 and 4 in the afternoon (a downward trend); and a rate of change close to 0 between 8 and 10 in the evening (a stable trend).

[0095] Identify state change points based on change rate and change trend. A state change point is the point in time when an indicator changes from one change mode to another, usually manifested as a significant change in the change rate or a reversal of the change trend. By setting the change rate threshold and the trend change threshold, state change points can be automatically identified. For example, when the heart rate change rate changes from a positive value to a negative value, and the amplitude exceeds 3 times per minute, it is marked as a state change point; when the number of steps changes from a continuous increase to a continuous decrease, and the trend change exceeds 500 steps per hour, it is marked as a state change point. For the above user, the state change points that may be identified include: 8 o'clock in the morning (heart rate changes from rising to stable), 12 o'clock in the afternoon (heart rate changes from stable to falling), 5 o'clock in the afternoon (heart rate changes from falling to rising), etc.

[0096] The adjustment cycle is determined based on the temporal distribution of state change points. The temporal distribution of state change points reflects the natural cyclical changes in the user's physiology and behavior and is an important basis for formulating the adjustment cycle. By statistically analyzing the time intervals and patterns of state change points, the optimal adjustment time is determined. For example, if a user's heart rate shows state change points at 8:00 AM, 1:00 PM, and 7:00 PM every day, these times may be the best times to implement adjustment measures. If the user's activity level shows similar patterns from Monday to Friday but is significantly different on weekends, different adjustment plans may be needed for weekdays and weekends. Based on this analysis, a period distribution table is generated, recording the optimal adjustment time points and cyclical patterns for each indicator.

[0097] The adjustment order in the sequence matrix is ​​grouped according to the indicator impact matrix. Based on the influence relationship between indicators, the adjustment indicators are divided into multiple adjustment groups. The indicators in each group have similar influencing paths or affected paths. For example, the number of steps and sleep duration can be grouped together (basic behavior adjustment group), and the heart rate and blood pressure can be grouped together (physiological indicator adjustment group). This grouping method facilitates the development of a coordinated adjustment strategy and avoids mutual interference between different adjustment measures.

[0098] Within each group, the adjustment step size is determined based on the adjustment amplitude table. For each indicator within the adjustment group, the specific adjustment step size is determined by referring to the adjustment amplitude table. For example, in the basic behavior adjustment group, the adjustment step size for step count is 2,000 steps per day, and the adjustment step size for sleep duration is 0.5 hours per day. In the physiological indicator adjustment group, by indirectly affecting heart rate and blood pressure through adjusting basic behaviors, it is expected that the heart rate will decrease by 5 beats / minute and the blood pressure will decrease by 10 / 5 mmHg.

[0099] Set the adjustment time based on the period distribution table. Set the execution time for each adjustment measure based on the optimal adjustment time and periodic pattern recorded in the period distribution table. For example, measures to increase the number of steps can be scheduled for execution between 8 and 10 in the morning and 3 and 5 in the afternoon, with an increase of 1,000 steps each time. Measures to improve sleep can be scheduled for execution before 10 p.m., such as maintaining a quiet sleeping environment and avoiding the use of electronic devices. Heart rate and blood pressure monitoring can be set after waking up in the morning, during lunch break, and before bedtime to fully evaluate the adjustment effect.

[0100] Generate a health management plan that includes the adjustment sequence, adjustment amplitude and adjustment cycle. Integrate the above analysis and decision-making results into a complete health management plan, including: adjustment sequence (first adjust the number of steps and sleep duration, and observe the changes in heart rate and blood pressure), adjustment amplitude (increase the number of steps by 2,000 steps per day, and increase the sleep duration by 0.5 hours per day) and adjustment cycle (increase the number of steps twice a day, morning and evening, perform sleep improvement measures at night, and evaluate the overall effect once a week). The plan is presented in an intuitive and easy-to-understand form, such as: Phase 1 of the health management plan (4 weeks): Walk 1,000 steps every morning from 8 to 10 am, walk 1,000 steps from 3 to 5 pm, turn off electronic devices before 10 pm and keep the sleeping environment quiet. It is expected that the heart rate will drop from 78 beats / minute to 73 beats / minute, and the blood pressure will drop from 135 / 85 mmHg to 125 / 80 mmHg; conduct a mid-term evaluation 2 weeks after the implementation of the plan, and adjust subsequent measures based on the actual results.

