Health condition comprehensive evaluation and intervention decision method and system based on ai

Abstract

This invention discloses an AI-based method and system for comprehensive health status evaluation and intervention decision-making, comprising: collecting multi-source health monitoring data to construct a multimodal health data set; extracting associated features through a multimodal health feature fusion inference model to generate a high-dimensional fused health feature vector; inputting the vector into a multi-dimensional physical condition dynamic assessment model to obtain physical condition assessment results; combining health demand preference data with a personalized intervention strategy generation and deduction model to generate a preliminary set of intervention strategies; inputting both into a health risk level AI quantitative analysis platform to output health risk quantification results and intervention strategy optimization suggestions; and integrating these to generate a comprehensive evaluation report and a personalized intervention implementation plan. The data collection in this invention is achieved through dedicated hardware collaboration. Through multi-model collaboration and multi-unit linkage, it realizes deep fusion of health data, dynamic assessment of physical condition, precise risk quantification, and personalized generation of intervention strategies, thereby improving the comprehensiveness and accuracy of health management.

Description

Technical Field

[0001] This invention relates to the field of health status assessment and intervention technology, and in particular to an AI-based comprehensive health status assessment and intervention decision-making method and system. Background Technology

[0002] With the increasing aging population and the high incidence of chronic diseases, health management needs are shifting from traditional diagnosis and treatment to proactive prevention and personalized intervention. The integrated analysis and intelligent decision-making of multi-source health data have become core industry demands. Currently, the demand for comprehensive utilization of multi-dimensional data, including physiological indicators, behavioral patterns, and environmental influences, in the field of health management is increasingly urgent. There is a pressing need to overcome the limitations of traditional assessment methods through intelligent technologies, achieving closed-loop management of the entire process from data collection to intervention execution. This would provide more comprehensive and dynamic technical support for health status assessment and meet the personalized health management needs of different populations. Existing technologies mostly achieve health assessment through single-modal data collection and basic algorithm models. They typically rely on independent monitoring devices to obtain physiological or behavioral data, which is then input into the assessment model after simple feature extraction, outputting basic health status results, and then combined with generalized intervention plans to form recommendations. Some technologies attempt to integrate small amounts of multi-source data, but lack a systematic feature fusion mechanism. They often use fixed weight allocation methods to process information from different sources, and intervention strategy generation is mostly based on preset rules, failing to fully consider individual differences in physical condition, risk levels, and needs and preferences. The overall process lacks dynamic adjustment and in-depth optimization technical design.

[0003] Existing technologies have two main drawbacks: First, the multimodal data fusion effect is insufficient, failing to achieve deep correlation and complementary reasoning of health data from different sources. The feature extraction process lacks accurate mining of data correlations, making it difficult for the assessment results to fully reflect the true health status. Second, personalization is lacking. The intervention strategy generation does not fully incorporate individual dynamic changes in physical condition, differences in health risk levels, and personalized needs and preferences. Furthermore, it lacks an interactive optimization mechanism with risk quantification results, making the intervention plan insufficiently targeted and feasible, and unable to effectively adapt to the specific health management needs of different users. Summary of the Invention

[0004] To overcome the shortcomings and deficiencies of existing technologies, this invention provides an AI-based method and system for comprehensive health status evaluation and intervention decision-making.

[0005] The technical solution adopted in this invention is an AI-based comprehensive health status evaluation and intervention decision-making method, comprising the following steps: S1, collecting multi-source health monitoring data, including physiological indicator detection data, behavioral pattern recording data, environmental exposure correlation data, and genetic characteristic correlation data, to construct a multimodal health data set; S2, calling a multimodal health feature fusion inference model to perform feature extraction and correlation inference on the multimodal health data set, generating a high-dimensional fused health feature vector, wherein the fusion process is achieved through multi-dimensional feature weight allocation and cross-modal information complementary inference; S3, inputting the high-dimensional fused health feature vector into a multi-dimensional physical fitness dynamic assessment model, and performing multi-dimensional physical fitness indicator correlation analysis. S4. Based on the physical condition assessment results, a deduction model is generated using personalized intervention strategies, and a preliminary set of intervention strategies is generated by combining health demand preference data and intervention measure suitability analysis; S5. The physical condition assessment results and the preliminary set of intervention strategies are input into the health risk level AI quantitative analysis platform, and the health risk quantification results and intervention strategy optimization suggestions are output through multi-dimensional risk factor correlation calculation and level division rule matching; S6. The health risk quantification results and intervention strategy optimization suggestions are integrated through the health status comprehensive evaluation and intervention decision module to generate the final health status comprehensive evaluation report and personalized intervention implementation plan.

[0006] Furthermore, the expression for the multimodal health feature fusion inference model is: ,in To create a high-dimensional fusion health feature vector, Let i be the feature weights of the i-th type of physiological indicator detection data. Let i be the feature mapping function for the i-th type of physiological indicator detection data. Let be the original feature matrix of the i-th type of physiological indicator detection data. Let i be the feature offset of the i-th type of physiological indicator detection data. It is a non-linear activation function. Record the feature weights of the data for the j-th type of behavioral pattern. The feature mapping function for recording data of the j-th type of behavioral pattern. Here, n represents the original feature matrix of the data recording the j-th type of behavioral pattern, and n is the number of categories of physiological indicator detection data. The number of categories for recording data on behavioral patterns is determined; the feature fusion confidence expression is calculated as follows: ,in For feature fusion confidence, The maximum value of the high-dimensional fused health feature vector. Let k be the confidence correction function for the k-th type of environmental exposure association data. Let be the feature parameter of the k-th type of environmental exposure associated data, and p be the number of categories of environmental exposure associated data.

[0007] Furthermore, the expression for the multidimensional physical condition dynamic assessment model is: ,in This is the result of the physical fitness assessment. Let q be the evaluation weight of the qth physical fitness indicator. Let be the time-series evaluation function for the q-th physical fitness indicator. This represents a high-dimensional fused health feature vector, where t is the current evaluation time point. Associating weights with historical states This is a function relating physical condition to historical correlation. This refers to the previous assessment time point. The number of physical fitness indicators; the expression for the trend of physical fitness status change is calculated as follows: ,in This represents the trend value of changes in physical condition. This represents the physical condition assessment results at the current point in time. This is the result of the physical condition assessment at the previous time point. Let r be the weight of the change in the physical fitness influencing factor. Let r be the correlation function for the change of the r-th physical fitness influencing factor. This represents the temporal variation of the high-dimensional fused health feature vector. For the evaluation time interval, s is the number of items in the physical fitness influencing factors.

[0008] Furthermore, the expression for the personalized intervention strategy generation and deduction model is as follows: ,in, This is a preliminary set of intervention strategies. The basic adaptation weights for the u-th intervention are: Let u be the fitness function between the u-th intervention and the physical condition. This is the result of the physical fitness assessment. Let be the execution parameter matrix for the u-th intervention. To assign risk-adapted weights to the u-th intervention, Let u be the fit function between the u-th intervention and the health risk. To quantify health risks, Let be the risk control parameter matrix for the u-th intervention, and k be the total number of interventions.

