A personalized health management scheme data mining method

Abstract

This invention relates to the field of health management technology and discloses a data mining method for personalized health management solutions. The method includes: acquiring multi-source heterogeneous health record data of target users; performing spatiotemporal alignment processing on the multi-source heterogeneous health record data to generate a standard health data sequence; constructing a dynamic health entropy value to divide health level intervals; inputting the health level intervals into a preset health intervention knowledge base; determining intervention measure primitives that match the health feature vector set; associating and mapping the intervention measure primitives with the dynamic health entropy value to construct an intervention mapping matrix; optimizing the intervention measure primitives in the intervention mapping matrix by minimizing the dynamic health entropy value to generate a personalized health candidate set; and performing expected effect deduction on the personalized health candidate set to obtain an optimized health management solution. This invention can improve the efficiency of data mining for personalized health management solutions.

Description

Technical Field

[0001] This invention relates to the field of health management technology, and in particular to a data mining method for personalized health management programs. Background Technology

[0002] Existing technologies have significant shortcomings in health data processing. They cannot perform unified spatiotemporal alignment and normalization of multi-source heterogeneous health record data; the methods for correcting missing data values ​​are simplistic and fail to form standardized health data sequences; they cannot accurately extract the correlation features between physiological indicator fluctuations and behavioral patterns, and cannot construct a complete set of health feature vectors that reflect the user's true state, resulting in a lack of reliable data foundation for subsequent analysis.

[0003] Existing technologies have significant shortcomings in the health assessment and solution generation stages. They fail to quantify and classify health status using dynamic health entropy values, resulting in a crude and unscientific approach to health level classification. They cannot accurately match intervention measures by incorporating a health intervention knowledge base, nor can they construct an intervention mapping matrix to achieve quantitative correlation. Furthermore, they cannot select and prioritize intervention combinations based on optimization goals, nor can they extrapolate expected effects. Consequently, the resulting health management solutions suffer from poor adaptability and insufficient scientific rigor, failing to meet the needs of personalized health management.

[0004] While there have been attempts to use Mahalanobis distance for medical data analysis, no technical solution has been disclosed that uses it to nonlinearly aggregate the deviation and dispersion rate of health feature vectors to construct dynamic health entropy values. Summary of the Invention

[0005] This invention provides a data mining method for personalized health management solutions to address the problems mentioned in the background section.

[0006] To achieve the above objectives, the present invention provides a data mining method for personalized health management solutions, comprising:

[0007] A1: Obtain multi-source heterogeneous health record data of the target user, perform spatiotemporal alignment processing on the multi-source heterogeneous health record data, and generate a standard health data sequence of the target user.

[0008] A2: Extract the physiological indicator fluctuation characteristics and behavioral pattern correlation characteristics from the standard health data sequence to construct the health feature vector set of the target user;

[0009] A3: Nonlinearly aggregate the deviation and dispersion rate of the health feature vector set to obtain the dynamic health entropy value of the target user, so as to divide the health level range of the target user;

[0010] A4: Input the health level range into a preset health intervention knowledge base, determine the intervention measure primitives that match the health feature vector set, associate and map the intervention measure primitives with the dynamic health entropy value, and construct the intervention mapping matrix for the target user;

[0011] A5: With minimizing the dynamic health entropy as the optimization objective, the intervention measure primitives in the intervention mapping matrix are combined, screened, and prioritized to generate a personalized health candidate set for the target user;

[0012] A6: Perform expected effect deduction on the personalized health candidate set to obtain the optimized health management plan for the target user.

[0013] In a preferred embodiment, the step of acquiring multi-source heterogeneous health record data of the target user, performing spatiotemporal alignment processing on the multi-source heterogeneous health record data, and generating a standard health data sequence for the target user includes:

[0014] Simultaneously collect wearable device monitoring data, electronic medical record data, and physical examination report text data of the target user, and integrate them into multi-source heterogeneous health record data of the target user;

[0015] Using the timestamp of the wearable device monitoring data as a reference, linear interpolation is performed to align the timestamps in the electronic medical record data and the physical examination report text data to obtain the target user's synchronized health data;

[0016] Based on the synchronized health data, the electronic medical record data and the physical examination report text data are normalized to obtain the normalized health data of the target user.

[0017] The normalized health data is classified, and the missing values ​​in the classified data are corrected by difference based on the historical health data mean of the target user, so as to obtain the standard health data sequence of the target user.

[0018] In a preferred embodiment, the step of extracting the physiological indicator fluctuation characteristics and behavioral pattern correlation characteristics from the standard health data sequence to construct the health feature vector set of the target user includes:

[0019] Data decoupling is performed on the standard health data sequence to obtain the behavioral parameter subsequence and physiological parameter subsequence of the target user;

[0020] Fluctuation analysis is performed on the physiological indicators in the physiological parameter subsequence to obtain the fluctuation characteristics of the target user's physiological indicators;

[0021] A time correlation analysis was performed on the step count and sleep duration indicators in the behavioral parameter subsequence to obtain the behavioral pattern association features of the target user.

[0022] The physiological indicator fluctuation characteristics and the behavioral pattern association characteristics are vectorized and concatenated to generate the health feature vector set of the target user.

[0023] In a preferred embodiment, the step of nonlinearly aggregating the deviation and dispersion rate of the health feature vector set to obtain the dynamic health entropy value of the target user, in order to divide the health level range of the target user, includes:

[0024] Extract the center point of each dimension of the feature vector in the health feature vector set, and calculate the Mahalanobis distance from each feature vector in the health feature vector set to the center point;

[0025] The deviation of the Mahalanobis distance is measured to obtain the degree of deviation of the health feature vector set;

[0026] Extract the standard deviation and mean of each dimension of the feature vector in the health feature vector set, and use the ratio of the standard deviation to the mean as the dispersion rate of the health feature vector set.

[0027] In a preferred embodiment, the step of using the ratio of the standard deviation to the mean as the dispersion rate of the health feature vector set includes:

[0028] The deviation degree and the dispersion rate are nonlinearly fused to generate the dynamic health entropy value of the target user, wherein the calculation formula of the dynamic health entropy value is:

[0029] ;

[0030] In the formula, The dynamic health entropy value, The weighting coefficient for the degree of deviation. The degree of deviation, It is an exponential function. The weighting coefficients for the dispersion rate are, This is the preset deviation saturation adjustment coefficient. The dispersion rate is... It is a logarithmic function;

[0031] Obtain a preset set of health level thresholds, compare the dynamic health entropy value with the preset set of health level thresholds, and determine the threshold range of the dynamic health entropy value.

