Personalized diet and exercise health management system fused with large model analysis capability

Abstract

The invention relates to the technical field of health management, in particular to a personalized diet and exercise health management system fusing large model analysis ability, which comprises a multi-source data acquisition module, an integrated intelligent wearable equipment interface and the like, and is used for acquiring multi-dimensional health data; the health portrait modeling module fuses multi-modal data based on an improved Transform architecture, and outputs a dynamic portrait containing risk early warning and trend prediction; the personalized scheme generation module integrates the medical knowledge graph and the user gene features to generate a personalized health scheme; the scenarized recommendation engine matches diet and exercise resources in combination with a real-time scene; the closed-loop supervision optimization module implements scheme execution evaluation and adaptive adjustment through federated learning; and the risk early warning subsystem constructs an acute and chronic disease prediction model to realize risk early warning. According to the invention, precise health management is realized, the scheme effectiveness and the user experience are improved, the data security is guaranteed, and the intelligent development of health management is promoted.

Description

Technical Field

[0001] The present invention relates to the field of health management technology, and specifically to a personalized diet and exercise health management system integrating large model analysis capabilities. Background Art

[0002] In today's society, people are paying more and more attention to their own health management, and the pursuit of scientific and personalized health management methods has become a mainstream trend; however, traditional health management models and technologies have many limitations and are difficult to meet people's growing health needs, which are specifically reflected in the following aspects:

[0003] Limitations of data collection and analysis: The data sources of previous health management systems were relatively single, mostly relying on physiological indicator data manually entered by users or collected by a small number of wearable devices, which could not fully reflect the health status of users. At the same time, there was a lack of effective integration and in-depth analysis methods for the collected data, making it difficult to mine the potential health information behind the data, resulting in inaccurate and incomplete health assessments, and unable to provide strong support for personalized health management.

[0004] Insufficient personalized program formulation: Traditional health management programs are often based on universal health standards and templates, ignoring the differences in genetic characteristics, health conditions, living habits, etc. of individual users. This "one-size-fits-all" program formulation method makes health management measures lack of specificity, making it difficult to achieve the desired health improvement effect, and user compliance is also low;

[0005] Lack of scenario adaptation capabilities: When recommending diet and exercise resources, existing health management systems rarely consider the user's real-time scenario, such as geographic location, time, environmental conditions and other factors; this causes the recommendation results to be out of touch with the user's actual needs, and it is impossible to provide appropriate health management advice and resources at the time and place where the user needs them, reducing the practicality and operability of the health management plan.

[0006] Therefore, to address the above issues, a personalized diet and exercise health management system that integrates large model analysis capabilities is proposed. Summary of the invention

[0007] The purpose of the present invention is to provide a personalized diet and exercise health management system that integrates large model analysis capabilities to solve the problems raised in the above background technology.

[0008] To achieve the above object, the present invention provides the following technical solutions:

[0009] A personalized diet and exercise health management system that integrates large model analysis capabilities, including:

[0010] Multi-source data acquisition module, integrating intelligent wearable device interfaces, medical data APIs, and mobile terminal sensors, to collect users' multi-dimensional health data in real time. The users' multi-dimensional health data includes:

[0011] Physiological index data: Heart rate variability (HRV), ambulatory blood pressure, blood oxygen saturation (SpO2), body fat percentage;

[0012] Behavioral trajectory data: GPS positioning trajectory, motion accelerometer data, screen usage time log;

[0013] Environment-related data: Temperature and humidity obtained from the meteorological API, air quality index (AQI), ultraviolet intensity;

[0014] Medical history data: Structured fields of electronic medical records, medication contraindication list, allergen database;

[0015] Health portrait modeling module, performing multi-modal data fusion based on an improved Transformer architecture, and outputting a dynamic health portrait including short-term risk warnings and long-term trend predictions;

[0016] Personalized solution generation module, integrating a medical knowledge graph and the user's genetic characteristics, to generate an executable solution including nutritional ratio, calorie distribution, and exercise intensity threshold;

[0017] Scene-based recommendation engine, combining real-time geographical location and the user's behavior pattern to dynamically match diet and exercise scene resources;

[0018] Closed-loop supervision and optimization module, realizing the evaluation of the solution execution effect and adaptive adjustment through the federated learning framework, and forming a closed-loop of health management.

[0019] As a preferred solution, the multi-source data acquisition module preprocesses heterogeneous data using a spatio-temporal feature alignment algorithm. The specific steps are as follows:

[0020] Divided into three time periods: morning, daytime, and night according to the data generation time period;

[0021] Encode the geographical location into a grid coordinate system with a preset precision;

[0022] Calculate the mean and standard deviation of the data within each spatio-temporal partition;

[0023] Assign entropy weight fusion weight values according to the data source type;

[0024] Perform spatio-temporal partition-based standardization processing on the original data and multiply by the corresponding source weight value.

[0025] As a preferred solution, the health portrait modeling module adopts a phased training strategy:

[0026] The contrastive learning objective function is adopted in the pre-training stage:

[0027] Calculate the cosine similarity of the data representation vectors of the same user at different times;

[0028] Construct a loss function to maximize the proportion of the exponential similarity of similar samples;

[0029] Set the temperature hyperparameter to control the sample discrimination;

[0030] In the fine-tuning stage, a multi-task loss function is adopted:

[0031] Combine the cross-entropy loss of acute health risk classification;

[0032] The mean squared error loss for chronic disease development prediction;

[0033] The data reconstruction autoencoder loss;

[0034] Dynamically adjust the weighting coefficients of each loss term.

[0035] As a preferred solution, the personalized solution generation module includes:

[0036] A nutrition optimization sub-module that uses an improved linear programming model:

[0037] The objective function is to minimize the sum of the weighted absolute deviations of the food ingredient cost and the metabolic adaptation coefficient;

[0038] The constraint conditions include the minimum intake of essential nutrients and the maximum intake of restricted nutrients;

[0039] The decision variable is a set of food ingredients with binary choices;

[0040] A motion planning sub-module that uses a genetic algorithm for parameter optimization:

[0041] The fitness function is inversely proportional to the deviation from the median of the target heart rate range;

[0042] Introduce the estimated value of the metabolic equivalent cost and the fatigue adjustment factor;

[0043] Retain the optimal motion parameter combination through iterative selection.

[0044] As a preferred solution, the scenario-based recommendation engine includes:

[0045] A diet recommendation algorithm based on an improved collaborative filtering model:

[0046] Comprehensively sum the weighted historical rating deviations of similar users;

[0047] Overlay the geographical location attenuation function and the price logarithmic suppression factor;

[0048] Dynamically adjust the weight distribution of the user's similar neighbor set;

[0049] Sports venue matching algorithm, based on spatio-temporal reachability analysis:

[0050] Calculate the proportion of the intersection of the user's free time period and the venue opening period;

[0051] Introduce an exponential decay factor based on real-time traffic time;

[0052] The comprehensive evaluation result is the venue availability score.

[0053] As an optimal solution, the closed-loop supervision optimization module realizes:

[0054] Dietary compliance monitoring, meal recognition through an improved residual network:

[0055] Calculate the relative deviation between the recognized nutrient content and the planned value;

[0056] Introduce a zero-elimination constant to handle the case of micronutrients;

[0057] Dynamically adjust the deviation weight according to the user's health risk profile;

[0058] Exercise effect evaluation, using the dynamic time warping algorithm:

[0059] Calculate the minimum alignment distance between the user's action sequence and the standard sequence;

[0060] Normalize according to the maximum physiological range of joint movement;

[0061] Generate an action standardization score.

[0062] As an optimal solution, it also includes a risk warning subsystem, and the risk warning subsystem includes:

[0063] Acute risk prediction model:

[0064] Weighted sum of the coefficients of each frequency band of heart rate variability;

[0065] Superimpose the sensitivity coefficient of the blood oxygen decline rate;

[0066] Output the acute risk probability value through the sigmoid function;

[0067] Chronic disease development scoring model:

[0068] Weighted sum based on the weekly change rate of key physiological indicators;

[0069] Introduce a historical score decay factor to achieve time series fusion;

[0070] Output a dynamically updated chronic disease development index.