[0009] Furthermore, the multi-source health monitoring data acquisition is achieved collaboratively through a flexible wearable biosensor array, a portable multi-parameter detection terminal, and a cloud-based distributed data acquisition node. The flexible wearable biosensor array integrates a photoelectric volume pulse detection unit, a skin impedance detection unit, and a motion posture sensing unit, with a sampling frequency set to 500Hz to 800Hz and data transmission using a millimeter-wave communication protocol. The portable multi-parameter detection terminal has a built-in microfluidic chip detection module and a quantum dot sensing module, supporting the simultaneous detection of more than 15 physiological indicators. The cloud-based distributed data acquisition node adopts an edge computing architecture, deploys a lightweight data filtering algorithm, and has a data cache capacity of no less than 1TB. The feature mapping function of the multimodal health feature fusion inference model adopts a deep residual network structure with 32 network layers and the Mish function as the activation function. The temporal evaluation function of the multidimensional physical condition dynamic assessment model adopts a long short-term memory network variant with 256 hidden layer neurons.

[0010] Further, S2 includes the following sub-steps: S21, performing feature dimension analysis on various types of data in the multimodal health dataset, filtering out features strongly correlated with physical condition through feature correlation analysis, and establishing a feature index table; S22, using a multi-scale feature extraction method to extract features hierarchically, obtaining local and global features at different scales, and constructing a multi-scale feature set; S23, performing correlation analysis on different modal features in the multi-scale feature set through a cross-modal feature association algorithm, establishing feature mapping relationships between modalities, and generating a feature association matrix; S24, inputting the feature association matrix into the multimodal health feature fusion inference model, performing deep fusion of multimodal features through model computation, and outputting a high-dimensional fused health feature vector.

[0011] Further, step S3 includes the following sub-steps: S31, dimensional normalization of the high-dimensional fused health feature vector, extraction of feature subsets associated with physical fitness assessment, and determination of physical fitness assessment index system; S32, based on the assessment index system, construction of time-series feature sequence, mining of dynamic change patterns of physical fitness through time-series data analysis methods, and generation of time-series change features; S33, inputting the feature subset and time-series change features into a multi-dimensional physical fitness dynamic assessment model, performing quantitative assessment of physical fitness through multi-dimensional calculations of the model, and obtaining preliminary assessment results; S34, consistency verification of the preliminary assessment results, correction of assessment deviations through multi-index cross-validation methods, and output of the final physical fitness assessment results.

[0012] Further, S4 includes the following sub-steps: S41, collecting user health need preference data, including intervention method preference, intervention intensity preference, and intervention cycle preference data, and establishing a preference data archive; S42, performing intervention need analysis on the physical condition assessment results, determining the health dimensions that need key intervention under the user's current physical condition, and generating an intervention need list; S43, based on the intervention need list and preference data archive, screening suitable intervention measure types, and constructing an intervention measure candidate set; S44, inputting the intervention measure candidate set into a personalized intervention strategy generation inference model, and combining the model's inference calculations to generate a preliminary intervention strategy set that meets the user's needs.

[0013] Further, S5 includes the following sub-steps: S51, extracting risk factors from the physical condition assessment results, identifying potential health risk factors, and establishing a risk factor database; S52, inputting the risk factors and the set of preliminary intervention strategies into the health risk level AI quantitative analysis platform, and using the platform's risk quantification algorithm to quantify each risk factor to obtain a risk quantification value; S53, based on the risk quantification value, classifying health risks according to preset risk level classification rules, and determining the risk level of each risk; S54, combining the risk level and the set of preliminary intervention strategies, adjusting and optimizing the preliminary intervention strategies using a strategy optimization algorithm, and outputting intervention strategy optimization suggestions and the final health risk quantification result.

[0014] This AI-based comprehensive health status assessment and intervention decision-making system applies AI-based comprehensive health status assessment and intervention decision-making methods. It includes: a multimodal health data collaborative acquisition and preprocessing unit, a high-dimensional fusion health feature intelligent extraction and inference unit, a multi-dimensional physical condition dynamic quantitative assessment unit, a personalized intervention strategy intelligent generation and deduction unit, a health risk level AI quantitative analysis unit, and a comprehensive health status assessment and intervention decision output unit. The multimodal health data collaborative acquisition and preprocessing unit and the high-dimensional fusion health feature intelligent extraction and inference unit are connected via a high-speed data transmission link for real-time transmission and feature extraction of multi-source health data. The high-dimensional fusion health feature intelligent extraction and inference unit and the multi-dimensional physical condition dynamic quantitative assessment unit are connected via a feature vector transmission interface to transmit high-dimensional fusion data. Health feature vectors are transmitted to a multidimensional physical condition dynamic quantitative assessment unit for physical condition evaluation. This unit connects to a personalized intervention strategy intelligent generation and deduction unit and a health risk level AI quantitative analysis unit, simultaneously outputting physical condition assessment results. The personalized intervention strategy intelligent generation and deduction unit and the health risk level AI quantitative analysis unit are connected via a strategy feedback link to perform interactive optimization of intervention strategies and risk quantification results. The health risk level AI quantitative analysis unit connects to a health status comprehensive evaluation and intervention decision output unit, transmitting health risk quantification results and intervention strategy optimization suggestions to this unit. The health status comprehensive evaluation and intervention decision output unit integrates all input information to generate and output a health status comprehensive evaluation report and a personalized intervention implementation plan.

[0015] Beneficial Effects: This invention proposes an AI-based comprehensive health status evaluation and intervention decision-making method and system. Through a multimodal health feature fusion and reasoning mechanism, it achieves correlation and complementarity of health data from different sources and precise feature extraction, overcoming the shortcomings of insufficient multimodal data fusion in existing technologies, allowing health assessment results to more comprehensively reflect the true state. By mining temporal change patterns through a multidimensional physical condition dynamic assessment model and combining it with AI-based quantitative analysis of health risk levels to achieve precise risk grading, the personalized intervention strategy generation and deduction model fully incorporates individual needs, preferences, and risk suitability analysis, solving the problem of insufficient personalization and adaptability in existing intervention programs and improving the targeting and feasibility of intervention measures. The system, through the coordinated linkage of six functional units and innovative hardware configuration, ensures the comprehensiveness and real-time nature of data collection. Deep optimization of model algorithms and step-by-step detailed execution ensure the accuracy of assessment and decision-making. The overall process forms an intelligent closed loop from data collection to intervention execution, providing users with more scientific and personalized health management support and effectively meeting the needs of proactive prevention and precise health management. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating the overall process of the method of the present invention. Figure 2 This is a flowchart of method step S2 of the present invention; Figure 3 This is a flowchart of method step S3 of the present invention; Figure 4 This is a flowchart of method step S4 of the present invention; Figure 5 This is a flowchart of step S5 of the method of the present invention; Figure 6 This is a diagram showing the system unit composition of the present invention. Detailed Implementation

[0017] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. The application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0018] like Figure 1 As shown, the AI-based comprehensive health status evaluation and intervention decision-making method includes the following steps: S1, collecting multi-source health monitoring data, including physiological indicator detection data, behavioral pattern recording data, environmental exposure correlation data, and genetic characteristic correlation data, to construct a multimodal health data set; S2, calling a multimodal health feature fusion inference model to perform feature extraction and correlation inference on the multimodal health data set, generating a high-dimensional fused health feature vector, wherein the fusion process is achieved through multi-dimensional feature weight allocation and cross-modal information complementary inference; S3, inputting the high-dimensional fused health feature vector into a multi-dimensional physical fitness dynamic assessment model, and performing multi-dimensional physical fitness indicator correlation analysis and time-series changes. S4. Based on the physical condition assessment results, a model is generated using personalized intervention strategies. Combined with health demand preference data and intervention measure suitability analysis, a preliminary set of intervention strategies is generated. S5. The physical condition assessment results and the preliminary set of intervention strategies are input into the health risk level AI quantitative analysis platform. Through multi-dimensional risk factor correlation calculation and level division rule matching, the health risk quantification results and intervention strategy optimization suggestions are output. S6. The health status comprehensive evaluation and intervention decision module integrates the health risk quantification results and intervention strategy optimization suggestions to generate a final health status comprehensive evaluation report and personalized intervention implementation plan.