[0032] Based on the threshold range, the health level labels of the preset health level threshold set are matched to divide the health level range of the target user.

[0033] In a preferred embodiment, the step of inputting the health level range into a preset health intervention knowledge base and determining the intervention measure primitives matching the health feature vector set includes:

[0034] Extract the level labels corresponding to the health level range;

[0035] Input the level label into the preset health intervention knowledge base and locate the intervention strategy index table associated with the level label;

[0036] Using the set of health feature vectors as query vectors, similarity matching is performed in the intervention strategy index table to obtain a set of candidate intervention measures for the target user;

[0037] Extract the execution conditions and objectives of the candidate intervention measures, and eliminate intervention measures that are incompatible with the health feature vector set to obtain the intervention measure primitives for the target user.

[0038] In a preferred embodiment, the step of associating the intervention measure primitives with the dynamic health entropy value to construct the intervention mapping matrix for the target user includes:

[0039] Obtain the effect dimension labels and expected adjustment magnitudes from the intervention measure primitives;

[0040] The dynamic health entropy value is decomposed into the action dimension label to obtain the entropy component of the dynamic health entropy value;

[0041] The expected adjustment range is multiplied by the entropy component to obtain the adjustment intensity value of the intervention measure primitive;

[0042] Using the intervention measure primitive as the row index and the effect dimension label as the column index, the adjustment intensity value is filled into the corresponding position in the matrix to construct the intervention mapping matrix for the target user.

[0043] In a preferred embodiment, the step of combining, screening, and prioritizing intervention measure primitives in the intervention mapping matrix to generate a personalized health candidate set for the target user, with the goal of minimizing the dynamic health entropy value, includes:

[0044] Based on the adjustment intensity value, identify the synergistic and repulsive relationships between the intervention measure primitives in the intervention mapping matrix;

[0045] By performing topological reconstruction on the cooperative and repulsive relationships, an interaction graph of the intervention measures for the target user is obtained;

[0046] With minimizing the dynamic health entropy as the optimization objective, the interaction graph of the intervention measures is traversed to generate a candidate combination set for the target user.

[0047] In a preferred embodiment, the step of generating a candidate combination set for the target user by traversing the interaction graph of the intervention measures with the optimization objective of minimizing the dynamic health entropy value includes:

[0048] Based on the predicted decrease in the dynamic health entropy value, the candidate combination set is sorted in descending order to obtain the first sorting sequence of the target user.

[0049] Obtain the implementation cost label and implementation risk level from the candidate combination set;

[0050] Based on the implementation cost label and the implementation risk level, the candidate combinations in the first sorting sequence are sorted a second time to obtain the personalized health candidate set of the target user.

[0051] In a preferred embodiment, the step of extrapolating the expected effects of the personalized health candidate set to obtain an optimized health management plan for the target user includes:

[0052] Obtain the implementation sequence relationship from the personalized health candidate set;

[0053] The implementation time sequence relationship is temporally correlated with the target user's historical health data, and the correlation result is used to perform path evolution to generate the target user's expected health evolution path;

[0054] The expected health evolution path is calibrated with indicators to obtain the effect evaluation indicators for the target users.

[0055] Based on the aforementioned effectiveness evaluation indicators, the personalized health candidate set is comprehensively evaluated to obtain an optimized health management plan for the personalized health candidate set.

[0056] Compared with the prior art, the present invention has the following beneficial effects:

[0057] 1. This invention provides precise data support for personalized health management through multi-source data fusion and feature extraction. It performs spatiotemporal alignment, normalization, and missing value correction on multi-source heterogeneous health record data to generate a standard health data sequence; it extracts the fluctuation characteristics of physiological indicators and the correlation features of behavioral patterns to construct a health feature vector set, comprehensively depicting the user's health status and laying a reliable foundation for subsequent analysis.

[0058] 2. This invention significantly improves the efficiency and adaptability of health management program generation through dynamic quantification and intelligent deduction. It calculates dynamic health entropy values ​​to divide health level intervals, matches intervention measure primitives with an intervention knowledge base, and constructs an intervention mapping matrix. It selects combination solutions with the goal of minimizing health entropy values, and generates optimized health management programs through expected effect deduction, achieving personalized, scientific, and efficient program output. Attached Figure Description

[0059] Figure 1 This is a flowchart illustrating a data mining method for a personalized health management plan, as provided in an embodiment of the present invention.

[0060] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0061] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0062] This application provides a data mining method for personalized health management solutions. The executing entity of this personalized health management solution data mining method includes, but is not limited to, at least one of the following electronic devices that can be configured to execute the method provided in this application embodiment: a server, a terminal, etc. In other words, the personalized health management solution data mining method can be executed by software or hardware installed on a terminal device or a server device. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster. The server can be an independent server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.

[0063] Reference Figure 1 The diagram shown is a flowchart illustrating a data mining method for personalized health management solutions according to an embodiment of the present invention. In this embodiment, the data mining method for personalized health management solutions includes:

[0064] A1: Obtain multi-source heterogeneous health record data of the target user, perform spatiotemporal alignment processing on the multi-source heterogeneous health record data, and generate a standard health data sequence of the target user.

[0065] In this embodiment of the invention, the step of acquiring multi-source heterogeneous health record data of a target user, performing spatiotemporal alignment processing on the multi-source heterogeneous health record data, and generating a standard health data sequence for the target user includes:

[0066] Simultaneously collect wearable device monitoring data, electronic medical record data, and physical examination report text data of the target user, and integrate them into multi-source heterogeneous health record data of the target user;

[0067] Using the timestamp of the wearable device monitoring data as a reference, linear interpolation is performed to align the timestamps in the electronic medical record data and the physical examination report text data to obtain the target user's synchronized health data;

[0068] Based on the synchronized health data, the electronic medical record data and the physical examination report text data are normalized to obtain the normalized health data of the target user.

[0069] The normalized health data is classified, and the missing values ​​in the classified data are corrected by difference based on the historical health data mean of the target user, so as to obtain the standard health data sequence of the target user.