[0071] As can be seen from the technical solutions provided by the present invention above, the personalized diet and exercise health management system integrating the large model analysis ability provided by the present invention has the following beneficial effects:

[0072] Comprehensive and accurate data collection and analysis: The multi-source data collection module integrates a variety of data interfaces and sensors, which can collect multi-dimensional health data such as physiological indicators, behavior trajectories, environmental correlations, and medical histories in real time, and preprocesses heterogeneous data through a spatio-temporal feature alignment algorithm to ensure the accuracy, real-time nature, and integrity of the data; On this basis, the health portrait modeling module, based on the improved Transformer architecture, deeply fuses and analyzes multi-modal data, and outputs a dynamic health portrait including short-term risk warnings and long-term trend predictions, providing a solid data foundation and analysis support for precise health management, making health assessment more comprehensive and accurate;

[0073] Highly personalized health management solutions: The personalized solution generation module integrates a medical knowledge graph and user gene characteristics, and uses an improved linear programming model and genetic algorithm for diet and exercise respectively to generate personalized solutions including key elements such as nutritional ratios, calorie distribution, and exercise intensity thresholds; This solution fully considers user individual differences, such as gene characteristics, health status, dietary preferences, and exercise ability, and can better meet the diverse health management needs of users compared with traditional general-purpose solutions, improving the pertinence and effectiveness of health management;

[0074] Intelligent recommendation services tailored to actual scenarios: The scenario-based recommendation engine combines scenario information such as the user's real-time geographical location, time, and behavior patterns, and uses an improved collaborative filtering model and spatio-temporal reachability analysis algorithm to dynamically match diet and exercise scenario resources; Whether it is recommending nearby restaurants that meet nutritional requirements or matching suitable exercise venues according to the user's free time and exercise goals, it can make the health management solution better integrate into the user's daily life, improve the user experience and the execution rate of the solution, and make health management more convenient and practical;

[0075] Closed-loop continuous optimization management: The closed-loop supervision and optimization module constructs a closed-loop management system covering diet compliance monitoring, exercise effect evaluation, and program adaptive adjustment through a federated learning framework; Using technologies such as improved residual networks and dynamic time warping algorithms, it monitors the user's implementation of personalized solutions in real time and makes dynamic optimization adjustments according to the evaluation results; This closed-loop management mode ensures that the health management solution always fits the actual situation of the user, forms a full-process closed-loop control from solution formulation, implementation to optimization, continuously guarantees the health management effect, and helps users better achieve health goals;

[0076] Effective health risk prevention mechanism: The risk early warning subsystem constructs two core modules, namely acute risk prediction and chronic disease development scoring. By using machine learning algorithms and mathematical models, it deeply analyzes the user's health data. It can monitor in real time and identify acute health risks and chronic disease development trends in advance, send early warning information to users and relevant parties in a timely manner, and provide personalized health intervention suggestions. This helps users take preventive measures before the occurrence of diseases, reduce the harm of health risks, and also provides a reference for medical institutions to reasonably allocate medical resources, achieving early detection and early intervention of diseases. Brief Description of the Drawings

[0077] Figure 1 It is a schematic diagram of the overall structure of the personalized diet and exercise health management system integrating the large model analysis ability of the present invention. Detailed Embodiment

[0078] In order to make the objectives, technical solutions and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below in conjunction with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0079] In order to better understand the above technical solutions, the above technical solutions will be described in detail below in conjunction with the specification drawings and specific embodiments.

[0080] As Figure 1 shown, the embodiment of the present invention provides a personalized diet and exercise health management system integrating the large model analysis ability, including:

[0081] A multi-source data acquisition module, which integrates intelligent wearable device interfaces, medical data APIs, and mobile terminal sensors to collect the user's multi-dimensional health data in real time. The user's multi-dimensional health data includes:

[0082] Physiological index data: heart rate variability HRV, ambulatory blood pressure, blood oxygen saturation SpO2, body fat percentage;

[0083] Behavioral trajectory data: GPS positioning trajectory, motion accelerometer data, screen usage time log;

[0084] Environment-related data: temperature and humidity, air quality index AQI, ultraviolet intensity obtained from the meteorological API;

[0085] Medical history data: structured fields of electronic medical records, medication contraindication list, allergen database;

[0086] A health portrait modeling module, which performs multi-modal data fusion based on an improved Transformer architecture and outputs a dynamic health portrait including short-term risk early warning and long-term trend prediction;

[0087] Personalized solution generation module, integrating medical knowledge graph and user gene characteristics, generates an executable solution including nutritional ratio, calorie distribution, and exercise intensity threshold;

[0088] Scene-based recommendation engine, combining real-time geographical location and user behavior patterns, dynamically matches diet and exercise scene resources;

[0089] Closed-loop supervision optimization module, through the federated learning framework, realizes the evaluation of the solution execution effect and adaptive adjustment, forming a health management closed loop.

[0090] In this embodiment, the multi-source data acquisition module is the basic component for obtaining users' multi-dimensional health data in the personalized diet and exercise health management system integrating large model analysis capabilities; in the field of health management, users' health conditions are affected by various factors such as physiology, behavior, and environment, and a single data source is difficult to comprehensively reflect users' health information; this module aims to integrate different types of data sources, collect various health-related data in real time and accurately, and provide rich and reliable data support for subsequent functions such as health portrait modeling and personalized solution generation. The multi-source data acquisition module preprocesses heterogeneous data using the spatio-temporal feature alignment algorithm, which is specifically reflected in:

[0091] I. Overall function overview:

[0092] This module constructs a collection environment integrating multiple data interfaces and sensors, and can collect users' multi-dimensional health data in real time from multiple channels such as smart wearable devices, medical data platforms, and mobile terminals; through the classified collection and preliminary processing of different types of data, it ensures the accuracy, real-time nature, and integrity of the data, provides basic data guarantee for the entire health management system, and enables the system to formulate personalized diet and exercise health management solutions for users based on comprehensive information;

[0093] II. Composition and functions of sub-modules:

[0094] (I) Data acquisition interface sub-module:

[0095] Smart wearable device interface: Responsible for establishing connections with various smart wearable devices (such as smart bracelets, smart watches, smart body fat scales, etc.), and obtaining users' physiological index data collected by the devices in real time through communication protocols such as Bluetooth and Wi-Fi; for example, obtaining data such as heart rate variability (HRV), dynamic blood pressure, and blood oxygen saturation (SpO2) from a smart bracelet, and obtaining body fat percentage data from a smart body fat scale to ensure the real-time nature and continuity of data acquisition;

[0096] Medical Data API: Connect to the electronic medical record systems and medical data platforms of institutions such as hospitals and physical examination centers, and obtain the user's medical history data through standardized API interfaces; including information such as structured fields in electronic medical records (such as disease diagnosis, treatment records, etc.), medication contraindication lists, and allergen databases, providing the system with the user's long-term health status and medical background data;

[0097] Mobile Terminal Sensors: Invoke various sensors built into mobile terminals (such as smartphones and tablets) to collect the user's behavioral trajectory data and environment-related data; use the GPS positioning module to obtain the user's geographical location trajectory, collect motion-related data through the motion accelerometer, and read the screen usage time log to understand the user's behavioral habits; at the same time, with the help of the meteorological API, combined with the positioning information of the mobile terminal, obtain environmental data such as temperature, humidity, air quality index (AQI), and ultraviolet intensity at the current location;

[0098] (2) Data Preprocessing and Verification Sub-module:

[0099] Data Format Unification: Since the data formats collected from different data sources are different, this sub-module performs format conversion and standardization processing on the collected data; for example, converting the binary data collected by smart wearable devices into a standard data format recognizable by the system, and standardizing the structured data obtained from the medical data API to ensure the consistency of all data in format, facilitating subsequent processing and analysis;

[0100] Data Quality Verification: Perform integrity, accuracy, and validity verification on the collected data; detect whether there are missing values, outliers, and incorrect data in the data by setting data range thresholds, logical verification rules, etc.; for example, for heart rate data, if a value outside the normal physiological range (such as less than 30 beats per minute or higher than 200 beats per minute) is detected, it is marked as abnormal data and corresponding processing is performed (such as requesting re-collection or data repair);

[0101] Timestamp Synchronization: To ensure the consistency of multi-source data in the time dimension, add accurate timestamps to all collected data and perform time synchronization processing; calibrate the time with the network time server to ensure the accuracy and synchronization of data from different data sources in time, providing a reliable time basis for subsequent spatio-temporal feature alignment and data analysis;

[0102] (3) Data Storage and Transmission Sub-module:

[0103] Local cache storage: During the data collection process, to prevent data loss and improve data processing efficiency, the collected data is first stored in the local cache; an efficient caching mechanism (such as an in-memory database) is adopted to ensure that data can be stored and retrieved quickly. At the same time, the cached data is regularly cleaned and managed to avoid data loss or system performance degradation caused by cache overflow;