[0019] Step S1 involves the comprehensive collection and aggregation of multi-source health monitoring data, specifically achieved through the collaborative operation of a flexible wearable biosensor array, a portable multi-parameter detection terminal, and a cloud-based distributed data acquisition node. The flexible wearable biosensor array integrates a photoelectric volumetric pulse detection unit, a skin impedance detection unit, and a motion posture sensing unit, with a sampling frequency set to 500Hz to 800Hz. Data transmission utilizes a millimeter-wave communication protocol, enabling real-time collection of physiological indicators such as heart rate, skin resistance, and motion amplitude. The portable multi-parameter detection terminal incorporates a microfluidic chip detection module and a quantum dot sensing module, supporting the simultaneous detection of more than 15 physiological indicators, including key data such as blood glucose, blood lipids, and blood oxygen saturation. The cloud-based distributed data acquisition node employs an edge computing architecture, deploying lightweight data filtering algorithms with a data cache capacity of no less than 1TB. It is responsible for collecting environmental exposure-related data such as ambient temperature and humidity, and air quality, as well as genetic characteristic-related data linked to gene sequences. During the data collection process, the three types of devices work synchronously according to a preset time sequence, completing a data summary every 10 minutes to ensure the continuity and integrity of data collection. All collected physiological indicator detection data, behavioral pattern recording data, environmental exposure-related data, and genetic characteristic-related data are uniformly stored in a cloud database. Through standardized data format processing, a multimodal health data set with a unified structure and comprehensive content is constructed, providing a high-quality data foundation for subsequent feature fusion and model calculation. This step, through multi-device collaboration and high-frequency sampling, ensures the comprehensiveness of data coverage dimensions and the accuracy of the data itself, laying a reliable data support for the entire health assessment and intervention decision-making process.

[0020] Step S2 involves the extraction and correlation inference of multimodal health features. A high-dimensional fused health feature vector is generated through a multimodal health feature fusion inference model. The specific implementation process includes four core operations. First, the feature dimensions of various data in the constructed multimodal health dataset are analyzed. Core features strongly correlated with physical condition are selected through feature correlation analysis, a core feature index table is established, and redundant data and irrelevant features are removed to improve computational efficiency. Then, a multi-scale feature extraction method is used to extract core features hierarchically, dividing the data into three scales: fine-grained, medium-grained, and coarse-grained. Local and global features at different scales are obtained respectively, constructing a multi-scale feature set to ensure multi-level coverage of feature information. Next, a cross-modal feature correlation algorithm is used to analyze the correlation between different modal features in the multi-scale feature set. The correlation strength between different modal features is calculated based on the Pearson correlation coefficient and mutual information entropy, establishing inter-modal feature correlation. The feature mapping relationship is used to generate a feature association matrix, realizing the association mapping of data from different sources. Finally, the feature association matrix is ​​input into a multimodal health feature fusion inference model. The feature mapping function of this model adopts a 32-layer deep residual network structure, and the activation function adopts the Mish function. Deep feature fusion is achieved through multi-dimensional feature weight allocation and cross-modal information complementary inference. The weight allocation is dynamically adjusted based on the feature importance assessment results, and the cross-modal information complementary inference is achieved through a modality missing information completion algorithm. The final output is a high-dimensional fused health feature vector with a dimension of 1024. This step, through multi-step refinement and model optimization design, achieves efficient fusion of multimodal data and accurate feature extraction, providing high-value feature input for subsequent physical condition assessment.

[0021] Step S3 outputs the physical fitness status assessment results through a multidimensional dynamic assessment model. The implementation process revolves around four key stages: feature processing, time-series analysis, model computation, and result verification. First, the high-dimensional fused health feature vector output from Step S2 is dimensionally normalized. Principal component analysis is used to extract principal components with a variance contribution rate greater than 85%, forming a key feature subset related to physical fitness status assessment. Combining traditional Chinese medicine constitution classification standards and modern medical health assessment indicators, a physical fitness status assessment indicator system is determined, comprising 36 indicators across six categories, including body function, metabolic level, immune status, and organ function. Based on this indicator system, a time-series feature sequence for the past 30 days is constructed at hourly intervals. The sliding window method and trend fitting algorithm are used to mine the dynamic changes in physical fitness status, generating time-series change features, including key information such as the rate of indicator change, the stability of the change trend, and the frequency of extreme values. The key feature subset and the time-series change features are then input together. A multidimensional physical fitness dynamic assessment model is proposed. This model employs a variant of a long short-term memory network with 256 hidden layer neurons for its temporal evaluation function. It achieves quantitative assessment of physical fitness status through multidimensional physical fitness index correlation analysis and temporal trend extrapolation, outputting preliminary assessment results. These results are expressed as a percentage, including overall physical fitness scores and individual index scores. Finally, the preliminary assessment results undergo consistency verification. A multi-index cross-validation method is used to calculate the consistency coefficient of results across different assessment dimensions. If the consistency coefficient is below 0.85, the model parameters are readjusted and the calculation is performed to correct assessment bias. The final output is a physical fitness assessment result with a consistency coefficient above 0.85. This step, through dynamic assessment and multiple rounds of verification, ensures the accuracy and reliability of the assessment results.

[0022] Step S4 generates a preliminary set of intervention strategies tailored to individual needs. A personalized intervention strategy generation and deduction model is used to achieve precise adaptation of the intervention strategies. The specific implementation process is divided into four key stages. First, user health needs and preferences data are collected. Users fill in their intervention method preferences (e.g., exercise intervention, dietary intervention, lifestyle intervention), intervention intensity preferences (mild, moderate, severe), and intervention period preferences (short-term, medium-term, long-term) through a terminal interface. Simultaneously, combined with historical intervention feedback data and health goal setting information, a preference data file including 12 preference indicators is established to ensure comprehensive collection of preference information. Next, the physical condition assessment results output in Step S3 are used for intervention needs analysis. Based on the differences between the scores of each indicator and the health standard thresholds, the health dimensions requiring key intervention under the user's current physical condition are determined, such as low metabolic level and weak immune status. Combining the importance weights of the indicators, an intervention needs list including intervention priority, intervention goals, and intervention focus is generated. Based on the intervention needs... The system uses a list and preference data archive to screen suitable intervention types from an intervention library containing over 200 intervention measures. These are then sorted according to their relevance to intervention effectiveness and preference matching, with the top 20 selected to construct a candidate set of intervention measures, ensuring both targeting and diversity. Finally, this candidate set is input into a personalized intervention strategy generation model. The model evaluates the suitability of each intervention measure to the individual's physical condition and health risks, combining the implementation parameters and risk control parameters of the intervention measures. Through multi-dimensional suitability calculations, it generates a preliminary set of intervention strategies. Each strategy includes 3-5 intervention measures, clearly defining the execution order, implementation details, and expected effects of each measure. This step, by fully considering individual differences and needs / preferences, achieves personalized generation of intervention strategies, providing a foundation for subsequent optimization and adjustments.