[0070] Through the health data collection channel authorized by the target user, real-time physiological monitoring data is collected from wearable device terminals, electronic medical record data is collected from medical institution information systems, and physical examination report text data is collected from physical examination institution databases. The three types of data are collected in a unified manner to form multi-source heterogeneous health record data of the target user, which includes different sources, structures and formats, providing a complete original data foundation for subsequent spatiotemporal alignment processing.

[0071] Using the continuous timestamps inherent in the wearable device monitoring data as a unified time reference, the discrete time tags inherent in the electronic medical record data and the physical examination report text data are matched and located. The health data with missing time points are linearly interpolated and aligned according to the chronological order, so that the time nodes of the electronic medical record data and the physical examination report text data are completely matched with the timestamps of the wearable device monitoring data. After processing, the target user's synchronized health data with completely unified time dimension is obtained.

[0072] Based on synchronized health data with a unified time dimension, and in accordance with the standardized processing specifications for medical and health data, the numerical and textual indicators of electronic medical record data and physical examination report text data are processed to unify the format and dimensions. Data from different sources and with different expressions are converted into the same standard expression, eliminating the differences in format and dimensions between data. After processing, the normalized health data of the target user is obtained.

[0073] Based on health data types such as physiological indicators, disease diagnoses, examination results, and lifestyle monitoring, normalized health data is clearly categorized, and data of the same type are grouped into the same category. The historical average of the target user's health data of the same type is retrieved as the correction basis. This historical average comes from the statistical results of the target user's long-term accumulated health data. Missing values ​​in each category of data after classification are corrected by difference, and missing health data content is supplemented to ensure data rationality. After correction, a standard health data sequence of the target user with uniform time, standard format, and complete data is obtained.

[0074] The beneficial effects are that it integrates multi-source heterogeneous health data, fills in the gaps in data sources and dimensions, and provides complete original support for health analysis.

[0075] A unified timestamp is used to achieve spatiotemporal alignment, eliminate data time misalignment, and ensure time sequence consistency; normalization processing unifies the data format and units, solving the problem of incomparability of heterogeneous data.

[0076] By correcting missing values ​​using historical averages, we can improve data integrity and accuracy, and form a standardized health data sequence.

[0077] A2: Extract the physiological indicator fluctuation characteristics and behavioral pattern correlation characteristics from the standard health data sequence to construct the health feature vector set of the target user;

[0078] In this embodiment of the invention, the step of extracting the physiological indicator fluctuation characteristics and behavioral pattern correlation characteristics from the standard health data sequence to construct the health feature vector set of the target user includes:

[0079] Data decoupling is performed on the standard health data sequence to obtain the behavioral parameter subsequence and physiological parameter subsequence of the target user;

[0080] Fluctuation analysis is performed on the physiological indicators in the physiological parameter subsequence to obtain the fluctuation characteristics of the target user's physiological indicators;

[0081] A time correlation analysis was performed on the step count and sleep duration indicators in the behavioral parameter subsequence to obtain the behavioral pattern association features of the target user.

[0082] The physiological indicator fluctuation characteristics and the behavioral pattern association characteristics are vectorized and concatenated to generate the health feature vector set of the target user.

[0083] The standard health data sequence is decomposed according to preset data dimension division rules. Data describing daily activity status is divided into a separate behavioral parameter subsequence, and data describing bodily function status is divided into a separate physiological parameter subsequence. The standard health data sequence originates from complete health data content that has undergone time-standardized formatting and missing value correction in the early stages. The behavioral parameter subsequence includes data related to steps, sleep duration, and activity periods, while the physiological parameter subsequence includes data related to heart rate, blood pressure, blood oxygen, and sleep physiological indicators.

[0084] Each physiological indicator in the physiological parameter subsequence is compared and judged segment by segment in continuous time sequence. The values ​​of physiological indicators at adjacent time points are directly compared, and the specific changes in the values, such as rise, fall and stabilization, are recorded. The magnitude and frequency of change of each physiological indicator in different time periods are sorted out, and the dynamic changes of each physiological indicator over time are fully presented, forming the fluctuation characteristics of physiological indicators that can intuitively reflect the stability of bodily functions. All physiological parameter subsequences are derived from the daily health monitoring and regular physical examination data of the target users.

[0085] The step count and sleep duration indicators in the behavioral parameter subsequences were synchronously compared according to completely corresponding time nodes. The step count data and sleep duration data within the same statistical period were directly correlated and compared to determine the correspondence between the number of steps and the length of sleep. The direct impact of changes in step count at different times on sleep duration and the effect of sleep duration on subsequent step counts were analyzed. The correlation and influence patterns between the two behavioral indicators were clarified, forming behavioral pattern correlation characteristics that can reflect the daily activity patterns. All behavioral parameter subsequences were derived from the target users' daily wearable device monitoring and health behavior record data.

[0086] All data from the physiological indicator fluctuation characteristics and the behavioral pattern correlation characteristics are directly combined and organized in a unified order, maintaining the original structure and correspondence of the two types of characteristics. The combination forms a complete and standardized health characteristic data set, which can comprehensively reflect the physiological dynamic changes and behavioral pattern correlation of the target user. Finally, a health characteristic vector set of the target user is generated. Both the physiological indicator fluctuation characteristics and the behavioral pattern correlation characteristics are derived from the target user's personal health-related data content that has been standardized and normalized.

[0087] The beneficial effect is that by decoupling the data, the standard health data sequence is split into behavioral and physiological parameter subsequences, which clearly divides the data dimensions and provides a well-structured and clearly categorized data foundation for accurately extracting health features.

[0088] By analyzing the fluctuations of physiological indicators and the correlation between behavioral indicators, we can comprehensively capture the dynamic changes in physiology and the patterns of behavioral relationships, fully reflecting the user's true health status and potential health risks.

[0089] By vectorizing and concatenating physiological fluctuation characteristics and behavioral correlation characteristics, a standardized and unified health feature vector set is generated, providing high-quality and highly usable feature support for subsequent health assessments and classification.

[0090] A3: Nonlinearly aggregate the deviation and dispersion rate of the health feature vector set to obtain the dynamic health entropy value of the target user, so as to divide the health level range of the target user;

[0091] In this embodiment of the invention, the step of nonlinearly aggregating the deviation and dispersion rate of the health feature vector set to obtain the dynamic health entropy value of the target user, in order to divide the health level range of the target user, includes:

[0092] Extract the center point of each dimension of the feature vector in the health feature vector set, and calculate the Mahalanobis distance from each feature vector in the health feature vector set to the center point;

[0093] The deviation of the Mahalanobis distance is measured to obtain the degree of deviation of the health feature vector set;

[0094] Extract the standard deviation and mean of each dimension of the feature vector in the health feature vector set, and use the ratio of the standard deviation to the mean as the dispersion rate of the health feature vector set.