[0101] In this embodiment, by constructing an indicator influence matrix and performing feature decomposition, the influence paths between health indicators are identified, thereby scientifically formulating the adjustment sequence, breaking through the subjective problem of setting the sequence based on traditional expert experience, and realizing the structuring and personalization of the adjustment process. The adjustment amplitude table is constructed by combining historical adjustment data and current standardized deviation, making the adjustment amplitude both responsive and controllable, avoiding the situation where the system oscillates due to too fast adjustment or the intervention is ineffective due to too slow adjustment. By dynamically analyzing the changing trends and state change points of real-time indicator data and extracting the distribution characteristics of the adjustment cycle, adaptive cycle regulation is achieved, replacing the static adjustment mode based on fixed cycles or uniform time intervals in the existing technology. Ultimately, by integrating the three elements of adjustment sequence, amplitude and cycle into a structural closed-loop health management strategy, the solution improves the accuracy and response efficiency of regulation, significantly enhances the stability and sustainability of health intervention, and meets the needs of personalized dynamic regulation under complex and changing health conditions.

[0102] A second aspect of an embodiment of the present invention provides a health management data processing system based on multimodal data, the system comprising: The first unit is used to receive user health data, extract the time series characteristic parameters of each indicator in the user health data, calculate the correlation coefficient between the time series characteristic parameters of different indicators, and write the indicator combination whose correlation coefficient exceeds a preset correlation threshold into the correlation indicator table; The second unit is used to construct an indicator influence network based on the correlation indicator table, perform interference suppression tests on the correlation indicators in the indicator influence network by introducing dynamic compensation signals, calculate the attenuation coefficient and distortion degree during signal transmission, write indicator combinations with stable attenuation coefficients and distortion degrees below a preset threshold into the causal relationship table, and record the signal transmission characteristics and stabilization range in the causal relationship table; The third unit is used to perform multi-source fusion of physiological, behavioral, and environmental data based on the causal relationship table to generate multimodal health data. Based on the multimodal health data, a feature space of health status is constructed. The state migration path is identified in the feature space. The risk trigger point is determined by combining the signal transmission characteristics. The adjustment target value is determined by analyzing the coordinated change pattern of different modal data within the stabilization interval. The fourth unit is used to generate a health management plan including the adjustment sequence, adjustment amplitude and adjustment cycle by calculating the deviation between the current indicator value and the adjustment target value, and dynamically optimize the health management plan based on real-time feedback during the indicator adjustment process.

[0103] According to a third aspect of an embodiment of the present invention, an electronic device is provided, including: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0104] According to a fourth aspect of an embodiment of the present invention, a computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the method described above is implemented.

[0105] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0106] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A health management data processing method based on multimodal data, characterized in that: include: Receive user health data, extract the time series characteristic parameters of each indicator in the user health data, calculate the correlation coefficient between the time series characteristic parameters of different indicators, and write the indicator combination whose correlation coefficient exceeds the preset correlation threshold into the correlation indicator table; An indicator influence network is constructed based on the correlation indicator table. By introducing a dynamic compensation signal, interference suppression tests are performed on the correlation indicators in the indicator influence network. The attenuation coefficient and distortion degree during signal transmission are calculated. The indicator combinations with stable attenuation coefficients and distortion degrees below a preset threshold are written into the causal relationship table. The signal transmission characteristics and stabilization range are also recorded in the causal relationship table. Based on the causal relationship table, multi-source fusion of physiological, behavioral, and environmental data is performed to generate multimodal health data. Based on the multimodal health data, a feature space of health status is constructed. The state migration path is identified in the feature space, and the risk trigger point is determined by combining the signal transmission characteristics. The adjustment target value is determined by analyzing the coordinated change rules of different modal data within the stabilization action range. By calculating the deviation between the current indicator value and the adjustment target value, a health management plan including the adjustment sequence, adjustment amplitude and adjustment cycle is generated, and the health management plan is dynamically optimized based on real-time feedback during the indicator adjustment process.

2. The method according to claim 1, characterized in that Receive user health data, extract the time series characteristic parameters of each indicator in the user health data, calculate the correlation coefficient between the time series characteristic parameters of different indicators, and write the indicator combination whose correlation coefficient exceeds the preset correlation threshold into the correlation indicator table, including: Receive user health data, calculate the change rate of each indicator in the user health data, determine the fluctuation range of the indicator based on the change rate, set a time series sampling window based on the fluctuation range, automatically shrink the sampling interval of the time series sampling window in the fluctuation-enhanced area, and automatically expand the sampling interval in the fluctuation-weakened area, and obtain a sampled data sequence within the time series sampling window; Extracting time series characteristic parameters from the sampled data sequence, the time series characteristic parameters including fluctuation amplitude ratio, fluctuation duration, and waveform characteristic value, calculating signal entropy values ​​for the time series characteristic parameters to construct a frequency component matrix, generating an eigendecomposition basis based on the frequency component matrix, performing multi-scale decomposition on the time series characteristic parameters using the eigendecomposition basis to obtain a characteristic change sequence, and writing the characteristic change sequence into a time series characteristic table; Based on the time series feature table, the correlation coefficient between the time series feature parameters of different indicators is calculated, the phase of the feature change sequence is calibrated, the sampling deviation is eliminated through dynamic compensation, the frequency domain cross-correlation of the calibrated sequence is calculated, the frequency domain cross-correlation and the time domain correlation coefficient are combined to form the correlation strength, and the indicator combination whose correlation strength exceeds the preset correlation threshold is written into the correlation indicator table.