[0023] Step S5 utilizes the AI-powered health risk grading analysis platform to output health risk grading results and intervention strategy optimization suggestions. The process includes four core steps: risk factor extraction, risk grading calculation, risk grading, and strategy optimization. First, risk factors are extracted from the physical condition assessment results output in Step S3. Based on a disease risk association database, potential health risks corresponding to abnormal indicators are identified, including chronic disease risks and functional decline risks. A risk factor database is established, including information on risk type, risk triggers, and scope of impact, encompassing 48 common health risk factors. The key risk factors and the preliminary intervention strategy set generated in Step S4 are input into the AI-powered health risk grading analysis platform. This platform employs a hybrid architecture combining deep learning and a rule engine. Its built-in risk grading algorithm quantifies each risk factor, calculating a weighted sum based on the probability of occurrence, degree of impact, and speed of development. The quantified value ranges from 0 to 100 points. Based on the risk grading value, and according to preset risk grading rules, health risks are categorized... Risk is categorized into five levels: extremely low risk (0-20 points), low risk (21-40 points), medium risk (41-60 points), high risk (61-80 points), and extremely high risk (81-100 points) to determine the risk level of each risk. Finally, combining the risk level with the initial intervention strategy set, the initial intervention strategies are adjusted and optimized through a strategy optimization algorithm. For high-risk and extremely high-risk items, the intensity and frequency of intervention measures are increased, and targeted intervention methods are added. For low-risk items, the intervention process is simplified, core measures are retained, and the synergistic and antagonistic effects between intervention measures are considered to optimize the combination of measures. The final output includes a quantitative result of health risk, including the risk level, quantitative score, risk description, and optimized intervention strategy recommendations. This step improves the scientificity and feasibility of the intervention strategy through precise risk quantification and dynamic strategy optimization.

[0024] Step S6 integrates all previously output information through the comprehensive health status evaluation and intervention decision-making module to generate the final comprehensive health status evaluation report and personalized intervention implementation plan. The implementation process includes four key operations: information integration, comprehensive evaluation, plan refinement, and output presentation. First, it receives the health risk quantification results and intervention strategy optimization suggestions from Step S5 via a high-speed data interface. Simultaneously, it calls upon the physical condition assessment results from Step S3 and the key information from the high-dimensional fusion health feature vector from Step S2 to establish a unified information integration framework. Information is associated and integrated according to the logical links of data, features, assessment, risk, and strategy to ensure the completeness and relevance of the information. Then, based on the integrated information, a comprehensive health status evaluation is conducted, constructing a three-dimensional evaluation system including a basic health score, a risk warning score, and an intervention suitability score. The basic health score is calculated based on the physical condition assessment results, the risk warning score is determined based on the health risk quantification results, and the intervention suitability score is based on the assessment of the suitability between the intervention strategy and the individual. The three scores are weighted and summed to obtain a comprehensive health score, while generating detailed evaluation descriptions for each indicator, clarifying the strengths and weaknesses. Next, the intervention strategy optimization suggestions are further refined. The process is refined by taking into account users' lifestyles, work rhythms, geographical environments, and other actual circumstances. The specific implementation time, operation methods, execution standards, and monitoring frequency for each intervention measure are clearly defined. For example, exercise intervention specifies the type of exercise, duration, intensity, weekly frequency, and monitoring indicators; dietary intervention specifies the daily dietary structure, ingredient selection, cooking methods, and intake control, forming a directly executable personalized intervention plan. Finally, a comprehensive health status evaluation report is generated according to a unified report template, including five parts: comprehensive health score, detailed indicator evaluation, risk level explanation, intervention implementation plan, and precautions. This report is presented to users and health managers through three formats: a terminal APP, a web page, and a paper report. This step, through information integration, comprehensive evaluation, plan refinement, and multi-format output, achieves a closed-loop output of health evaluation and intervention decision-making, providing users with comprehensive, accurate, and actionable health management guidance.

[0025] Preferably, the expression of the multimodal health feature fusion inference model is: ,in To create a high-dimensional fusion health feature vector, Let i be the feature weights of the i-th type of physiological indicator detection data. Let i be the feature mapping function for the i-th type of physiological indicator detection data. Let be the original feature matrix of the i-th type of physiological indicator detection data. Let i be the feature offset of the i-th type of physiological indicator detection data. It is a non-linear activation function. Record the feature weights of the data for the j-th type of behavioral pattern. The feature mapping function for recording data of the j-th type of behavioral pattern. Here, n represents the original feature matrix of the data recording the j-th type of behavioral pattern, and n is the number of categories of physiological indicator detection data. The number of categories for recording data on behavioral patterns is determined; the feature fusion confidence expression is calculated as follows: ,in For feature fusion confidence, The maximum value of the high-dimensional fused health feature vector. Let k be the confidence correction function for the k-th type of environmental exposure association data. Let be the feature parameter of the k-th type of environmental exposure associated data, and p be the number of categories of environmental exposure associated data.

[0026] Specifically, the multimodal health feature fusion inference model is based on the principle of complementarity of multi-source data and the law of feature association. First, it addresses the heterogeneity of different modal data by achieving preliminary feature fusion through hierarchical mapping and weight allocation. During calculation, it first analyzes the feature distribution characteristics of physiological indicator detection data and behavioral pattern record data. Considering the core supporting role of the two types of data in health assessment, it uses product operation to reflect the synergistic effect of feature fusion. Then, it introduces a nonlinear activation function to enhance the model's ability to fit complex relationships, forming the first-stage fusion formula. Based on the high-dimensional fusion feature vector output from the first stage, it further combines the confidence correction requirements of environmental exposure-related data. Through normalization processing and product operation, it integrates the confidence correction factor to calculate the feature fusion confidence formula, ensuring the reliability of the fusion result. The formula is based on the correlation and complementarity of multimodal data. It dynamically adjusts the weights to adapt the importance of different data. The number of categories for physiological indicator detection data is set to 8, for behavioral pattern recording data to 6, and for environmental exposure-related data to 4. Feature weights range from 0.1 to 0.9, optimized using a gradient descent algorithm based on feature importance assessment results, with offset values ​​ranging from 0.01 to 0.05. In implementation, feature mapping is first performed on each type of data, followed by sequential fusion calculations and confidence corrections according to the formula. This model achieves deep fusion of multimodal data through two steps, effectively mining the correlation value of data from different sources, improving the comprehensiveness and accuracy of feature vectors, and providing high-quality input for subsequent physical condition assessment.