[0104] Secure transmission channel: Establish a secure data transmission channel to transmit the preprocessed and verified data to the subsequent modules of the system; use an encrypted communication protocol (such as SSL / TLS) to encrypt the data during transmission to prevent the data from being stolen or tampered with during transmission. At the same time, monitor and manage the data transmission process to ensure that the data can be transmitted to the target module stably and efficiently;

[0105] Cloud backup storage: To ensure data security and traceability, the important collected data is synchronously backed up to the cloud storage server; adopt distributed storage technology to achieve redundant storage and disaster recovery backup of data, preventing data loss caused by local storage failures. At the same time, regularly maintain and manage the data stored in the cloud to ensure data integrity and availability;

[0106] III. Key technical principles:

[0107] (I) Principle of heterogeneous data collection technology:

[0108] According to the characteristics of different types of data sources such as smart wearable devices, medical data platforms, and mobile terminals, corresponding technical means are adopted to achieve data collection; for smart wearable devices, connect to the device and obtain data by following the communication protocol and development interface provided by the device manufacturer; for medical data APIs, make data requests and receive data according to standard interface specifications; for mobile terminal sensors, use the sensor access permissions and development framework provided by the operating system to achieve real-time collection of sensor data; in this way, it is possible to be compatible with various types of data sources and achieve unified collection of heterogeneous data;

[0109] (II) Principle of spatio-temporal feature alignment algorithm:

[0110] To effectively integrate multi-source heterogeneous data, a spatio-temporal feature alignment algorithm is used to preprocess the collected data; its core idea is to normalize and standardize the data from different sources according to the timestamp and geographical location information of the data, so that the data is comparable in the spatio-temporal dimension; the specific formula is: (where, is the data preprocessed by the spatio-temporal feature alignment algorithm: is the original data; represents the time period to which the data timestamp belongs (morning Daytime Night); Represents the grid code of the geographical location for data collection; Represents the data source type (medical device Mobile terminal Environmental sensor); Is the source weight value calculated by the entropy weight fusion algorithm, used to measure the importance of different data sources; Is for the corresponding spatio-temporal partition (time is Geographical location is The mean value of the data; Is for the corresponding spatio-temporal partition (time is Geographical location is The standard deviation of the data); Through this algorithm, the differences in the spatio-temporal distribution of the data can be eliminated, laying a foundation for subsequent multi-modal data fusion and analysis;

[0111] (III) Principle of data encrypted transmission:

[0112] During the data transmission process, an encrypted communication protocol (such as SSL / TLS) is used to encrypt and protect the data; The SSL / TLS protocol establishes a secure connection through a handshake process, negotiates the encryption algorithm and key; During the data transmission stage, the negotiated encryption algorithm (such as AES, RSA, etc.) is used to encrypt the data, converting the plaintext data into ciphertext data for transmission; After receiving the ciphertext data, the receiving end uses the corresponding key to decrypt it and restore the original data; In this way, the security of the data during the transmission process is ensured, preventing the data from being stolen or tampered with;

[0113] IV. Working process of the module:

[0114] (I) Initialization stage:

[0115] When the system starts, the multi-source data collection module performs initialization configuration, including setting parameters such as the frequency, range, and priority of data collection;

[0116] Establish connections with data sources such as smart wearable devices, medical data APIs, and mobile terminal sensors, and perform authentication and authorization of the devices or interfaces to ensure that data can be obtained normally;

[0117] Initialize the data preprocessing and verification rules, set the thresholds and logics for data quality verification, and configure the relevant parameters of the spatio-temporal feature alignment algorithm to prepare for data collection and processing;

[0118] (II) Data collection stage:

[0119] According to the set collection frequency, real-time collect the multi-dimensional health data of users from data sources such as smart wearable devices, medical data APIs, and mobile terminal sensors;

[0120] Perform preliminary processing on the collected data, such as adding timestamps and converting data formats, and store the data in the local cache;

[0121] Regularly check the connection status of the data source and the data collection situation. If a connection interruption or abnormal data collection is found, troubleshoot and recover in a timely manner to ensure the continuity of data collection;

[0122] (III) Data preprocessing and verification stage:

[0123] Read the collected data from the local cache, perform unified and standardized processing on the data format to make it meet the data requirements of the system;

[0124] Perform quality verification on the data to detect whether there are missing values, outliers, and incorrect data in the data; for abnormal data, repair, mark, or request re-collection according to the preset processing rules;

[0125] Use the spatio-temporal feature alignment algorithm to preprocess the data. According to the timestamp and geographical location information of the data, perform normalization and standardization processing on the data to make the data comparable in the spatio-temporal dimension;

[0126] (IV) Data storage and transmission stage:

[0127] Store the preprocessed and verified data in the local cache and synchronously back it up to the cloud storage server to ensure the security and traceability of the data;

[0128] Through a secure transmission channel, transmit the processed data to subsequent modules such as the health portrait modeling module of the system to provide data support for other functions of the health management system;

[0129] During the data transmission process, monitor and manage the transmission status to ensure that the data can be stably and efficiently transmitted to the target module; if a transmission failure occurs, retransmit in a timely manner or take other fault recovery measures;

[0130] (V) End stage:

[0131] When the system stops running or receives an instruction to stop collection, the multi-source data collection module stops data collection and transmission operations; close the connection with the data source, release relevant resources, and clean up and archive the logs and temporary data during the collection process for subsequent query and analysis.

[0132] In this embodiment, the health image modeling module is one of the core components of the personalized diet and exercise health management system that integrates the analysis capabilities of large models. It deeply processes and analyzes the multi-dimensional health data of users obtained by the multi-source data acquisition module to construct a dynamic model that can reflect the health status of users. In the health management scenario, the health status of users is complex and changeable, and data from a single dimension is difficult to comprehensively depict the health situation. This module aims to use advanced artificial intelligence technologies to achieve the effective integration and in-depth mining of multi-modal data, providing an accurate basis for the formulation of subsequent personalized health management plans. Specifically, it is embodied as follows:

[0133] I. Overall function overview:

[0134] Based on the improved Transformer architecture, this module constructs an efficient multi-modal data processing and analysis platform. It can deeply integrate multi-source heterogeneous data such as physiological indicators, behavior trajectories, and environmental correlations from different channels. Through a phased training strategy, it outputs a dynamic health image that includes short-term risk warnings and long-term trend predictions. This image can not only reflect the current health status of users in real time but also make forward-looking judgments on potential health risks, providing scientific and reliable support for personalized diet and exercise health management.

[0135] II. Composition and functions of sub-modules:

[0136] (I) Data preprocessing and feature extraction sub-module:

[0137] Multi-modal data fusion: Receives the multi-dimensional health data transmitted by the multi-source data acquisition module, and integrates different types of data such as physiological index data, behavior trajectory data, environmental correlation data, and medical history data. According to the characteristics of each type of data, specific fusion methods are adopted. For example, time-series physiological data and spatial information behavior trajectory data are associated through timestamps and geographical locations, enabling different modal data to complement and cooperate with each other.

[0138] Feature extraction: Uses technologies such as convolutional neural networks (CNNs) and recurrent neural networks (RNNs) in deep learning to extract features from the fused multi-modal data. For physiological index data, local features such as heart rate variability and blood pressure changes are extracted through CNNs. For behavior trajectory data, RNNs are used to capture temporal features such as movement patterns and activity rules. At the same time, combined with artificially designed domain knowledge features, such as physiological index threshold features based on medical knowledge, a comprehensive and representative feature set is constructed.