3. The method according to claim 2, characterized in that Calculating the signal entropy value of the time series characteristic parameters to construct a frequency component matrix, generating an eigendecomposition basis according to the frequency component matrix, and performing multi-scale decomposition on the time series characteristic parameters using the eigendecomposition basis to obtain a characteristic change sequence includes: The time series characteristic parameter sequence is segmented, and the reweighted Shannon entropy value of each segment is calculated. The normalized information entropy sequence is constructed by modifying the sequence using the adaptive weighting coefficient of the signal fluctuation amplitude. The normalized information entropy sequence is arranged according to the time scale to form an entropy distribution matrix. Constructing a frequency component matrix using the entropy distribution matrix, extracting entropy changes at adjacent time scales in the entropy distribution matrix, dividing the distribution intervals of the frequency components based on the entropy changes, calculating the mutual information between the frequency components in different distribution intervals, and writing the frequency component combinations whose mutual information exceeds a preset coupling threshold into a frequency coupling table; Extracting singular values ​​from the frequency component matrix, arranging the singular values ​​in descending order of energy contribution to obtain dominant frequency components, constructing the dominant frequency components as reference basis functions, adjusting the reference basis functions according to the frequency coupling table, minimizing reconstruction errors through iterative optimization, and generating an eigendecomposition basis; The feature decomposition basis is used to perform multi-scale decomposition on the time series feature parameters, the start and end positions of the decomposition interval are determined based on the frequency coupling table, a matching pursuit operation with an adaptive step size is performed in each decomposition interval, and the scale features obtained by decomposition are reorganized in chronological order to generate a feature change sequence.

4. The method according to claim 1, wherein An indicator influence network is constructed based on the correlation indicator table. By introducing a dynamic compensation signal, an interference suppression test is performed on the correlation indicators in the indicator influence network. The attenuation coefficient and distortion degree during the signal transmission process are calculated. The indicator combinations with stable attenuation coefficients and distortion degrees below the preset threshold are written into the causal relationship table. The signal transmission characteristics and stable action ranges are recorded in the causal relationship table, including: An indicator influence network is constructed based on the association indicator table, wherein nodes represent health indicators and edges between nodes represent association strengths. The delay time of signal transmission on the edges is calculated, an adjustment period of a time-varying weight is determined based on the delay time, the time-varying weight is updated based on the signal fluctuation frequency within the adjustment period, and the topology of the indicator influence network is optimized based on the time-varying weight; Selecting related indicators in the optimized indicator influence network, injecting a test signal between the related indicators, detecting a transmission distortion of the test signal within a phase calibration interval, and generating a dynamic compensation signal based on the time-varying weight and the transmission distortion; Superimposing the dynamic compensation signal on the test signal to obtain a compensated test signal, extracting the instantaneous frequency characteristics of the compensated test signal, calculating the frequency offset relative to the original test signal, and obtaining the attenuation coefficient and distortion degree of signal transmission according to the frequency offset; The correlation index pair of the attenuation coefficient fluctuation amplitude less than the first threshold and the distortion degree less than the second threshold is written into the causal relationship table, the stable action interval is determined based on the frequency offset, and the signal transmission characteristics within the stable action interval are recorded in the causal relationship table.

5. The method according to claim 1, wherein Based on multimodal health data, a feature space of health status is constructed. The state migration path is identified in the feature space. The risk trigger point is determined by combining the signal transmission characteristics. The adjustment target value is determined by analyzing the coordinated change rules of different modal data within the stabilization range. The following are included: Dynamically normalize and preprocess the multimodal health data, calculate the information entropy based on the preprocessed data to obtain the mutual information value between modalities, construct a data association matrix based on the mutual information value, use the data association matrix to design an adaptive weight function, dynamically weight the preprocessed data to obtain fusion features, and generate a feature association table; Segmenting the fused features into sliding windows, extracting statistical features and rate-of-change features within each window to obtain a feature sequence, performing density clustering on the feature sequence in combination with a feature association table to obtain state clusters, analyzing the distribution characteristics of the state clusters to determine state transition regions, and constructing a feature space of health status based on the state transition regions; Extracting a state transition trajectory in the feature space, calculating the geometric features of the state transition trajectory to obtain a state transition point, analyzing the system response curve at the state transition point in combination with signal transfer characteristics, determining a risk trigger point based on the convergence of the system response curve, and generating a risk distribution map; The periodic characteristics and trend characteristics of multimodal health data are extracted within the stabilization interval to obtain a change feature sequence, the phase relationship and amplitude relationship of the change feature sequence are analyzed to identify the coordinated change pattern, the safe activity range is determined based on the coordinated change pattern and the risk distribution map, and the state point with the optimal stability within the safe activity range is selected as the adjustment target value.