[0027] Preferably, the expression for the multidimensional physical condition dynamic assessment model is: ,in This is the result of the physical fitness assessment. Let q be the evaluation weight of the qth physical fitness indicator. Let be the time-series evaluation function for the q-th physical fitness indicator. This represents a high-dimensional fused health feature vector, where t is the current evaluation time point. Associating weights with historical states This is a function relating physical condition to historical correlation. This refers to the previous assessment time point. The number of physical fitness indicators; the expression for the trend of physical fitness status change is calculated as follows: ,in This represents the trend value of changes in physical condition. This represents the physical condition assessment results at the current point in time. This is the result of the physical condition assessment at the previous time point. Let r be the weight of the change in the physical fitness influencing factor. Let r be the correlation function for the change of the r-th physical fitness influencing factor. This represents the temporal variation of the high-dimensional fused health feature vector. For the evaluation time interval, s is the number of items in the physical fitness influencing factors.

[0028] Specifically, the multidimensional physical fitness dynamic assessment model takes the temporal dynamic change pattern of physical fitness as its core basis. First, it considers the comprehensive influence of multiple core indicators on physical fitness, using a summation operation to integrate the assessment results of each indicator. Simultaneously, it introduces historical state correlation terms to reflect the continuity of physical fitness, calculating the basic formula for physical fitness assessment. Based on the assessment results output by the basic formula, it further analyzes the changing trend of physical fitness. By calculating the difference between the current and historical assessment results, and combining the correlation effects of changes in physical fitness influencing factors, a summation operation is used to integrate the changes in each influencing factor, calculating the formula for the changing trend of physical fitness, thus achieving dynamic assessment. Based on the multidimensional characteristics and temporal correlation of physical fitness, the number of core physical fitness indicators is set to 36, with assessment weights determined using the analytic hierarchy process, ranging from 0.02 to 0.08. The historical state correlation weight is set to 0.3, and the number of physical fitness influencing factors is set to 12, with change weights ranging from 0.05 to 0.15. During implementation, high-dimensional fused feature vectors from different time points are first collected. After being input into the model, the physical condition assessment and trend calculation are completed sequentially according to the formula. The time interval for time-series assessment is set to 1 hour, and the historical data retrieval period is 30 days. This model achieves dynamic quantitative assessment of physical condition through two-step calculation, comprehensively considering the current core indicator status while fully combining historical change patterns and changes in influencing factors, ensuring that the assessment results accurately reflect the true physical condition and trend.

[0029] Preferably, the expression for the personalized intervention strategy generation and deduction model is: ,in, This is a preliminary set of intervention strategies. The basic adaptation weights for the u-th intervention are: Let u be the fitness function between the u-th intervention and the physical condition. This is the result of the physical fitness assessment. Let be the execution parameter matrix for the u-th intervention. To assign risk-adapted weights to the u-th intervention, Let u be the fit function between the u-th intervention and the health risk. To quantify health risks, Let be the risk control parameter matrix for the u-th intervention, and k be the total number of interventions.

[0030] Specifically, the personalized intervention strategy generation and deduction model is based on the principle of adaptability between individual physical condition, health risk, and intervention measures. Considering that intervention strategies need to adapt to both physical condition and health risk level, and that the adaptation weights of different intervention measures differ, a product operation is used to integrate the basic adaptability and risk adaptability of various intervention measures to form a single fusion formula. The formula is established based on the precise adaptation requirements of personalized intervention. A dual-weight design is used to reflect the degree of adaptability between intervention measures and physical condition and health risk. The total number of intervention measures is set to 20, and the basic adaptability weight and risk adaptability weight both range from 0.2 to 0.8. These values ​​are optimized and determined through a collaborative filtering algorithm combined with historical user feedback data. The execution parameter matrix includes 5 parameters such as intervention intensity and execution frequency, and the risk control parameter matrix includes 4 parameters such as risk mitigation coefficient and safety threshold. During implementation, the physical condition assessment results and health risk quantification results are obtained first, then the parameter matrices of each intervention measure are extracted, and multi-dimensional adaptability calculations are performed according to the formula to generate a preliminary set of intervention strategies. This formula strengthens the weight of highly adaptable interventions through product operations and weakens the impact of poorly adaptable measures. At the same time, it integrates execution parameters and risk control parameters to ensure that the generated intervention strategy not only conforms to individual physical characteristics but also accurately addresses health risks, thereby improving the pertinence and feasibility of intervention measures.

[0031] Preferably, the multi-source health monitoring data acquisition is achieved collaboratively through a flexible wearable biosensor array, a portable multi-parameter detection terminal, and a cloud-based distributed data acquisition node. The flexible wearable biosensor array integrates a photoelectric volume pulse detection unit, a skin impedance detection unit, and a motion posture sensing unit, with a sampling frequency set to 500Hz to 800Hz and data transmission using a millimeter-wave communication protocol. The portable multi-parameter detection terminal has a built-in microfluidic chip detection module and a quantum dot sensing module, supporting the simultaneous detection of more than 15 physiological indicators. The cloud-based distributed data acquisition node adopts an edge computing architecture, deploys a lightweight data filtering algorithm, and has a data cache capacity of no less than 1TB. The feature mapping function of the multimodal health feature fusion inference model adopts a deep residual network structure with 32 network layers and the Mish function as the activation function. The temporal evaluation function of the multidimensional physical condition dynamic assessment model adopts a long short-term memory network variant with 256 hidden layer neurons.

[0032] Preferred, such as Figure 2 As shown, S2 includes the following sub-steps: S21, performing feature dimension analysis on various types of data in the multimodal health dataset, filtering out features strongly correlated with physical condition through feature correlation analysis, and establishing a feature index table; S22, using a multi-scale feature extraction method to extract features hierarchically, obtaining local and global features at different scales, and constructing a multi-scale feature set; S23, performing correlation analysis on different modal features in the multi-scale feature set through a cross-modal feature association algorithm, establishing feature mapping relationships between modalities, and generating a feature association matrix; S24, inputting the feature association matrix into the multimodal health feature fusion inference model, performing deep fusion of multimodal features through model computation, and outputting a high-dimensional fused health feature vector.

[0033] Specifically, step S2 involves the precise extraction and fusion inference of multimodal health features. S21 first analyzes the feature dimensions of physiological indicator detection data, behavioral pattern recording data, environmental exposure-related data, and genetic characteristic-related data in the multimodal health dataset. A correlation analysis algorithm is used to calculate the correlation coefficient between each feature and physical condition. A correlation coefficient threshold of 0.6 is set, and core features with correlation coefficients higher than this threshold are selected. A core feature index table including feature name, data source, and dimensional information is established to ensure that subsequent processing focuses on high-value features. S22 employs a multi-scale feature extraction method, dividing the data into three scales: 16×16, 32×32, and 64×64. Local and global features at each scale are extracted through convolution and pooling operations, constructing a multi-scale feature set including features at different scales. Each feature is accompanied by a scale identifier and dimensional parameters. 23. Using a cross-modal feature association algorithm, the mutual information entropy and Pearson correlation coefficient between different modal features are calculated. The correlation coefficient threshold is set to 0.5, and a feature mapping relationship between modalities that meets the threshold requirement is established, generating a feature association matrix with a dimension of 512×512. The matrix elements represent the association strength between corresponding features. S24. The feature association matrix is ​​input into a multimodal health feature fusion inference model. The model adopts a 32-layer deep residual network structure with the Mish function as the activation function. A dynamic weight allocation algorithm is used to assign weights of 0.1 to 0.9 to different modal features. After multiple rounds of iterative operation, the deep fusion of multimodal features is achieved, and the final output is a high-dimensional fused health feature vector with a dimension of 1024. This step ensures the comprehensiveness of feature extraction and the accuracy of fusion, providing high-quality input for subsequent physical fitness assessment.