[0095] The step of using the ratio of the standard deviation to the mean as the dispersion rate of the health feature vector set includes:

[0096] The deviation degree and the dispersion rate are nonlinearly fused to generate the dynamic health entropy value of the target user, wherein the calculation formula of the dynamic health entropy value is:

[0097] ;

[0098] In the formula, The dynamic health entropy value, The weighting coefficient for the degree of deviation. The degree of deviation, It is an exponential function. The weighting coefficients for the dispersion rate are, This is the preset deviation saturation adjustment coefficient. The dispersion rate is... It is a logarithmic function;

[0099] Obtain a preset set of health level thresholds, compare the dynamic health entropy value with the preset set of health level thresholds, and determine the threshold range of the dynamic health entropy value.

[0100] Based on the threshold range, the health level labels of the preset health level threshold set are matched to divide the health level range of the target user.

[0101] The geometric center point of all feature vectors in the health feature vector set is identified and extracted. This center point is the average representative point of the health feature vector set in the feature space. The data source is all dimensional feature data stored within the health feature vector set. By calculating the distance between each feature vector in the health feature vector set and this center point in the multidimensional feature space, Mahalanobis distance is used as the distance metric. Mahalanobis distance can eliminate the correlation and dimensional differences between the dimensions of the feature vectors, accurately reflecting the deviation position and degree of each feature vector from the overall center. After the calculation is completed, the set of Mahalanobis distances from each feature vector in the health feature vector set to the center point is obtained, providing basic data for subsequent measurement of the degree of deviation.

[0102] Based on the calculated Mahalanobis distance from each feature vector to the center point in the health feature vector set, all Mahalanobis distance values ​​are summarized and statistically analyzed. The degree of deviation is used as a quantitative indicator to measure the degree of dispersion and tightness of each feature vector in the health feature vector set relative to the center point. By statistically analyzing the distribution, deviation magnitude, and deviation frequency of each Mahalanobis distance, the overall deviation status of the health feature vector set is comprehensively determined. The greater the degree of deviation, the more dispersed the distribution of the health feature vector set and the more obvious the difference between each feature vector and the center point. Conversely, the smaller the degree of deviation, the more clustered the health feature vector set is. The analysis results directly reflect the overall deviation level of the target user's health feature vector set.

[0103] The standard deviation of each feature vector in the health feature vector set is extracted. This standard deviation is derived from the statistical analysis of the dispersion of all feature vector values ​​in the same dimension of the health feature vector set. At the same time, the mean of each feature vector is extracted. This mean is derived from the statistical analysis of the average level of all feature vector values ​​in the same dimension of the health feature vector set. The ratio between the standard deviation and the mean is calculated and used as the dispersion rate of the health feature vector set. The dispersion rate can standardize the reflection of the fluctuation and distribution of the data of each feature vector, eliminate the influence of the difference in the range of feature values ​​in different dimensions, realize a unified quantitative description of the dispersion of the data in each dimension of the health feature vector set, and provide a reliable dispersion rate index for subsequent nonlinear aggregation.

[0104] The determined deviation and the calculated dispersion rate are superimposed and integrated layer by layer according to feature weights. The deviation comes from the quantitative result of the overall distribution of the health feature vector set, and the dispersion rate comes from the quantitative result of the fluctuation of data in each dimension of the health feature vector set. The two indicators are nonlinearly fused by superimposing and integrating layer by layer. During the fusion process, the numerical characteristics and change patterns of the deviation and dispersion rate are fully preserved. Finally, a dynamic health entropy value that can comprehensively reflect the dynamic changes of the target user's health status is generated. The dynamic health entropy value can intuitively reflect the stability and fluctuation risk of the target user's health characteristics.

[0105] The system retrieves a pre-set set of health level thresholds from the standard configuration library of the health management system. This set of thresholds was developed by professional medical personnel in conjunction with clinical health assessment standards. The data source is authoritative medical and health assessment systems and standard data accumulated from long-term health management practices. It contains a complete range of thresholds corresponding to different health states and can be directly used for the determination and classification of health status levels.

[0106] The generated dynamic health entropy value is compared with the preset health level threshold set obtained from the standard configuration library item by item. During the comparison, the value range of the dynamic health entropy value in the threshold set is matched one by one to accurately determine the specific threshold interval corresponding to the dynamic health entropy value. This interval directly corresponds to the current health status level of the target user.

[0107] Based on the determined threshold range, the corresponding health level label is searched and matched in the preset health level threshold set. The health level label is uniformly formulated by the professional medical assessment system and contains clear division standards for different health levels. Based on the matched health level label, the final division of the target user's health level range is completed. The division result can be directly used for the target user's health status assessment and subsequent health management services. The health level threshold set is determined by clinical experts based on the distribution of health data of a large sample population. Specifically, health level H1 corresponds to the entropy value range [0, 0.3), H2 corresponds to [0.3, 0.6), etc.

[0108] The weighting coefficient of the deviation degree comes from the overall contribution evaluation result of the target user's health feature vector set. This evaluation result is obtained by traversing and statistically analyzing the influence of each dimension of the health feature vector set one by one. The statistical process takes the actual correlation of health status as the sole criterion and requires that the sum of the deviation degree weighting coefficient and the dispersion rate weighting coefficient be one.

[0109] The degree of deviation is derived from the quantitative result of the difference between the health feature vector set and the center position of the standard health feature. This quantitative result is obtained by comparing the distribution difference between each feature vector in the health feature vector set and the center position item by item. The comparison process takes the actual distribution state of the health feature as the sole criterion.

[0110] The preset deviation saturation adjustment coefficient is derived from the statistical results of the saturation critical value of the deviation change in health data. This saturation critical value is obtained by fitting the deviation change trend of a large number of healthy people piece by piece. The fitting process takes the actual stable boundary of the health status as the sole criterion.

[0111] The weighting coefficient of the dispersion rate comes from the contribution evaluation results of the discrete states of each dimension of the target user's health feature vector set. This evaluation result is obtained by traversing and statistically analyzing the degree of fluctuation impact of each dimension of the health feature vector set data one by one. The statistical process takes the actual fluctuation sensitivity of the health status as the sole criterion.