[0139] Data Dimensionality Reduction and Normalization: To reduce the data dimension and computational complexity, methods such as principal component analysis (PCA) and singular value decomposition (SVD) are used to perform dimensionality reduction on the extracted features; at the same time, data normalization operations are carried out to convert data with different scales and distributions into a unified range, improving the stability and efficiency of model training;

[0140] (2) Model Architecture Construction Sub-module:

[0141] Improved Transformer Architecture Design: Based on the Transformer architecture, improvements are made according to the data characteristics and task requirements in the field of health management; in the encoder part, the number of heads of the multi-head attention mechanism is increased to enhance the model's ability to capture complex relationships between different modality data; in the decoder part, a multi-output layer structure is designed, corresponding to short-term risk warning and long-term trend prediction tasks respectively; at the same time, residual connections and layer normalization techniques are introduced to alleviate the vanishing gradient problem and improve the training speed and stability of the model;

[0142] Parameter Initialization: Appropriate parameter initialization methods, such as Xavier initialization or Kaiming initialization, are used to initialize the parameters of the model; according to the characteristics and functions of different layers, reasonable initialization parameter ranges are set to ensure that the model can converge quickly in the initial stage of training and avoid falling into local optimal solutions;

[0143] Model Configuration and Optimization: According to the data scale and computing resources, configure the hyperparameters of the model, such as the hidden layer dimension, number of attention heads, learning rate, etc.; through methods such as cross-validation, optimize and adjust the hyperparameters to find the optimal model configuration and improve the performance and generalization ability of the model;

[0144] (3) Model Training and Optimization Sub-module:

[0145] Implementation of the Phased Training Strategy:

[0146] Pre-training Stage: The model is pre-trained using a contrastive learning objective function; by constructing positive and negative sample pairs of data of the same user at different times, the model learns the similarity and difference features of the data through contrastive learning; the specific formula is: (where, is the value of the contrastive learning objective function: , are the data representation vectors of the same user at different times: is the cosine similarity calculation function; is the temperature hyperparameter (set to 0.07); (where is the number of data in the dataset); by minimizing this objective function, the model can better capture the intrinsic characteristics of the data, laying a foundation for the subsequent fine-tuning stage;

[0147] Fine-tuning stage: The model is fine-tuned using a multi-task loss function to adapt to short-term risk warning and long-term trend prediction tasks; the formula for the multi-task loss function is: (where, is the total value of the multi-task loss function: is the cross-entropy loss for acute health risk classification: is the mean squared error loss for chronic disease development prediction: is the data reconstruction autoencoder loss; are dynamic adjustment coefficients (initial values are 0.6, 0.3, 0.1 respectively); by adjusting these coefficients, the weights between different tasks are balanced, enabling the model to achieve good performance on multiple tasks;

[0148] Application of optimization algorithms: Optimization algorithms such as Stochastic Gradient Descent (SGD) and Adaptive Moment Estimation (Adam) are selected to train the model; according to the training situation of the model and the characteristics of the data, the learning rate is dynamically adjusted, such as adopting a learning rate decay strategy, gradually decreasing the learning rate as the number of training rounds increases, avoiding overfitting of the model in the later stage of training, and improving the convergence speed and training effect of the model;

[0149] Model evaluation and improvement: During the training process, the model is regularly evaluated using the validation set, and performance metrics such as accuracy, recall rate, and mean squared error on short-term risk warning and long-term trend prediction tasks are calculated; based on the evaluation results, problems existing in the model are analyzed, and adjustments and improvements are made to the model architecture, parameter settings, or training strategies to continuously optimize the model performance;

[0150] (4) Health portrait generation and update sub-module:

[0151] Health portrait generation: The trained and optimized model outputs a dynamic health portrait containing short-term risk warning and long-term trend prediction based on the input multi-dimensional health data of the user; for short-term risk warning, the model analyzes the current physiological indicators and behavior data to determine whether the user has acute health risks and gives corresponding risk levels and warning information; for long-term trend prediction, the model predicts the future health development trend of the user based on historical data and the current state, such as the occurrence probability of chronic diseases and the change trend of physical functions;

[0152] Health Image Update: As the user's health data is continuously updated, this sub-module receives new data in real-time and inputs it into the model for processing; dynamically updates the health image based on the new data to promptly reflect changes in the user's health status, ensuring the timeliness and accuracy of the health image; meanwhile, records the update history of the health image for easy viewing and analysis by users and professionals;

[0153] III. Key Technical Principles:

[0154] (I) Principle of the Improved Transformer Architecture:

[0155] The Transformer architecture is based on the attention mechanism and can process input sequences in parallel, effectively capturing long-range dependencies in the data; in the health image modeling module, the improved Transformer architecture enhances the model's ability to model complex data relationships by increasing the number of heads in the multi-head attention mechanism, enabling the model to extract and fuse features from multi-modal data from multiple different perspectives; meanwhile, the design of the multi-output layer structure enables the model to handle multiple tasks such as short-term risk warning and long-term trend prediction simultaneously, improving the practicality and flexibility of the model;

[0156] (II) Principle of Contrastive Learning:

[0157] The core idea of contrastive learning is to maximize the similarity between similar samples and minimize the similarity between dissimilar samples, enabling the model to learn the intrinsic feature representations of the data; in the pre-training stage, by constructing contrastive samples of data from the same user at different times, the model learns the change patterns and feature differences of the user's health data in the time dimension, thereby improving the model's ability to represent the data and providing a better foundation for subsequent task learning;

[0158] (III) Principle of Multi-Task Learning:

[0159] Multi-task learning aims to improve the generalization ability and performance of the model by simultaneously learning multiple related tasks and leveraging the shared information and complementarity between tasks; in the fine-tuning stage, by designing a multi-task loss function, tasks such as short-term risk warning and long-term trend prediction are jointly trained; the shared features and knowledge between different tasks can promote each other, enabling the model to achieve better results on multiple tasks, while also reducing the training parameters and computational costs of the model;

[0160] IV. Workflow of the Module:

[0161] (I) Initialization Phase:

[0162] After the health image modeling module is started, the model architecture is initialized, the improved Transformer architecture is loaded, and configured according to the set parameters;

[0163] Initialize the data preprocessing and feature extraction sub-module, and set the relevant parameters and methods for data fusion, feature extraction, dimensionality reduction, and normalization;

[0164] Determine the phased strategy for model training, and set the objective function, optimization algorithm, hyperparameters, etc. for the pre-training phase and the fine-tuning phase;

[0165] (2) Data processing phase:

[0166] Receive the user's multi-dimensional health data transmitted by the multi-source data acquisition module, and perform multi-modal data fusion in the data preprocessing and feature extraction sub-module;

[0167] Apply the corresponding feature extraction technology to extract features from the fused data, and perform data dimensionality reduction and normalization processing to obtain the feature vectors suitable for model input;

[0168] (3) Model training phase:

[0169] Pre-training phase: Input the processed data into the model, perform pre-training according to the contrastive learning objective function, and adjust the model parameters through the optimization algorithm to enable the model to learn the intrinsic feature representation of the data;

[0170] Fine-tuning phase: On the basis of pre-training, use the health data with annotation information to fine-tune the model according to the multi-task loss function, further optimize the model parameters, and enable the model to adapt to the short-term risk warning and long-term trend prediction tasks;

[0171] During the training process, regularly evaluate the model using the validation set, and adjust the hyperparameters and training strategy according to the evaluation results until the model performance reaches the optimal;

[0172] (4) Health portrait generation and update phase:

[0173] When the model training is completed, input the new user health data into the model, and the health portrait generation and update sub-module generates the user's dynamic health portrait according to the output result of the model;

[0174] As the user's health data is continuously updated, receive new data in real time and input it into the model to dynamically update the health portrait, and timely reflect the changes in the user's health status;

[0175] (5) End phase:

[0176] When the system stops running or receives a stop instruction, the health portrait modeling module stops the data processing and model training operations, saves the trained model and related parameters, closes the relevant resources, and makes preparations for the next run.

[0177] In this embodiment, the personalized solution generation module is the core execution module of the personalized diet and exercise health management system that integrates the analysis capabilities of large models. It undertakes the key task of transforming user health data into actionable health management solutions. In the field of health management, individual differences among users are significant, and traditional general solutions are difficult to meet diverse needs. This module aims to integrate medical knowledge and user personalized characteristics to customize accurate diet and exercise plans for users, helping to achieve scientific health management, specifically manifested as follows:

[0178] I. Overall function overview:

[0179] Based on the dynamic health portrait output by the health portrait modeling module, this module deeply integrates the medical knowledge graph and user gene characteristics to build a set of accurate personalized solution generation systems. By applying optimization algorithms and intelligent models, it designs solutions for diet nutrition and exercise planning respectively, generating executable solutions containing key elements such as nutritional ratios, calorie distribution, and exercise intensity thresholds, realizing the personalization and precision of health management;

[0180] II. Composition and functions of sub - modules:

[0181] (I) Nutrition optimization sub - module:

[0182] Data integration and analysis: Receive the user health information provided by the health portrait modeling module, including physiological indicators, disease risks, diet preferences and other data. At the same time, combine the information related to nutritional metabolism in the user's gene test report and the knowledge about nutrient requirements and food nutritional components in the medical knowledge graph for comprehensive analysis. For example, based on the user's body fat percentage and blood sugar level, judge their metabolic ability for carbohydrates and fats, and combine the information of specific gene loci in the gene test to determine personalized nutritional needs;