[0034] Preferred, such as Figure 3 As shown, step S3 includes the following sub-steps: S31, dimensional normalization of the high-dimensional fused health feature vector, extraction of feature subsets associated with physical fitness assessment, and determination of physical fitness assessment index system; S32, based on the assessment index system, construction of time-series feature sequence, mining of dynamic change patterns of physical fitness through time-series data analysis methods, and generation of time-series change features; S33, inputting the feature subset and time-series change features into a multi-dimensional physical fitness dynamic assessment model, performing quantitative assessment of physical fitness through multi-dimensional calculations of the model, and obtaining preliminary assessment results; S34, consistency verification of the preliminary assessment results, correction of assessment deviations through multi-index cross-validation methods, and output of the final physical fitness assessment results.

[0035] Specifically, step S3 performs a dynamic quantitative assessment of multidimensional physical condition. S31 performs dimensional regularization on the 1024-dimensional high-dimensional fused health feature vector output from step S2, and uses principal component analysis to extract principal components with a cumulative variance contribution rate reaching 90%, forming a key feature subset including 64 principal components. Combining traditional Chinese medicine constitution classification standards and modern medical health assessment indicators, a physical condition assessment index system is determined, comprising 36 indicators across six categories: physical function, metabolic level, immune status, organ function, psychological state, and adaptability. Each indicator has a clearly defined assessment dimension and quantitative standard. S32, based on this assessment index system, collects high-dimensional fused health feature vectors from the past 30 days at hourly intervals, constructing a time-series feature sequence. Using a sliding window method (window size set to 24) and a linear trend fitting algorithm, the dynamic change patterns of physical condition are mined, generating indicators including the rate of change and the slope of the trend. The model analyzes the temporal variation characteristics of information such as the frequency of extreme values. In step S33, the key feature subset and temporal variation characteristics are input into a multidimensional physical fitness dynamic assessment model. The model uses a variant of a long short-term memory network with 256 hidden layer neurons. Through multi-dimensional index correlation calculations and temporal trend extrapolation, 36 assessment indicators are quantitatively scored from 0 to 100 points, and the scores are summed to obtain preliminary assessment results. In step S34, the preliminary assessment results are validated for consistency. A multi-index cross-validation method is used to calculate the consistency coefficient of results under different assessment dimensions. A consistency coefficient threshold of 0.85 is set. If the result does not reach the threshold, the model parameters are adjusted and recalculated until a final physical fitness assessment result with a consistency coefficient higher than 0.85 is output. This step, through feature processing, temporal analysis, model calculation, and result validation, ensures the accuracy and dynamism of the physical fitness assessment.

[0036] Preferred, such as Figure 4 As shown, S4 includes the following sub-steps: S41, collecting user health need preference data, including intervention method preference, intervention intensity preference, and intervention cycle preference data, and establishing a preference data archive; S42, performing intervention need analysis on the physical condition assessment results, determining the health dimensions that need to be focused on intervention under the user's current physical condition, and generating an intervention need list; S43, based on the intervention need list and preference data archive, screening suitable intervention measure types, and constructing an intervention measure candidate set; S44, inputting the intervention measure candidate set into a personalized intervention strategy generation inference model, and combining the model's inference calculations to generate a preliminary intervention strategy set that meets the user's needs.

[0037] Specifically, step S4 generates a personalized preliminary intervention strategy set. S31 collects user health need preference data, gathering user preferences for intervention methods such as exercise, diet, lifestyle, and medication through the terminal interface, setting preferences for intervention intensity (mild, moderate, severe) and intervention period (short-term 1-3 months, medium-term 3-6 months, long-term 6-12 months), and retrieving intervention feedback information from the user's historical health management data to establish a preference data archive including 12 preference indicators, with the archive data updated in real time. S32 performs intervention needs analysis on the physical condition assessment results output in step S3, comparing the differences between each assessment indicator and health standard thresholds, ranking them by the degree of difference to determine the health dimensions requiring key intervention, and generating an intervention needs list including intervention priority, intervention target value, and intervention period requirements. Each need in the list corresponds to a specific assessment indicator. S33, based on the intervention needs list… The system uses a single preference data archive to select suitable intervention types from an intervention database containing over 200 intervention measures. These interventions are weighted and scored based on their relevance to the intervention effect (weight 0.6) and their preference matching degree (weight 0.4). The top 20 measures with the highest scores are selected to construct a candidate set of intervention measures. Each measure in the set is labeled with its applicable health dimension and preference fit. In step S34, the candidate set of intervention measures is input into a personalized intervention strategy generation model. The model combines physical condition assessment results and health risk prediction information, and assigns an adaptation weight of 0.2 to 0.8 to each candidate measure through multi-dimensional adaptability calculations. The top 5-8 measures are selected by weight to form a preliminary intervention strategy set. Each strategy clarifies the execution logic and relationships of each measure. This step, through preference collection, needs analysis, measure selection, and strategy generation, ensures the personalization and adaptability of the intervention strategy.

[0038] Preferred, such as Figure 5 As shown, S5 includes the following sub-steps: S51, extracting risk factors from the physical condition assessment results, identifying potential health risk factors, and establishing a risk factor database; S52, inputting the risk factors and the set of preliminary intervention strategies into the health risk level AI quantitative analysis platform, and using the platform's risk quantification algorithm to quantify each risk factor to obtain a risk quantification value; S53, based on the risk quantification value, classifying health risks according to preset risk level classification rules, and determining the risk level of each risk; S54, combining the risk level and the set of preliminary intervention strategies, adjusting and optimizing the preliminary intervention strategies using a strategy optimization algorithm, and outputting intervention strategy optimization suggestions and the final health risk quantification result.