6. The method according to claim 5, characterized in that Determining the scope of security activities based on the collaborative change pattern and risk distribution map includes: Acquire phase information of multimodal health data, calculate the phase coupling strength between the data, construct a coupling matrix based on the ratio of the maximum to the minimum phase coupling strength, mark the indicator combinations in the coupling matrix whose ratio exceeds a preset coupling threshold as strong coupling pairs, and extract the change trajectory of the strong coupling pairs within the stable working range to construct a collaborative change envelope; Calculating the gradient value of the risk distribution map to obtain a risk gradient, identifying a risk change boundary region based on the risk gradient, and shrinking and adjusting the coordinated change envelope according to the position of the risk change boundary region and the value of the risk gradient to obtain a modified envelope; An optimization objective function is constructed using the modified envelope as a constraint condition, the optimization objective function is solved to obtain a multi-dimensional activity range, and a state transition constraint condition is constructed according to the boundary characteristics of the multi-dimensional activity range; The safe activity area is divided based on the state transition constraint condition, the state stability index is extracted from the safe activity area, and the safe activity area is optimized and adjusted according to the state stability index to obtain the final safe activity range.

7. The method according to claim 1, characterized in that By calculating the deviation between the current indicator value and the adjustment target value, a health management plan including the adjustment sequence, adjustment amplitude and adjustment cycle is generated, including: Obtaining the current indicator value and the adjustment target value, calculating the standardized deviation between the current indicator value and the adjustment target value, constructing an indicator impact matrix based on the standardized deviation, performing eigendecomposition on the indicator impact matrix to obtain eigenvectors, extracting the influence transfer path based on the eigenvectors to determine the adjustment order, and generating a sequence matrix; Acquire historical adjustment data, calculate an indicator response curve in combination with the sequence matrix, determine an adjustment step threshold based on a rate of change of the indicator response curve, adopt an initial adjustment step when the standardized deviation is greater than the adjustment step threshold, and adjust the adjustment step based on real-time feedback data when the standardized deviation is less than the adjustment step threshold, thereby generating an adjustment amplitude table; Collecting real-time indicator data, calculating the rate of change and the trend of change of the real-time indicator data, identifying state change points based on the rate of change and the trend of change, determining the adjustment period according to the time distribution of the state change points, and generating a period distribution table; The adjustment sequence in the sequence matrix is ​​grouped according to the indicator influence matrix, the adjustment step is determined in each group according to the adjustment amplitude table, the adjustment time is set in combination with the cycle distribution table, and a health management plan including the adjustment sequence, adjustment amplitude and adjustment cycle is generated.

8. A health management data processing system based on multimodal data, used to implement the method according to any one of claims 1 to 7, characterized in that: include: The first unit is used to receive user health data, extract the time series characteristic parameters of each indicator in the user health data, calculate the correlation coefficient between the time series characteristic parameters of different indicators, and write the indicator combination whose correlation coefficient exceeds a preset correlation threshold into the correlation indicator table; The second unit is used to construct an indicator influence network based on the correlation indicator table, perform interference suppression tests on the correlation indicators in the indicator influence network by introducing dynamic compensation signals, calculate the attenuation coefficient and distortion degree during signal transmission, write indicator combinations with stable attenuation coefficients and distortion degrees below a preset threshold into the causal relationship table, and record the signal transmission characteristics and stabilization range in the causal relationship table; The third unit is used to perform multi-source fusion of physiological, behavioral, and environmental data based on the causal relationship table to generate multimodal health data. Based on the multimodal health data, a feature space of health status is constructed. The state migration path is identified in the feature space. The risk trigger point is determined by combining the signal transmission characteristics. The adjustment target value is determined by analyzing the coordinated change pattern of different modal data within the stabilization interval. The fourth unit is used to generate a health management plan including the adjustment sequence, adjustment amplitude and adjustment cycle by calculating the deviation between the current indicator value and the adjustment target value, and dynamically optimize the health management plan based on real-time feedback during the indicator adjustment process.

9. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images