[0112] The dispersion rate is derived from the quantitative result of the distribution dispersion of data in each dimension of the health feature vector set. This dispersion is obtained by statistically analyzing the distribution of all data in the same dimension of the health feature vector set item by item. The statistical process takes the actual dispersion of health features as the sole criterion.

[0113] The exponential function is used to perform saturation adjustment calculations on the degree of deviation. During the calculation, the degree of deviation is first combined with the preset deviation degree saturation adjustment coefficient, and then a continuous numerical transformation is completed based on the natural constant. The transformation process can limit the excessive growth of the degree of deviation, making the result more consistent with the actual changes in health status.

[0114] The logarithmic function is used to perform a smoothing transformation on the dispersion rate. During the operation, the dispersion rate is first added to a fixed value to eliminate the influence of zero values, and then the continuous value is transformed by the natural logarithm rule. The transformation process can smooth the sharp fluctuations of the dispersion rate, making the result more in line with the expression requirements of the stability of the health state.

[0115] The dynamic health entropy value is obtained by weighting and fusing the result of the deviation degree after saturation adjustment and the result of the dispersion rate after stationary transformation. The weighting and fusion process uses the weight coefficients of the deviation degree and the dispersion rate to balance and adjust, and finally forms a unified quantitative value of health status.

[0116] The overall significance of dynamic health entropy lies in comprehensively quantifying the stability and fluctuation risk of the target user's health status. The higher the value, the more dispersed the distribution of health characteristics, the more violent the fluctuations, and the lower the stability. The lower the value, the more concentrated the distribution of health characteristics, the smoother the fluctuations, and the higher the stability.

[0117] Dynamic health entropy can be directly used as the core basis for health level classification, intervention screening and health management program optimization, providing target users with an objective and unified health status assessment standard.

[0118] The beneficial effect is that by using Mahalanobis distance to calculate the degree of deviation between the feature vector and the center point, the dimensional correlation and dimensional difference can be eliminated, the overall deviation of health characteristics from the normal level can be accurately quantified, and the objectivity of the assessment can be improved.

[0119] Using the ratio of standard deviation to mean as the dispersion rate can uniformly measure the degree of fluctuation of health data in various dimensions, eliminate the influence of differences in numerical range, and reliably reflect the stability and dispersion of health characteristics.

[0120] The degree of deviation and the dispersion rate provide core quantitative indicators for the calculation of dynamic health entropy value, laying a stable and accurate data foundation for subsequent nonlinear fusion and scientific division of health level intervals.

[0121] By using exponential and logarithmic functions to nonlinearly fuse the degree of deviation and dispersion rate, numerical fluctuations can be smoothed and abnormal increases can be suppressed, making the dynamic health entropy value more consistent with the actual changes in health status.

[0122] Dynamic health entropy can comprehensively quantify the stability and fluctuation risk of a user's health. The numerical value intuitively reflects the dispersion and fluctuation of health characteristics, providing a unified and objective quantitative basis for health assessment.

[0123] By matching dynamic health entropy values ​​within a range based on preset level thresholds, health levels can be scientifically classified, making health status grading more accurate, authoritative, and clinically valuable.

[0124] A4: Input the health level range into a preset health intervention knowledge base, determine the intervention measure primitives that match the health feature vector set, associate and map the intervention measure primitives with the dynamic health entropy value, and construct the intervention mapping matrix for the target user;

[0125] In this embodiment of the invention, the step of inputting the health level range into a preset health intervention knowledge base and determining the intervention measure primitives that match the health feature vector set includes:

[0126] Extract the level labels corresponding to the health level range;

[0127] Input the level label into the preset health intervention knowledge base and locate the intervention strategy index table associated with the level label;

[0128] Using the set of health feature vectors as query vectors, similarity matching is performed in the intervention strategy index table to obtain a set of candidate intervention measures for the target user;

[0129] Extract the execution conditions and objectives of the candidate intervention measures, and eliminate intervention measures that are incompatible with the health feature vector set to obtain the intervention measure primitives for the target user.

[0130] The step of associating and mapping the intervention measure primitives with the dynamic health entropy value to construct the intervention mapping matrix for the target user includes:

[0131] Obtain the effect dimension labels and expected adjustment magnitudes from the intervention measure primitives;

[0132] The dynamic health entropy value is decomposed into the action dimension label to obtain the entropy component of the dynamic health entropy value;

[0133] The expected adjustment range is multiplied by the entropy component to obtain the adjustment intensity value of the intervention measure primitive;

[0134] Using the intervention measure primitive as the row index and the effect dimension label as the column index, the adjustment intensity value is filled into the corresponding position in the matrix to construct the intervention mapping matrix for the target user.

[0135] The level labels corresponding to the health level ranges are directly extracted from the target user's health status assessment results. These assessment results are derived from the complete process of monitoring the target user's full-dimensional health characteristics and classifying health levels. The level labels are used to uniquely identify the target user's current health level category.

[0136] The pre-set health intervention knowledge base is used to store intervention strategy information corresponding to various health levels. After the extracted level labels are directly input into the knowledge base, the knowledge base will automatically locate the intervention strategy index table uniquely associated with the level label based on the association between the level labels and intervention strategies stored internally. This index table completely records the basic information of all available intervention strategies under the corresponding health level.

[0137] The health feature vector set fully contains all the health feature information of the target user. After the vector set is directly input into the intervention strategy index table as a query vector, each intervention strategy in the table will be compared with the feature content of the health feature vector set. During the comparison process, the feature category and feature attribute are fully matched, and finally a set of candidate intervention measures that match the health features of the target user are obtained.

[0138] The execution conditions and objectives of each intervention in the candidate intervention set are directly extracted from the intervention strategy index table. After extraction, the execution conditions, objectives and health feature vector set of each intervention are checked for compatibility. Interventions whose execution conditions cannot be met or whose objectives do not match are eliminated. The remaining interventions directly constitute the intervention primitives for the target user.

[0139] The function dimension labels are directly extracted from the functional classification information of the intervention measure primitives. This functional classification information fully records the specific areas in which the intervention measures play a role in the user's health status. The source is the functional positioning and classification work carried out in the early stage of intervention measure primitives.