[0039] Specifically, step S5 involves quantifying health risk levels and optimizing intervention strategies. S51 extracts risk factors from the physical condition assessment results output in step S3. Based on a disease risk association database including 48 common health risks, it matches the potential health risks corresponding to each abnormal assessment indicator, identifying key information such as risk type, risk triggers, scope of impact, and development rate, and establishing a risk factor database. The database is stored and sorted by risk occurrence probability. S52 inputs the key risk factors and the preliminary intervention strategy set generated in step S4 into the health risk level AI quantitative analysis platform. The platform uses a hybrid architecture combining deep learning and a rule engine. Through a risk quantification algorithm, it quantifies each risk factor from 0 to 100 points based on three dimensions: risk occurrence probability (weight 0.3), impact degree (weight 0.4), and development speed (weight 0.3), obtaining a risk quantification value. S53, based on the risk quantification value, classifies it into five risk levels according to preset rules, with 0-20 points being extremely low risk. Risk is categorized into four risk levels: 21-40 (low risk), 41-60 (medium risk), 61-80 (high risk), and 81-100 (extremely high risk). Each risk level is then assigned a specific label. S54 combines the risk level with the initial intervention strategy set, using a strategy optimization algorithm to adjust the initial intervention strategies. For high-risk and extremely high-risk items, the intensity of intervention measures is increased by 30%, the execution frequency is increased by 50%, and 2-3 additional targeted intervention measures are added. For low-risk and extremely low-risk items, the intervention process is simplified, core measures are retained, and the synergy and exclusion coefficients between intervention measures are calculated to optimize the combination of measures. Combinations with a synergy coefficient below 0.3 are eliminated. The final output includes a health risk quantification result including risk level, quantitative score, risk description, and optimized intervention strategy recommendations. This step, through risk extraction, quantification, classification, and strategy optimization, enhances the scientific validity and feasibility of the intervention strategy.

[0040] like Figure 6As shown, an AI-based comprehensive health status evaluation and intervention decision-making system is applied to an AI-based comprehensive health status evaluation and intervention decision-making method. It includes: a multimodal health data collaborative acquisition and preprocessing unit, a high-dimensional fusion health feature intelligent extraction and inference unit, a multi-dimensional physical condition dynamic quantitative assessment unit, a personalized intervention strategy intelligent generation and deduction unit, a health risk level AI quantitative analysis unit, and a comprehensive health status evaluation and intervention decision output unit. The multimodal health data collaborative acquisition and preprocessing unit and the high-dimensional fusion health feature intelligent extraction and inference unit are connected via a high-speed data transmission link for real-time transmission and feature extraction of multi-source health data. The high-dimensional fusion health feature intelligent extraction and inference unit and the multi-dimensional physical condition dynamic quantitative assessment unit are connected via a feature vector transmission interface to transmit high-dimensional data. The system integrates health feature vectors and transmits them to a multidimensional physical condition dynamic quantitative assessment unit for physical condition evaluation. This unit connects to a personalized intervention strategy intelligent generation and deduction unit and a health risk level AI quantitative analysis unit, simultaneously outputting physical condition assessment results. The personalized intervention strategy intelligent generation and deduction unit connects to the health risk level AI quantitative analysis unit via a strategy feedback link, enabling interactive optimization of intervention strategies and risk quantification results. The health risk level AI quantitative analysis unit connects to a comprehensive health status evaluation and intervention decision output unit, transmitting health risk quantification results and intervention strategy optimization suggestions to this unit. The comprehensive health status evaluation and intervention decision output unit integrates all input information to generate and output a comprehensive health status evaluation report and a personalized intervention implementation plan.

[0041] This AI-based comprehensive health status assessment and intervention decision-making method and system constructs a multi-source data collaborative collection system, integrates various health-related data types, ensures the comprehensiveness and real-time nature of data collection through innovative hardware configuration, and achieves deep fusion of multimodal features using a dedicated model to fully explore the correlation value of data from different sources. It establishes a dynamic physical fitness assessment and risk quantification mechanism, combining temporal change patterns to achieve accurate assessment of physical fitness status and hierarchical classification of health risks. Simultaneously, it generates personalized intervention strategies through a dedicated inference model, achieving deep linkage between assessment, risk analysis, and intervention recommendations. Through the coordinated linkage of six functional units and the detailed design of each step, a complete technical chain is formed from data collection to decision output, ensuring the intelligence and precision of the entire process.

[0042] This method and system address the problem of insufficient multimodal data fusion by achieving deep correlation and efficient fusion of different types of health data through multi-dimensional feature weight allocation, cross-modal information complementary reasoning, and a dedicated fusion model. This significantly improves the comprehensiveness and accuracy of feature extraction, making the assessment results more consistent with the actual health status. Addressing the lack of personalized adaptability, it deeply integrates dynamic physical fitness assessment results, quantitative data on health risk levels, and user needs and preferences. Through intervention strategy generation and deduction models, it achieves precise adaptation of intervention measures. Simultaneously, by utilizing the interactive mechanism of risk analysis and strategy optimization, it continuously optimizes intervention plans, ensuring the pertinence and feasibility of intervention recommendations and effectively meeting the personalized health management needs of different users.

[0043] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "set," "install," "connect," "link," and "fix" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal communication between two components. Those skilled in the art will understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0044] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various equivalent changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An AI-based comprehensive health status assessment and intervention decision-making method, characterized in that, Includes the following steps: S1. Collect multi-source health monitoring data, including physiological indicator detection data, behavioral pattern recording data, environmental exposure correlation data, and genetic characteristic correlation data, to construct a multimodal health dataset. S2. Call the multimodal health feature fusion inference model to extract features and perform correlation inference on the multimodal health dataset, generating a high-dimensional fused health feature vector. The fusion process is achieved through multi-dimensional feature weight allocation and cross-modal information complementary inference. S3. Input the high-dimensional fused health feature vector into a multi-dimensional physical condition dynamic assessment model. Through multi-dimensional physical condition indicator correlation analysis and temporal trend deduction, output the physical condition assessment result. S4. Based on the physical condition assessment result, use personalized intervention strategies to generate a deduction model. Combine health demand preference data and intervention measure suitability analysis to generate a preliminary intervention strategy set. S5. Input the physical condition assessment result and the preliminary intervention strategy set into a health risk level AI quantitative analysis platform. Through multi-dimensional risk factor correlation calculation and level division rule matching, output the health risk quantification result and intervention strategy optimization suggestions. S6 integrates the quantitative results of health risks and the optimization suggestions for intervention strategies through the comprehensive health status evaluation and intervention decision-making module to generate the final comprehensive health status evaluation report and personalized intervention implementation plan.

2. The AI-based comprehensive health status evaluation and intervention decision-making method according to claim 1, characterized in that, The expression for the multimodal health feature fusion inference model is: ,in To create a high-dimensional fusion health feature vector, Let i be the feature weights of the i-th type of physiological indicator detection data. Let i be the feature mapping function for the i-th type of physiological indicator detection data. Let be the original feature matrix of the i-th type of physiological indicator detection data. Let i be the feature offset of the i-th type of physiological indicator detection data. It is a non-linear activation function. Record the feature weights of the data for the j-th type of behavioral pattern. The feature mapping function for recording data of the j-th type of behavioral pattern. Here, n represents the original feature matrix of the data recording the j-th type of behavioral pattern, and n is the number of categories of physiological indicator detection data. The number of categories for recording data on behavioral patterns; The calculated confidence expression for feature fusion is as follows: ,in For feature fusion confidence, The maximum value of the high-dimensional fused health feature vector. Let k be the confidence correction function for the k-th type of environmental exposure association data. Let be the feature parameter of the k-th type of environmental exposure associated data, and p be the number of categories of environmental exposure associated data.