[0183] Construction of an improved linear programming model: Based on the analysis results, use an improved linear programming model for nutrition solution design. The objective function of the model is: (where, is the number of types of food ingredients; is the unit cost of food ingredient ; is the target cost of food ingredient ; is the metabolic adaptation coefficient (calculated based on the user's gene test report), aiming to minimize the difference between the food ingredient cost and the target cost while meeting nutritional requirements and considering the user's metabolic adaptability. The constraint conditions include: (where, is the quantity of food ingredient; is food ingredient Essential nutrients content; For food ingredients content of limited control nutrients as follows: is the lower limit of the intake of essential nutrients stipulated by medical guidelines ; is the upper limit of the intake of limited control nutrients stipulated by medical guidelines ; is the food ingredient selection decision variable, indicating the selection of this food ingredient, indicating not to select this food ingredient), ensuring sufficient intake of essential nutrients and non-exceedance of limited control nutrients;

[0184] Plan generation and optimization: By solving the linear programming model, determine the selection combination and ratio of food ingredients to generate a preliminary nutrition plan; at the same time, considering factors such as the user's dietary preferences and food allergy history, optimize and adjust the plan, such as replacing food ingredients that the user does not like or is allergic to, and improving the acceptability of the plan on the premise of meeting nutritional requirements;

[0185] (2) Exercise planning sub-module:

[0186] Exercise demand analysis: Based on the information of the user's physical condition, exercise ability, health goals, etc. provided by the health portrait modeling module, analyze the user's exercise demand; for example, for users with overweight, determine the main goal of fat loss; for users with weak cardiovascular function, focus on improving cardiopulmonary function; combined with the impact of different exercises on various systems of the body in the medical knowledge graph, clarify the general range of exercise type, intensity and duration;

[0187] Genetic algorithm parameter optimization: Use the genetic algorithm to optimize the parameters of the exercise plan, and the fitness function is: (wherein, is the actual exercise heart rate; is the median of the target heart rate range; is the estimated value of the metabolic equivalent cost; is the fatigue adjustment factor (default 0.3)); this function evaluates the quality of the exercise plan by measuring the closeness of the actual exercise heart rate to the target heart rate, as well as the metabolic cost and fatigue of the exercise; by simulating the selection, crossover and mutation operations in the biological evolution process, continuously optimize the parameters of the exercise plan, such as exercise type, exercise intensity, exercise duration, etc.;

[0188] Personalized exercise plan generation: Based on the parameters optimized by the genetic algorithm and combined with the actual situation of the user's exercise environment, time arrangement, etc., generate a personalized exercise plan that includes specific content such as exercise items, exercise intensity, exercise frequency, and exercise duration; for example, design fragmented exercise plans for users with busy work, and recommend suitable outdoor projects and routes for users who like outdoor sports;

[0189] (3) Scheme integration and output sub-module:

[0190] Scheme integration: Integrate the nutrition plan generated by the nutrition optimization sub-module and the exercise plan generated by the exercise planning sub-module to ensure that they are coordinated with each other in terms of goals, time arrangements, etc.; for example, adjust the dietary calorie intake according to the exercise intensity and duration to make the nutritional supplement match the exercise consumption;

[0191] Scheme visual output: Output the integrated personalized scheme in an intuitive and easy-to-understand form, such as generating a graphic-rich diet plan, exercise schedule, etc.; at the same time, provide a detailed description of the scheme to explain the basis and precautions for scheme formulation to facilitate user understanding and implementation;

[0192] III. Key technical principles:

[0193] (1) Principle of improved linear programming model:

[0194] Linear programming is a mathematical method for solving the optimal solution of the objective function under a set of linear constraint conditions; in the nutrition optimization sub-module, the improved linear programming model introduces a metabolic adaptation coefficient to consider the impact of the user's individual genetic differences on nutrient metabolism, making the model more in line with the actual needs of the user; by setting the upper and lower limit constraint conditions of nutrient intake and combining the objective function of ingredient cost, the optimal solution of ingredient selection and ratio is achieved under the dual requirements of meeting nutritional needs and cost control, and a scientific and reasonable nutrition plan is formulated for the user;

[0195] (2) Principle of genetic algorithm:

[0196] The genetic algorithm is a stochastic search algorithm that simulates the process of natural evolution. Through operations such as selection, crossover, and mutation, the individuals in the population are evolved to find the optimal solution; in the exercise planning sub-module, the parameters of the exercise plan are encoded as individuals, and the fitness function is used to evaluate the quality of the individuals, and the parameters of the exercise plan are continuously optimized by simulating the biological evolution process; this algorithm can perform efficient search in the complex parameter space, adapt to the personalized exercise needs of different users, and generate an exercise plan that meets the user's physical condition and goals;

[0197] (3) Principle of the integration of medical knowledge graph and gene characteristics:

[0198] The medical knowledge graph integrates a large amount of medical knowledge and clinical experience, including structured information such as disease diagnosis, treatment plans, and nutritional requirements; the user's genetic characteristics reflect individual differences at the genetic level and are closely related to health management; integrating the two with the user's health data can provide more comprehensive knowledge support and accurate individual difference information for personalized plan generation; for example, judging the user's absorption ability of certain nutrients based on genetic characteristics, and formulating a more suitable nutrition plan in combination with the nutrition standards in the medical knowledge graph; designing a personalized exercise plan based on the association between genes and exercise ability and the impact of exercise on health in medical knowledge.

[0199] IV. Workflow of the module:

[0200] (I) Initialization stage:

[0201] After the personalized plan generation module is started, it loads the medical knowledge graph data and establishes connections with the health portrait modeling module and the gene detection data interface to ensure that the required user information can be obtained.

[0202] Initialize the relevant parameters of the nutrition optimization sub-module and the exercise planning sub-module, such as the constraint conditions of the linear programming model, the population size of the genetic algorithm, the crossover and mutation probability, etc.

[0203] (II) Data reception and analysis stage:

[0204] Receive the user's health portrait data output by the health portrait modeling module, including information such as physiological indicators, health risks, and behavior habits.

[0205] Obtain the user's gene detection report data and extract the gene characteristic information related to nutrition metabolism and exercise ability.

[0206] Integrate and analyze the received data, and combine with the medical knowledge graph to clarify the user's personalized nutritional needs and exercise goals.

[0207] (III) Plan generation stage:

[0208] Nutrition plan generation: The nutrition optimization sub-module constructs an improved linear programming model according to the data analysis results, solves the model to obtain a preliminary nutrition plan, and optimizes and adjusts it according to factors such as user preferences.

[0209] Exercise plan generation: The exercise planning sub-module optimizes the exercise plan parameters using the genetic algorithm based on the exercise demand analysis and generates a personalized exercise plan.

[0210] The plan integration and output sub-module integrates the nutrition plan and the exercise plan to ensure their coordination with each other.

[0211] (IV) Plan output stage:

[0212] Visualize the integrated personalized plan and output it in a form that is easy for users to understand, such as displaying a diet plan, exercise schedule, etc. through a mobile application, web interface, etc., and providing a detailed plan description;

[0213] (V) End stage:

[0214] When new user data is received or the user requests plan adjustment, repeat the above process to update the personalized plan; when the system stops running, save the relevant data and plan templates, and close the module operation resources.

[0215] In this embodiment, the scenario-based recommendation engine is an important component for realizing precise service push in the personalized diet and exercise health management system that integrates the analysis ability of large models; in the health management scenario, the diet and exercise needs of users will change due to scenario factors such as geographical location, time, and behavior patterns, and traditional recommendation methods are difficult to meet diverse needs; the engine aims to combine the user's real-time scenario information, dynamically match diet and exercise scenario resources, and provide personalized recommendation services that are more in line with the actual needs of users, improving the practicality and user experience of the health management system, specifically manifested as:

[0216] I. Overall function overview:

[0217] The engine constructs an intelligent recommendation system based on the analysis of the user's real-time scenario and behavior pattern. Through the real-time monitoring and analysis of data such as the user's geographical location, time information, and behavior trajectory, combined with an improved recommendation algorithm, it dynamically generates recommendations for diet and exercise resources that conform to the user's current scenario; it can accurately match corresponding diet venues, exercise programs, and related services according to different scenario requirements, making the health management plan better integrate into the user's daily life, and improving the user's usage frequency and satisfaction with the health management system;

[0218] II. Composition and functions of sub-modules:

[0219] (I) Scenario perception and analysis sub-module:

[0220] Scenario data collection: Connect to the multi-source data collection module to obtain real-time scenario-related data such as the user's geographical location information (through GPS positioning), time information (system time), and behavior trajectory data (such as exercise accelerometer data, screen usage time log); at the same time, obtain environment-related data from the outside, such as temperature, humidity, air quality index (AQI), ultraviolet intensity, etc. provided by the meteorological API, to provide comprehensive data support for scenario analysis;

[0221] Scene Feature Extraction: Process and analyze the collected scene data to extract key scene features. For example, determine the type of area where the user is located based on geographical location information (such as business district, residential area, park, etc.); combine time information and behavioral trajectory data to judge the user's current activity status (such as working, resting, exercising, etc.); evaluate the impact of current environmental conditions on diet and exercise based on environmental data (such as hot weather is suitable for light food and indoor exercise).