[0140] The expected adjustment range is directly extracted from the effect evaluation information of the intervention element. This effect evaluation information fully records the range of health status adjustment that can be achieved after the implementation of the intervention, and is derived from the previous work on the actual effect verification and quantitative analysis of the intervention element.

[0141] The dynamic health entropy value is split according to the specific functional domain marked by the function dimension label. The splitting process is divided according to the health impact range corresponding to each function dimension. After the division, the dynamic health entropy value component corresponding to each function dimension is obtained.

[0142] The values ​​of the dynamic health entropy components are derived from the comprehensive monitoring and quantitative assessment of the user's health status in the early stages, with each component precisely corresponding to the quantitative results of health status in a specific health function area.

[0143] The expected adjustment magnitude is directly multiplied by the entropy component. The original quantification standards of the two values ​​remain unchanged during the calculation process. The resulting value is directly used as the adjustment intensity value of the intervention measure primitive.

[0144] The value of the adjustment intensity is derived from the direct combination of the range of health adjustments that the intervention can achieve and the quantitative results of the current health status, accurately reflecting the actual adjustment ability of the intervention on the user's health status.

[0145] Intervention measure primitives are directly used as row indices in the matrix. Each row index uniquely corresponds to an independent intervention measure. The row indices are derived from the independent identification and sorting work carried out on the intervention measure primitives in the early stage.

[0146] The function dimension labels are directly used as column indices of the matrix. Each column index uniquely corresponds to an independent health function function area. The column indexes are derived from the previous independent identification and classification work carried out on the health function areas.

[0147] The adjustment intensity value is precisely filled into the corresponding position in the matrix according to the corresponding row index and column index. The original quantification standard of the value remains unchanged during the filling process. After the filling is completed, the intervention mapping matrix of the target user is fully constructed.

[0148] All data in the intervention mapping matrix comes from the comprehensive monitoring of users' health status in the early stage, the functional positioning and effect evaluation of intervention measures, and the matrix fully presents the actual regulatory capacity of each intervention measure for different health function areas.

[0149] The beneficial effects are that by extracting health level tags and locating the intervention strategy index table, the search scope for intervention measures can be quickly narrowed, invalid matches can be reduced, and the initial matching efficiency between intervention measures and user health levels can be improved.

[0150] By using health feature vector sets for similarity matching, candidate intervention measures that match the user's health characteristics can be accurately screened, thereby improving the targeting and personalization of intervention measures.

[0151] Verifying the execution conditions and objectives and eliminating incompatible measures can ensure the effectiveness and feasibility of intervention primitives, avoid ineffective interventions, and provide a reliable foundation for subsequent intervention mapping.

[0152] Obtaining the dimensions of action and expected adjustment range of intervention measures can clarify the direction of action and improvement potential of each measure, providing a clear parameter basis for subsequent quantitative correlation and intensity calculation.

[0153] Decomposing dynamic health entropy values ​​into entropy components according to their function dimensions can accurately pinpoint the risk level of each health dimension, making intervention and regulation more targeted and improving the effectiveness of health improvement.

[0154] By calculating the adjustment intensity by multiplying the expected adjustment magnitude by the entropy component, the intervention effect can be objectively quantified, avoiding subjective judgment bias and providing a reliable numerical basis for the selection of intervention combinations.

[0155] By constructing a mapping matrix based on the basic elements and dimensions of intervention, the correspondence between intervention and health dimensions can be systematically presented, facilitating rapid retrieval and combination optimization, and improving the efficiency of solution generation.

[0156] A5: With minimizing the dynamic health entropy as the optimization objective, the intervention measure primitives in the intervention mapping matrix are combined, screened, and prioritized to generate a personalized health candidate set for the target user;

[0157] In this embodiment of the invention, the step of combining, screening, and prioritizing the intervention measure primitives in the intervention mapping matrix to generate a personalized health candidate set for the target user, with the goal of minimizing the dynamic health entropy value, includes:

[0158] Based on the adjustment intensity value, identify the synergistic and repulsive relationships between the intervention measure primitives in the intervention mapping matrix;

[0159] By performing topological reconstruction on the cooperative and repulsive relationships, an interaction graph of the intervention measures for the target user is obtained;

[0160] With minimizing the dynamic health entropy as the optimization objective, the interaction graph of the intervention measures is traversed to generate a candidate combination set for the target user.

[0161] The step of minimizing the dynamic health entropy value as the optimization objective, traversing the interaction graph of the intervention measures, and generating a candidate combination set for the target user, includes:

[0162] Based on the predicted decrease in the dynamic health entropy value, the candidate combination set is sorted in descending order to obtain the first sorting sequence of the target user.

[0163] Obtain the implementation cost label and implementation risk level from the candidate combination set;

[0164] Based on the implementation cost label and the implementation risk level, the candidate combinations in the first sorting sequence are sorted a second time to obtain the personalized health candidate set of the target user.

[0165] Based on the regulation intensity values ​​corresponding to each intervention primitive in the intervention mapping matrix, the combined effect of any two intervention primitives is judged. Two intervention primitives whose regulation intensity values ​​are superimposed and improve the health regulation effect are judged to have a synergistic relationship. Two intervention primitives whose regulation intensity values ​​cancel each other out and reduce the health regulation effect are judged to have a repulsive relationship. The regulation intensity values ​​are all derived from the filling values ​​of the intervention mapping matrix, which are calculated from the expected regulation amplitude and the entropy component.

[0166] All identified synergistic and exclusionary relationships are reorganized and arranged according to the function dimension of the intervention measure primitives. Intervention measure primitives with synergistic relationships are linked and integrated, while intervention measure primitives with exclusionary relationships are separated independently. After the topological structure of all relationships is reconstructed, an intervention measure interaction diagram is formed that intuitively shows the mutual influence relationship of each intervention measure primitive. This interaction diagram fully retains the original attributes and relationship characteristics of all intervention measure primitives.

[0167] Minimizing the dynamic health entropy value of the target user is the sole optimization direction. The dynamic health entropy value is the core quantitative result reflecting the stability of the user's health. Starting from the starting node of the interaction graph of intervention measures, all intervention measure primitives are traversed sequentially. Intervention measure primitives with synergistic relationships and that can significantly reduce the dynamic health entropy value are selected for combination. The principle of synergy priority and exclusion avoidance is followed throughout the traversal. After the traversal is completed, all combinations that meet the optimization objective are sorted and summarized to generate a candidate combination set for the target user.