3. The AI-based comprehensive health status evaluation and intervention decision-making method according to claim 1, characterized in that, The expression for the multidimensional physical condition dynamic assessment model is: ,in This is the result of the physical fitness assessment. Let q be the evaluation weight of the qth physical fitness indicator. Let be the time-series evaluation function for the q-th physical fitness indicator. This represents a high-dimensional fused health feature vector, where t is the current evaluation time point. Associating weights with historical states This is a function relating physical condition to historical correlation. This refers to the previous assessment time point. This represents the number of physical fitness indicators. The calculated expression for the trend of physical condition change is as follows: ,in This represents the trend value of changes in physical condition. This represents the physical condition assessment results at the current point in time. This is the result of the physical condition assessment at the previous time point. Let r be the weight of the change in the physical fitness influencing factor. Let r be the correlation function for the change of the r-th physical fitness influencing factor. This represents the temporal variation of the high-dimensional fused health feature vector. For the evaluation time interval, s is the number of items in the physical fitness influencing factors.

4. The AI-based comprehensive health status evaluation and intervention decision-making method according to claim 1, characterized in that, The expression for the personalized intervention strategy generation and deduction model is: ,in, This is a preliminary set of intervention strategies. The basic adaptation weights for the u-th intervention are: Let be the fit function between the u-th intervention and the physical condition. This is the result of the physical fitness assessment. Let be the execution parameter matrix for the u-th intervention. To assign risk-adapted weights to the u-th intervention, Let u be the fit function between the u-th intervention and the health risk. To quantify health risks, Let be the risk control parameter matrix for the u-th intervention, and k be the total number of interventions.

5. The AI-based comprehensive health status evaluation and intervention decision-making method according to claim 1, characterized in that, The multi-source health monitoring data acquisition is achieved collaboratively through a flexible wearable biosensor array, a portable multi-parameter detection terminal, and a cloud-based distributed data acquisition node. The flexible wearable biosensor array integrates a photoelectric volume pulse detection unit, a skin impedance detection unit, and a motion posture sensing unit, with a sampling frequency set to 500Hz to 800Hz and data transmission using a millimeter-wave communication protocol. The portable multi-parameter detection terminal has a built-in microfluidic chip detection module and a quantum dot sensing module, supporting the simultaneous detection of more than 15 physiological indicators. The cloud-based distributed data acquisition nodes adopt an edge computing architecture, deploy lightweight data filtering algorithms, and have a data cache capacity of no less than 1TB. The feature mapping function of the multimodal health feature fusion inference model adopts a deep residual network structure with 32 network layers and the activation function is the Mish function. The temporal evaluation function of the multidimensional physical condition dynamic assessment model adopts a variant of the long short-term memory network with 256 hidden layer neurons.

6. The AI-based comprehensive health status evaluation and intervention decision-making method according to claim 1, characterized in that, S2 includes the following sub-steps: S21, performing feature dimension analysis on various types of data in the multimodal health data set, filtering out features strongly correlated with physical condition through feature correlation analysis, and establishing a feature index table; S22, a multi-scale feature extraction method is used to extract features hierarchically, obtaining local and global features at different scales, and constructing a multi-scale feature set; S23, a cross-modal feature association algorithm is used to perform association analysis on different modal features in the multi-scale feature set, establish feature mapping relationships between modalities, and generate a feature association matrix; S24, the feature association matrix is ​​input into the multi-modal health feature fusion inference model, and deep fusion of multi-modal features is performed through model operation to output a high-dimensional fused health feature vector.

7. The AI-based comprehensive health status evaluation and intervention decision-making method according to claim 1, characterized in that, S3 includes the following sub-steps: S31, dimensional normalization of the high-dimensional fused health feature vector, extraction of feature subsets associated with physical fitness assessment, and determination of physical fitness assessment index system; S32, based on the assessment index system, construction of time-series feature sequence, mining of dynamic change patterns of physical fitness through time-series data analysis methods, and generation of time-series change features; S33, inputting the feature subset and time-series change features into a multi-dimensional physical fitness dynamic assessment model, performing quantitative assessment of physical fitness through multi-dimensional calculations of the model, and obtaining preliminary assessment results; S34, consistency verification of the preliminary assessment results, correction of assessment deviations through multi-index cross-validation methods, and output of the final physical fitness assessment results.

8. The AI-based comprehensive health status evaluation and intervention decision-making method according to claim 1, characterized in that, S4 includes the following steps: S41, collecting user health need preference data, including intervention method preference, intervention intensity preference and intervention cycle preference data, and establishing a preference data archive; S42, perform intervention needs analysis on the physical condition assessment results, determine the key health dimensions that need to be intervened in the user's current physical condition, and generate an intervention needs list; S43, based on the intervention needs list and preference data file, filter suitable intervention types and construct a candidate set of intervention measures; S44: Input the candidate set of intervention measures into the personalized intervention strategy generation model, and combine the model's inference calculations to generate a preliminary set of intervention strategies that meet the user's needs.

9. The AI-based comprehensive health status evaluation and intervention decision-making method according to claim 1, characterized in that, S5 includes the following steps: S51, extracting risk factors from the physical condition assessment results, identifying potential health risk factors, and establishing a risk factor database. S52, input the set of risk factors and preliminary intervention strategies into the health risk level AI quantitative analysis platform, and use the platform's risk quantification algorithm to quantify and calculate each risk factor to obtain the risk quantification value; S53, based on the risk quantification value, classify health risks according to the preset risk level classification rules, and determine the risk level of each risk; S54 combines risk levels with a set of initial intervention strategies, and uses a strategy optimization algorithm to adjust and optimize the initial intervention strategies, outputting intervention strategy optimization suggestions and the final health risk quantification results.

10. An AI-based comprehensive health status assessment and intervention decision-making system, characterized in that, This system is applied to the AI-based comprehensive health status evaluation and intervention decision-making method described in claim 1, comprising: a multimodal health data collaborative acquisition and preprocessing unit, a high-dimensional fusion health feature intelligent extraction and inference unit, a multi-dimensional physical condition dynamic quantitative assessment unit, a personalized intervention strategy intelligent generation and deduction unit, a health risk level AI quantitative analysis unit, and a comprehensive health status evaluation and intervention decision output unit; the multimodal health data collaborative acquisition and preprocessing unit and the high-dimensional fusion health feature intelligent extraction and inference unit are connected via a high-speed data transmission link to perform real-time transmission and feature extraction of multi-source health data; the high-dimensional fusion health feature intelligent extraction and inference unit and the multi-dimensional physical condition dynamic quantitative assessment unit are connected via a feature vector transmission interface to transmit high-dimensional fusion health feature vectors. The data is transmitted to a multidimensional dynamic quantitative assessment unit for physical fitness evaluation. This unit connects to a personalized intervention strategy intelligent generation and deduction unit and a health risk level AI quantitative analysis unit, simultaneously outputting the physical fitness assessment results. The personalized intervention strategy intelligent generation and deduction unit connects to the health risk level AI quantitative analysis unit via a strategy feedback link, enabling interactive optimization of intervention strategies and risk quantification results. The health risk level AI quantitative analysis unit connects to a comprehensive health status evaluation and intervention decision output unit, transmitting health risk quantification results and intervention strategy optimization suggestions to this unit. The comprehensive health status evaluation and intervention decision output unit integrates all input information, generating and outputting a comprehensive health status evaluation report and a personalized intervention implementation plan.

Images