[0222] Scene Recognition and Classification: Based on the extracted scene features, use machine learning algorithms (such as decision trees, support vector machines, etc.) to recognize and classify the user's current scene; divide the scene into different types, such as dining scenes (breakfast, lunch, dinner, snacks), exercise scenes (morning exercise, exercise after work, weekend exercise), special scenes (business trips, tourism), etc., to provide a clear scene basis for subsequent recommendations.

[0223] (2) Diet Recommendation Algorithm Sub-module:

[0224] Data Integration and Processing: Integrate the user's health profile data (such as nutritional needs, dietary preferences, allergy history), historical diet data (order records, evaluation information), and merchant information (dish information, price, location, user evaluation); clean, preprocess, and perform feature engineering on this data to extract effective features for the recommendation algorithm, such as the preference score of the user for different dishes, the comprehensive score of the merchant, etc.

[0225] Construction of an Improved Collaborative Filtering Model: Use an improved collaborative filtering model for diet recommendations. The recommendation score calculation formula is: (where is the recommendation score of user for merchant ; is the set of similar neighbors of user ; is the similarity weight between user and user ; is the rating of user for merchant ; is the average rating of user ; is the bias term of merchant ; is the spherical distance between the user and the merchant; is the distance decay function (using exponential decay ; is merchant The price)); Based on the traditional collaborative filtering algorithm, this model introduces the distance factor and price factor between users and merchants, making the recommendation results more in line with the actual needs of users;

[0226] Recommendation result generation and sorting: Calculate the recommendation score of each merchant for the user according to the improved collaborative filtering model, sort the merchants according to the score, and generate a personalized diet recommendation list; At the same time, combine the user's current scene information to screen and adjust the recommendation results. For example, recommend fast food restaurants near the user during lunchtime, and recommend special restaurants during special festivals;

[0227] (III) Sports venue matching algorithm sub-module:

[0228] Integration of sports resource data: Collect and organize various sports venue information, including gyms, parks, swimming pools, stadiums, etc., covering data such as the geographical location, opening hours, facility types, capacity, and user reviews of the venues; At the same time, obtain the user's health profile data (such as sports ability, health goals, sports preferences) and real-time scene data (such as current location, free time, weather conditions);

[0229] Construction of spatio-temporal accessibility analysis model: Build a sports venue matching algorithm based on spatio-temporal accessibility analysis. The formula for calculating the availability score is: (where is the availability score of the sports venue for the user ; is the geographical coding of the sports venue; is the opening period of the venue; is the user 's free period; is the user 's total free time; is the time (in minutes) for the user to reach the sports venue calculated based on real-time traffic data)); This model comprehensively considers the matching degree between the opening time of the sports venue and the user's free time, as well as the time cost for the user to reach the venue, and evaluates the availability of the sports venue;

[0230] Recommendation and screening of sports venues: Sort the sports venues according to the availability score, and generate a personalized sports venue recommendation list; Combine the user's sports preferences and health goals to screen and optimize the recommendation results. For example, recommend gyms with complete aerobic exercise facilities for users who want to lose weight, and recommend parks or hiking routes with beautiful scenery for outdoor sports enthusiasts; At the same time, adjust the recommendation according to the real-time weather conditions. For example, give priority to recommending indoor sports venues on rainy days;

[0231] (4) Recommendation Result Display and Interaction Sub-module:

[0232] Visualization of Recommendation Results: Display the diet and exercise recommendation results to users in an intuitive and user-friendly interface; adopt a combination of pictures and texts to show the pictures of recommended merchants, dish information, maps of sports venues, facility introductions, etc., so as to facilitate users to quickly understand the details of the recommendations;

[0233] User Interaction Processing: Receive feedback information from users on the recommendation results, such as click, favorite, evaluation and other operations; adjust and optimize the recommendation algorithm according to user feedback to further improve the accuracy of recommendations and user satisfaction; for example, if a user is not satisfied with a certain recommendation, the system records relevant information and reduces the weight of similar recommendations in subsequent recommendations;

[0234] Recommendation Information Push: Actively push recommendation information to users through mobile application push, SMS reminder, etc.; combine the user's usage habits and scenarios to select appropriate push timing and methods, such as pushing recommendations for nearby restaurants before lunch time and pushing suggestions for sports venues when the user has free time;

[0235] III. Key Technical Principles:

[0236] (1) Principle of Improved Collaborative Filtering Algorithm:

[0237] The collaborative filtering algorithm makes recommendations based on the similarity between users. By finding neighbor users with similar interests to the target user, it makes recommendations for the target user according to the preferences of the neighbor users; in the diet recommendation algorithm sub-module, the improved collaborative filtering model introduces a distance decay function and price factor, so that the recommendation results not only consider the similar preferences of users, but also combine the actual distance between users and merchants and price sensitivity; the distance decay function reduces the recommendation weight of merchants exponentially as the distance increases; the price factor undergoes logarithmic transformation, so that merchants with higher prices are inhibited to a certain extent in the recommendation score, thus making the recommendation results more in line with the user's selection behavior in the actual scenario;

[0238] (2) Principle of Spatiotemporal Accessibility Analysis:

[0239] Spatiotemporal accessibility analysis is used to evaluate the difficulty for users to reach a certain place under specific time and space conditions; in the sports venue matching algorithm sub-module, the intersection ratio of the opening hours of the sports venue and the user's free time is calculated to measure the matching degree in terms of time; combined with the arrival time calculated based on real-time traffic data, the spatial accessibility is evaluated; by combining time and space factors and constructing an availability score model in the form of an exponential function, it can accurately reflect the actual availability of the sports venue for users and recommend suitable sports venues for users;

[0240] (3) Principle of Scenario Recognition and Analysis Technology:

[0241] Scene recognition and analysis technology classifies and judges the user's current scene by fusing and processing multi-source data and extracting features, and uses machine learning algorithms. By combining the features of data such as geographical location, time, and behavior trajectory, a scene recognition model is trained to automatically recognize different types of scenes. For example, by analyzing the user's location and behavior patterns during specific working hours on weekdays, it is determined that the user is in a working scene. Based on the outdoor activity trajectory and time information on weekends, it is recognized that the user is in a leisure and sports scene. Accurate scene recognition provides an important basis for subsequent personalized recommendations.

[0242] IV. Workflow of the module:

[0243] (I) Initialization stage:

[0244] After the scene-based recommendation engine is started, connections are established with the multi-source data collection module and the health profile modeling module to ensure that the required user data and scene information can be obtained.

[0245] Load recommendation resources such as food merchant data and sports venue data, and perform preprocessing and index construction on the data to improve data query and processing efficiency.

[0246] Initialize the relevant parameters of the food recommendation algorithm and the sports venue matching algorithm, such as the number of neighbors in the collaborative filtering model, the time weight of the spatio-temporal reachability analysis model, etc.

[0247] (II) Scene perception and analysis stage:

[0248] Real-time receive user scene-related data transmitted by the multi-source data collection module, including geographical location, time, behavior trajectory, environmental information, etc.

[0249] Process and analyze the collected data, extract scene features, and use scene recognition algorithms to judge the type of the scene where the user is currently located.

[0250] Transmit the recognized scene information to the food recommendation algorithm sub-module and the sports venue matching algorithm sub-module to provide a scene basis for recommendations.

[0251] (III) Recommendation algorithm execution stage:

[0252] Food recommendation: The food recommendation algorithm sub-module calculates the recommendation scores of merchants using an improved collaborative filtering model based on the user's scene information, health profile data, and historical food data, and generates a personalized food recommendation list.