[0168] The dynamic health entropy value change after the implementation of each intervention combination in the candidate combination set is predicted, and the predicted decrease in dynamic health entropy value corresponding to each combination is calculated. The predicted decrease is derived from the comprehensive deduction of the interaction between the adjustment intensity value and the intervention measures in the intervention mapping matrix. All combinations in the candidate combination set are arranged in descending order of predicted decrease, and the first sorted sequence of the target users is obtained after the arrangement is completed.

[0169] The implementation cost label and implementation risk level of each intervention combination are retrieved from the candidate combination set in the pre-set health intervention knowledge base. The implementation cost label is used to identify the level of resource input required to implement the intervention combination, and the implementation risk level is used to identify the level of health impact that may occur during the implementation of the intervention combination. Both types of information are derived from the results of the compilation of professional medical and health assessment system and actual intervention implementation experience.

[0170] Based on the level of resource investment represented by the implementation cost label and the level of security represented by the implementation risk level, the order of all candidate combinations in the first ranking sequence is adjusted again, prioritizing the retention of combinations with lower implementation costs and lower implementation risks. During the adjustment process, the core priority of the predicted decrease in dynamic health entropy value remains unchanged. After the second ranking is completed, a personalized health candidate set that fits the actual implementation conditions of the target user is obtained.

[0171] The beneficial effects are that by identifying the synergistic and repulsive relationships of intervention primitives based on the adjustment intensity value, the interaction patterns between measures can be clarified, combination conflicts can be avoided, and a reliable basis can be provided for subsequent reasonable combinations.

[0172] Topological reconstruction of synergistic and repulsive relationships can clearly present the correlation structure of intervention measures, facilitating intuitive analysis and efficient traversal, and improving the efficiency and rationality of intervention combination selection.

[0173] By traversing the action graph with the goal of minimizing dynamic health entropy, highly efficient and synergistic intervention combinations can be prioritized to ensure that candidate solutions can specifically reduce health risks and improve the effectiveness of interventions.

[0174] Sort by the predicted decrease in dynamic health entropy value in descending order, which can give priority to combinations with better intervention effects, providing a scientific and intuitive basis for selecting personalized health plans.

[0175] Obtaining the implementation costs and risk levels of candidate combinations allows for a comprehensive assessment of the feasibility and safety of the proposed solutions, avoiding a focus solely on results while neglecting implementation conditions.

[0176] By combining cost and risk in a secondary ranking process, low-cost and low-risk options can be selected to improve the practicality of health management plans while ensuring the effectiveness of interventions.

[0177] A6: Perform expected effect deduction on the personalized health candidate set to obtain the optimized health management plan for the target user.

[0178] In this embodiment of the invention, the step of performing expected effect deduction on the personalized health candidate set to obtain an optimized health management plan for the target user includes:

[0179] Obtain the implementation sequence relationship from the personalized health candidate set;

[0180] The implementation time sequence relationship is temporally correlated with the target user's historical health data, and the correlation result is used to perform path evolution to generate the target user's expected health evolution path;

[0181] The expected health evolution path is calibrated with indicators to obtain the effect evaluation indicators for the target users.

[0182] Based on the aforementioned effectiveness evaluation indicators, the personalized health candidate set is comprehensively evaluated to obtain an optimized health management plan for the personalized health candidate set.

[0183] The execution sequence and interval of each intervention measure are extracted from the individual health candidate set. This execution sequence is derived from the standard intervention execution specifications recorded in the preset health intervention knowledge base, and is also compiled in combination with the collaborative constraints in the intervention interaction diagram.

[0184] The extracted implementation time sequence is bound to the target user's long-term accumulated historical health data according to the same time scale. The historical health data comes from the target user's wearable device monitoring records, electronic medical records, and physical examination report data. The bound results are continuously extrapolated in chronological order to gradually show the continuous change process of health status after the implementation of intervention measures, and generate an expected health evolution path that can fully reflect the future health change trend.

[0185] Key time points and key trends in the generated expected health evolution path are marked. The marked content includes the expected changes and stable states of physiological indicators, behavioral indicators, and dynamic health entropy values. All marked content is uniformly organized into an effect evaluation index that can quantify the intervention effect. This index is entirely derived from the actual evolution results of the expected health evolution path.

[0186] Using effectiveness evaluation indicators as the core evaluation criteria, each program in the personalized health candidate set is comprehensively compared and evaluated from multiple dimensions such as intervention effect, implementation cost, implementation risk, and health improvement. The program with the best effect evaluation and most suitable for the target user's health status is selected, and the selection result is the optimized health management program for the target user.

[0187] The beneficial effects are that obtaining the implementation sequence of personalized health candidate sets can clarify the execution logic of intervention measures and provide a standardized and clear process foundation for subsequent time-series correlation and effect extrapolation.

[0188] By linking the implementation timeline with historical health data and analyzing the path evolution, it is possible to scientifically simulate the health change trend after intervention and generate expected health evolution paths that fit the user's actual situation.

[0189] By defining indicators for the expected health evolution path, quantitative and objective performance evaluation indicators can be formed, providing a unified and reliable basis for evaluation of the program.

[0190] By comprehensively evaluating candidate programs based on effectiveness assessment indicators, we can select optimized health management programs with excellent intervention effects and strong feasibility, thereby improving the scientific nature and implementation of the programs.

[0191] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0192] This application embodiment can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0193] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A personalized health management program data mining method, characterized by, The method includes: A1: Obtain multi-source heterogeneous health record data of the target user, perform spatiotemporal alignment processing on the multi-source heterogeneous health record data, and generate a standard health data sequence of the target user; A2: Extract the physiological indicator fluctuation characteristics and behavioral pattern correlation characteristics from the standard health data sequence to construct the health feature vector set of the target user; A3: Nonlinearly aggregate the deviation and dispersion rate of the health feature vector set to obtain the dynamic health entropy value of the target user, so as to divide the health level range of the target user; A4: Input the health level range into a preset health intervention knowledge base, determine the intervention measure primitives that match the health feature vector set, associate and map the intervention measure primitives with the dynamic health entropy value, and construct the intervention mapping matrix for the target user; A5: With minimizing the dynamic health entropy as the optimization objective, the intervention measure primitives in the intervention mapping matrix are combined, screened, and prioritized to generate a personalized health candidate set for the target user; A6: Perform expected effect deduction on the personalized health candidate set to obtain the optimized health management plan for the target user.