[0253] Recommendation of sports venues: The sports venue matching algorithm sub-module combines the user's scenario information, health profile data, and sports resource data, calculates the availability score of sports venues through the spatio-temporal accessibility analysis model, and generates a personalized list of recommended sports venues;

[0254] (4) Recommendation result display and interaction stage:

[0255] The recommendation result display and interaction sub-module visualizes the diet and exercise recommendation lists and displays them to the user through the system interface;

[0256] Receive the feedback information of the user on the recommendation results, adjust and optimize the recommendation algorithm according to the feedback, and update the recommendation model parameters;

[0257] According to the user settings and scenario conditions, actively push the recommendation information to the user to remind the user to view and use the recommendation service;

[0258] (5) End stage:

[0259] When the system stops running or receives a stop instruction, the scenario-based recommendation engine stops data collection and recommendation services, saves the recommendation model and related data, closes the connection with other modules, and releases system resources.

[0260] In this embodiment, the closed-loop supervision optimization module is the core guarantee component for the personalized diet and exercise health management system integrating the large model analysis ability to achieve continuous and effective health management; in the practice of health management, the user's implementation of the personalized plan directly affects the health management effect, and the traditional management mode lacks an effective monitoring and dynamic adjustment mechanism for plan implementation; this module constructs a complete monitoring, evaluation, and adjustment system, quantitatively analyzes the implementation effect of the plan based on the federated learning framework, and realizes adaptive optimization, forming a health management closed-loop to ensure that the health management plan always fits the actual situation of the user, improving the effectiveness and sustainability of health management, specifically reflected in:

[0261] I. Overall function overview:

[0262] Based on the federated learning framework, this module constructs a closed-loop management system covering dietary compliance monitoring, exercise effect evaluation, and plan adaptive adjustment; through data collection and analysis during the user's implementation of the personalized diet and exercise plan, it monitors the implementation of the plan in real time, uses advanced algorithm models to evaluate the implementation effect, and dynamically optimizes and adjusts the plan according to the evaluation results, so as to achieve the full-process closed-loop control of health management from plan formulation, implementation to optimization, and ensure the achievement of health management goals;

[0263] II. Composition and functions of sub-modules:

[0264] (1) Dietary compliance monitoring sub-module:

[0265] Data collection and preprocessing: Collaborate with the multi-source data collection module to obtain user diet-related data, including meal images and weight information collected by intelligent devices (such as cameras and food scales), as well as diet records manually entered by users; preprocess the collected data, such as image enhancement and format conversion, to provide high-quality data for subsequent analysis;

[0266] Application of improved residual network: Use the improved residual network for meal recognition, and calculate the dietary compliance deviation through the following formula: (where, is the deviation value of dietary compliance; is the number of nutrient types; is the nutrient obtained by image recognition content (nutrients include II and III); is the intake of nutrients specified in the plan ; is the anti-zero constant (set to ); is the risk weight of nutrient (dynamically adjusted according to the user's health profile)); Utilize the powerful feature extraction ability of the residual network to accurately identify the ingredients and nutrient content in the meal, and compare with the nutrition goals in the personalized plan to quantify the deviation degree between the user's diet and the plan;

[0267] Abnormal warning and feedback: Set the dietary compliance deviation threshold. When the calculated exceeds the threshold, trigger the abnormal warning mechanism; Send warning messages to users through the system interface, mobile phone push, etc., prompt the difference between the current diet and the plan, and give adjustment suggestions; At the same time, synchronize the warning information to the system management end for professionals to intervene and guide;

[0268] (2) Exercise effect evaluation sub-module:

[0269] Exercise data collection and processing: With the help of intelligent wearable devices (such as smart bracelets and sports watches) and mobile terminal sensors, collect data during the user's exercise, including information such as exercise trajectory, heart rate, acceleration, and joint angle; Process the collected data, such as filtering and noise reduction, to remove interference factors and extract effective exercise features;

[0270] Application of dynamic time warping algorithm: Use the dynamic time warping algorithm (DTW) to calculate the action standard degree, and the formula is as follows: (where, is the action standard degree score; is the standard action sequence and the dynamic time warping distance between the user's executed action sequence ; is the length of the standard action sequence; is the length of the user's executed action sequence; is the maximum physiological range of joint movement); Through this algorithm, the user's actual movement action sequence is matched and compared with the standard action sequence to evaluate the standardization and accuracy of the user's movement actions;

[0271] Comprehensive evaluation of exercise effect: Combining multi-dimensional indicators such as action standard score, exercise intensity (such as heart rate data), and exercise duration, comprehensively evaluate the user's exercise effect; According to the user's health goals (such as fat loss, muscle gain, improving cardiopulmonary function, etc.), analyze whether the exercise achieves the expected effect, and generate an exercise effect evaluation report to provide a basis for program adjustment;

[0272] (3) Scheme adaptive adjustment sub-module:

[0273] Analysis of evaluation results: Receive the evaluation results of the dietary compliance monitoring sub-module and the exercise effect evaluation sub-module, and comprehensively analyze the user's program execution situation; Identify problems existing in the user's diet and exercise, such as unbalanced nutrient intake and insufficient exercise intensity;

[0274] Adaptive adjustment strategy: Based on the evaluation results, use the federated learning framework, combined with the user's health profile and historical data, to develop a personalized program adjustment strategy; For the dietary program, if it is found that a certain nutrient is insufficiently ingested, recommend alternative ingredients or adjust the recipe through a nutritional substitution decision tree; For the exercise program, adjust parameters such as exercise type, intensity, and duration according to the exercise effect evaluation results; During the adjustment process, fully consider the user's preferences and actual situation to ensure the feasibility and acceptability of the adjusted program;

[0275] Program update and push: Update the adjusted personalized diet and exercise program to the system and push it to the user; At the same time, explain the reasons and expected effects of the program adjustment to the user, and guide the user to execute according to the new program; During the subsequent monitoring process, continuously pay attention to the user's execution of the adjusted program to form a closed-loop optimization;

[0276] (4) Federated learning collaboration sub-module:

[0277] Data Interaction and Privacy Protection: Under the federated learning framework, coordinate data interaction among various participants (such as client side, medical institution side, and health management institution side); ensure the secure transmission and sharing of parameters through encryption technology without the data leaving the local area, protecting user data privacy; for example, when the client side uploads the data of the plan execution, homomorphic encryption technology is used to encrypt the data to ensure the security of the data during transmission and aggregation;

[0278] Model Collaborative Training: Organize all parties to participate in the collaborative training of the model, and collect the model parameters obtained by each party through local data training; use technologies such as federated averaging algorithm to aggregate the parameters by combining the data volume weight and the model performance weight, and update the global model; through continuous iterative training, enable the global model to integrate the data characteristics of multiple parties and improve the accuracy of evaluating and adjusting the execution of the user health management plan;

[0279] Knowledge Sharing and Optimization: Promote knowledge sharing among all parties, and integrate the professional knowledge and experience of different participants in the field of health management into the model; for example, the clinical data and medical knowledge provided by medical institutions can help optimize the adjustment strategies of diet and exercise plans, and the practical experience of health management institutions can improve the monitoring and evaluation methods, so as to achieve the continuous optimization of the entire closed-loop supervision and optimization module.

[0280] The present invention also includes a risk warning subsystem, which is a key line of defense for preventing health risks in the personalized diet and exercise health management system integrating the analysis ability of large models; in the health management scenario, the user's health status may change rapidly due to the sudden occurrence of acute diseases or the deterioration of chronic diseases. This subsystem constructs a scientific prediction model, real-time monitors and analyzes the user's health data, identifies potential risks in advance, and provides timely and effective warning information for users and relevant parties in health management, helping to reduce the harm of health risks, specifically manifested as:

[0281] I. Overall Function Overview:

[0282] Based on the user's multi-dimensional health data, this subsystem uses machine learning algorithms and mathematical models to construct two core modules: acute risk prediction and chronic disease development scoring; through in-depth analysis of information such as real-time physiological indicators and historical health data, it realizes the instant warning of acute health risks and the long-term tracking of the development trend of chronic diseases, providing forward-looking health risk prompts for users so as to take targeted preventive and intervention measures;

[0283] II. Composition and Functions of Sub-modules:

[0284] (1) Acute Risk Prediction Module:

[0285] Data integration and feature extraction: Obtain the user's real-time physiological index data from the multi-source data collection module, focusing on the RR interval data related to heart rate variability (HRV) and the change data of blood oxygen saturation (SpO2); perform preprocessing such as noise reduction and filtering on these data, and extract key features such as the mean and standard deviation of the RR interval, the blood oxygen decline rate, etc., to provide high-quality input for the risk prediction model;

[0286] Construction of acute risk prediction model: Construct an acute risk prediction model using the following formula: (where, is the probability of acute risk occurrence; is the sigmoid activation function, which maps the function output value between 0 and 1 to represent the probability: is the number of heart rate variability (HRV) frequency bands; is the weight coefficient of each heart rate variability (HRV) frequency band, reflecting the influence degree of different frequency bands on acute risk; is the interval value of each heart rate variability (HRV) frequency band; is the sensitivity coefficient of the blood oxygen decline rate; is the blood oxygen decline rate);

[0287] Risk warning and notification: Set the acute risk probability threshold. When the calculated exceeds the threshold, immediately trigger the warning mechanism; send acute risk warning information to the user through mobile application push, SMS, etc., and elaborate on the risk type, possible reasons and emergency response suggestions; at the same time, synchronize the warning information to the associated medical institution or health management expert terminal for professional personnel to intervene and guide in a timely manner;

[0288] (2) Chronic disease development scoring module:

[0289] Data collection and processing: Collect the user's long-term physiological index data, such as key indicators related to chronic diseases such as fasting blood glucose, cholesterol, blood pressure, etc., as well as historical health portrait data and medical diagnosis records; perform time series analysis on the data to calculate dynamic features such as the weekly change rate of each indicator, providing rich analysis materials for the scoring model;

[0290] Construction of chronic disease development scoring model: Construct a chronic disease development scoring model based on the following formula: (where, is the chronic disease development score at the moment; is the chronic disease development score at the moment, reflecting the influence of the historical score on the current score; is the historical impact attenuation factor (default 0.85), which is used to adjust the weight of historical scores; is the number of key physiological indicators; is the key physiological indicators specified by medical guidelines weight, reflecting the importance of different indicators in the development of chronic diseases: are key physiological indicators (fasting blood glucose, cholesterol, etc.) weekly change rate); this model combines historical scores and current indicator changes to dynamically evaluate the development trend of chronic diseases;

[0291] Trend analysis and intervention suggestions: Regularly update the chronic disease development score, and analyze the score change trend; When the score shows a significant increase or reaches the warning threshold, generate a detailed chronic disease development trend report, and push personalized health intervention suggestions to users, including diet adjustment, exercise plan change, medical examination reminder, etc.; At the same time, synchronize the report to the user's family doctor or health management team to assist in formulating long-term health management strategies;

[0292] (3) Data management and model optimization module:

[0293] Data storage and maintenance: Establish a dedicated database to classify and store all data involved in the acute risk prediction and chronic disease development scoring processes, including original health data, preprocessed data, model input and output data, etc.; Regularly back up and clean the data to ensure the integrity and validity of the data, and provide reliable data support for model training and risk analysis;

[0294] Model training and optimization: Use newly accumulated health data to regularly retrain and optimize the acute risk prediction model and chronic disease development scoring model; By adjusting model parameters, improving algorithm structures, etc., improve the prediction accuracy and adaptability of the model; For example, update the index weight coefficients according to new medical research results, and introduce more dimensional data features to improve model performance;

[0295] Performance evaluation and monitoring: Continuously monitor the operating performance of the risk warning subsystem, and evaluate indicators such as the accuracy rate and recall rate of the model by comparing the actual health events with the model prediction results; Once it is found that the model performance declines or there are deviations, analyze and adjust in a timely manner to ensure the reliability and effectiveness of risk warning.

[0296] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principle and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A personalized diet and exercise health management system integrating the analysis capabilities of large models, characterized in that: It includes: A multi-source data acquisition module that integrates interfaces of intelligent wearable devices, medical data APIs, and mobile terminal sensors to collect users' multi-dimensional health data in real time. The users' multi-dimensional health data includes: Physiological index data: heart rate variability (HRV), ambulatory blood pressure, blood oxygen saturation (SpO2), body fat percentage; Behavioral trajectory data: GPS positioning trajectory, motion accelerometer data, screen usage time log; Environment-related data: temperature and humidity obtained from the meteorological API, air quality index (AQI), ultraviolet intensity; Medical history data: structured fields of electronic medical records, medication contraindication list, allergen database; A health portrait modeling module that performs multi-modal data fusion based on an improved Transformer architecture and outputs a dynamic health portrait including short-term risk warnings and long-term trend predictions; A personalized solution generation module that integrates a medical knowledge graph and users' genetic characteristics to generate an executable solution including nutritional ratio, calorie distribution, and exercise intensity threshold; A scenario-based recommendation engine that combines real-time geographical location and users' behavior patterns to dynamically match diet and exercise scenario resources; A closed-loop supervision and optimization module that realizes the evaluation of the execution effect of the solution and adaptive adjustment through a federated learning framework to form a closed loop of health management.

2. The personalized diet and exercise health management system integrating the large model analysis ability according to claim 1, wherein: The multi-source data acquisition module preprocesses heterogeneous data using a spatio-temporal feature alignment algorithm. The specific steps are as follows: It is divided into three time periods: morning, daytime, and night according to the data generation time period; The geographical location is encoded into a grid coordinate system with a preset accuracy; Calculate the mean and standard deviation of the data within each spatio-temporal partition; Assign an entropy weight fusion weight value according to the data source type; Perform standardized processing on the original data based on spatio-temporal partitions and multiply by the corresponding source weight value.

3. The personalized diet and exercise health management system integrating the large model analysis ability according to claim 1, characterized in that: The health portrait modeling module adopts a phased training strategy: In the pre-training stage, a contrastive learning objective function is used: Calculate the cosine similarity of the data representation vectors of the same user at different time periods; Construct a loss function to maximize the proportion of the exponential similarity of similar samples; Set a temperature hyperparameter to control the sample discrimination; In the fine-tuning stage, a multi-task loss function is used: Combine the cross-entropy loss of acute health risk classification; The mean square error loss of chronic disease development prediction; The data reconstruction autoencoder loss; Dynamically adjust the weighted coefficients of each loss term.

4. The personalized diet and exercise health management system integrating the large model analysis ability according to claim 1, characterized in that: The personalized solution generation module includes: A nutrition optimization sub-module that uses an improved linear programming model: The objective function is to minimize the weighted absolute deviation sum of the food ingredient cost and the metabolic adaptation coefficient; The constraint conditions include the minimum intake of essential nutrients and the maximum intake of restricted nutrients; The decision variable is a set of food ingredients with binary choices; An exercise planning sub-module that uses a genetic algorithm for parameter optimization: The fitness function is inversely proportional to the deviation from the median of the target heart rate interval; Introduce the estimated value of the metabolic equivalent cost and the fatigue adjustment factor; Retain the optimal combination of exercise parameters through iterative selection.

5. The personalized diet and exercise health management system integrating the large model analysis ability according to claim 1, characterized in that: The scenario-based recommendation engine includes: A diet recommendation algorithm based on an improved collaborative filtering model: Comprehensively sum the weighted historical rating deviations of similar users; Overlay the geographical location attenuation function and the price logarithm suppression factor; Dynamically adjust the weight distribution of the set of similar user neighbors; A sports venue matching algorithm based on spatio-temporal reachability analysis: Calculate the intersection ratio of the user's idle period and the venue opening period; Introduce an exponential decay factor based on real-time traffic time; The comprehensive evaluation result is the venue availability score.

6. The personalized diet and exercise health management system integrating the large model analysis ability according to claim 1, characterized in that: The closed-loop supervision optimization module realizes: Dietary compliance monitoring, and meal recognition is performed through an improved residual network: Calculate the relative deviation between the recognized nutrient content and the planned value; Introduce an anti-zero constant to handle the situation of micronutrients; Dynamically adjust the deviation weight according to the user's health risk profile; Exercise effect evaluation, using the dynamic time warping algorithm: Calculate the minimum alignment distance between the user's action sequence and the standard sequence; Normalize according to the maximum physiological range of motion of the joints; Generate an action standardization score.

7. The personalized diet and exercise health management system integrating the large model analysis ability according to claim 1, characterized in that: It also includes a risk warning subsystem, and the risk warning subsystem includes: Acute risk prediction model: Weighted sum of the coefficients of each frequency band of heart rate variability; Superimpose the sensitivity coefficient of the blood oxygen decline rate; Output the acute risk probability value through the sigmoid function; Chronic disease development score model: Weighted sum based on the weekly change rate of key physiological indicators; Introduce a historical score decay factor to achieve time series fusion; Output a dynamically updated chronic disease development index.

Images