2. The personalized health management program data mining method of claim 1, wherein, The process of acquiring multi-source heterogeneous health record data of the target user, performing spatiotemporal alignment processing on the multi-source heterogeneous health record data, and generating a standard health data sequence for the target user includes: Simultaneously collect wearable device monitoring data, electronic medical record data, and physical examination report text data of the target user, and integrate them into multi-source heterogeneous health record data of the target user; Using the timestamp of the wearable device monitoring data as a reference, linear interpolation is performed to align the timestamps in the electronic medical record data and the physical examination report text data to obtain the target user's synchronized health data; Based on the synchronized health data, the electronic medical record data and the physical examination report text data are normalized to obtain the normalized health data of the target user. The normalized health data is classified, and the missing values ​​in the classified data are corrected by difference based on the historical health data mean of the target user, so as to obtain the standard health data sequence of the target user.

3. The data mining method for a personalized health management plan as described in claim 1, characterized in that, The step of extracting physiological indicator fluctuation characteristics and behavioral pattern correlation characteristics from the standard health data sequence to construct the health feature vector set of the target user includes: Data decoupling is performed on the standard health data sequence to obtain the behavioral parameter subsequence and physiological parameter subsequence of the target user; Fluctuation analysis is performed on the physiological indicators in the physiological parameter subsequence to obtain the fluctuation characteristics of the target user's physiological indicators; A time correlation analysis was performed on the step count and sleep duration indicators in the behavioral parameter subsequence to obtain the behavioral pattern association features of the target user. The physiological indicator fluctuation characteristics and the behavioral pattern association characteristics are vectorized and concatenated to generate the health feature vector set of the target user.

4. The data mining method for a personalized health management plan as described in claim 1, characterized in that, The step of nonlinearly aggregating the deviation and dispersion rate of the health feature vector set to obtain the dynamic health entropy value of the target user, in order to divide the health level range of the target user, includes: Extract the center point of each dimension of the feature vector in the health feature vector set, and calculate the Mahalanobis distance from each feature vector in the health feature vector set to the center point; The deviation of the Mahalanobis distance is measured to obtain the degree of deviation of the health feature vector set; Extract the standard deviation and mean of each dimension of the feature vector in the health feature vector set, and use the ratio of the standard deviation to the mean as the dispersion rate of the health feature vector set.

5. The data mining method for a personalized health management plan as described in claim 4, characterized in that, The step of using the ratio of the standard deviation to the mean as the dispersion rate of the health feature vector set includes: The deviation degree and the dispersion rate are nonlinearly fused to generate the dynamic health entropy value of the target user, wherein the calculation formula of the dynamic health entropy value is: ； In the formula, The dynamic health entropy value, The weighting coefficient for the degree of deviation. The degree of deviation, It is an exponential function. The weighting coefficients for the dispersion rate are, This is the preset deviation saturation adjustment coefficient. The dispersion rate is... It is a logarithmic function; Obtain a preset set of health level thresholds, compare the dynamic health entropy value with the preset set of health level thresholds, and determine the threshold range of the dynamic health entropy value. Based on the threshold range, the health level labels of the preset health level threshold set are matched to divide the health level range of the target user.

6. The data mining method for a personalized health management plan as described in claim 4, characterized in that, The step of inputting the health level range into a preset health intervention knowledge base and determining the intervention measure primitives that match the health feature vector set includes: Extract the level labels corresponding to the health level range; Input the level label into the preset health intervention knowledge base and locate the intervention strategy index table associated with the level label; Using the set of health feature vectors as query vectors, similarity matching is performed in the intervention strategy index table to obtain a set of candidate intervention measures for the target user; Extract the execution conditions and objectives of the candidate intervention measures, and eliminate intervention measures that are incompatible with the health feature vector set to obtain the intervention measure primitives for the target user.

7. The data mining method for a personalized health management plan as described in claim 6, characterized in that, The step of associating and mapping the intervention measure primitives with the dynamic health entropy value to construct the intervention mapping matrix for the target user includes: Obtain the effect dimension labels and expected adjustment magnitudes from the intervention measure primitives; The dynamic health entropy value is decomposed into the action dimension label to obtain the entropy component of the dynamic health entropy value; The expected adjustment range is multiplied by the entropy component to obtain the adjustment intensity value of the intervention measure primitive; Using the intervention measure primitive as the row index and the effect dimension label as the column index, the adjustment intensity value is filled into the corresponding position in the matrix to construct the intervention mapping matrix for the target user.

8. The data mining method for a personalized health management plan as described in claim 7, characterized in that, The step of minimizing the dynamic health entropy value as the optimization objective involves combining, screening, and prioritizing the intervention measure primitives in the intervention mapping matrix to generate a personalized health candidate set for the target user, including: Based on the adjustment intensity value, identify the synergistic and repulsive relationships between the intervention measure primitives in the intervention mapping matrix; By performing topological reconstruction on the cooperative and repulsive relationships, an interaction graph of the intervention measures for the target user is obtained; With minimizing the dynamic health entropy as the optimization objective, the interaction graph of the intervention measures is traversed to generate a candidate combination set for the target user.

9. The data mining method for a personalized health management plan as described in claim 8, characterized in that, The step of minimizing the dynamic health entropy value as the optimization objective, traversing the interaction graph of the intervention measures, and generating a candidate combination set for the target user, includes: Based on the predicted decrease in the dynamic health entropy value, the candidate combination set is sorted in descending order to obtain the first sorting sequence of the target user. Obtain the implementation cost label and implementation risk level from the candidate combination set; Based on the implementation cost label and the implementation risk level, the candidate combinations in the first sorting sequence are sorted a second time to obtain the personalized health candidate set of the target user.

10. The data mining method for a personalized health management plan as described in claim 2, characterized in that, The step of extrapolating the expected effects from the personalized health candidate set to obtain an optimized health management plan for the target user includes: Obtain the implementation sequence relationship from the personalized health candidate set; The implementation time sequence relationship is temporally correlated with the target user's historical health data, and the correlation result is used to perform path evolution to generate the target user's expected health evolution path; The expected health evolution path is calibrated with indicators to obtain the effect evaluation indicators for the target users. Based on the aforementioned effectiveness evaluation indicators, the personalized health candidate set is comprehensively evaluated to obtain an optimized health management plan for the personalized health candidate set.

Images