Health risk assessment method and early warning system based on multi-source data analysis

Abstract

The invention discloses a health risk assessment method and early warning system based on multi-source data analysis. The health risk assessment method comprises the following steps: S1, collecting and preprocessing multi-source health data through medical detection equipment; s2, extracting key health indexes, sequence features and statistical features based on the health data set; s3, adopting a recurrent neural network model to construct a health risk assessment model; s4, optimizing structural parameters of the health risk assessment model by adopting an improved dragonfly algorithm; s5, performing performance evaluation on the health risk evaluation model by using the optimal parameter set; s6, deploying the final health risk assessment model in a health risk assessment and early warning system; and S7, when the health risk assessment result exceeds a preset risk threshold, automatically generating early warning information. According to the method, the recurrent neural network model, the improved dragonfly algorithm and the multi-source health data fusion optimization technology are combined, and health risk assessment and early warning based on multi-source data analysis are realized.

Description

Technical Field

[0001] The present invention relates to the technical field of health risk assessment, and in particular to a health risk assessment method and early warning system based on multi-source data analysis. Background Art

[0002] With the rapid development of society and the accelerated aging of the population, the incidence of chronic non-communicable diseases such as cardiovascular disease and diabetes has increased year by year, seriously threatening people's safety and quality of life. In recent years, sudden health emergencies such as sudden death have become frequent, placing a huge medical and economic burden on families and society. Early identification and intervention of health risks have become important issues that urgently need to be addressed in the fields of public health and health management services. However, the existing health management service system still has many shortcomings in practical application.

[0003] Currently, most health management services rely primarily on traditional methods such as hospital physical examinations, regular follow-up visits, or manual questionnaires. Monitoring of individual health status is often static and intermittent, lacking continuity and timeliness. These services typically focus only on physical examination results at a specific point in time and are unable to track individual health changes in real time and dynamically, making it difficult to promptly capture early warning signals for the progression of chronic diseases or sudden health risks. At the same time, with the prevalence of health monitoring technologies such as wearable devices and mobile medical devices, the sources of personal health data are becoming increasingly diverse, but existing health management systems are often unable to achieve real-time integration of data from multiple devices and platforms. Various types of health data are scattered across different terminals and platforms, lacking a unified and standardized collection, storage, and sharing mechanism. This has led to prominent data silos and information fragmentation, severely restricting the comprehensive utilization and value mining of health data.

[0004] In addition, existing health risk assessment methods are mostly based on a single data source or limited health indicators, and mainly use simple statistical analysis or expert experience rules to classify and judge risks. These methods have limited ability to identify health risks and are difficult to adapt to the increasingly complex and heterogeneous health data environment. For the integration and analysis of multi-source heterogeneous information including vital sign monitoring, laboratory tests, medical imaging, medical history files, lifestyle, etc., current health risk assessment models lack efficient feature extraction, data fusion and dynamic modeling capabilities. Especially when faced with health scenarios with strong time series, large individual differences, and many interference factors, traditional models find it difficult to accurately depict the evolution of health risks over time, making it difficult to implement early warning and intervention for high-risk individuals.

[0005] A more prominent problem is that most existing systems lack immediate early warning and intelligent feedback mechanisms for health risks. When individuals experience abnormal physiological parameters or elevated health risks, the systems often fail to automatically identify and promptly deliver warnings, lacking the ability to rapidly intervene and address emergencies. In actual health management, doctors and healthcare professionals struggle to keep abreast of each user's dynamic health risks, resulting in delayed personalized interventions and targeted prevention and control measures, and missed opportunities for optimal treatment and intervention.

[0006] In terms of data processing and model optimization, the data preprocessing and feature engineering of traditional health risk assessment systems mostly rely on manual settings, with a low degree of automation, making it difficult to adapt to the processing needs of large-scale, multi-type, and multi-dimensional health data. Although advanced artificial intelligence technologies such as deep learning have demonstrated significant advantages in fields such as medical imaging and genomics, their application in multi-source fusion of health data, time series modeling, and risk prediction still needs to be deepened. Existing models mostly use manual parameter adjustment or simple search strategies in terms of structural parameter selection and performance optimization, lacking an intelligent global optimization mechanism. Model training is prone to falling into local optimality, affecting the accuracy and reliability of risk assessment. In addition, the system lacks an extensible automatic early warning push and health management recommendation generation module, making it difficult to provide users with full-process and life-cycle health protection.

[0007] To sum up, the existing health management and risk assessment systems generally have the following defects: First, there is a lack of continuous and dynamic monitoring of chronic disease data, which makes it impossible to track and intervene in the entire process of individual health risks; second, multi-source health data is difficult to integrate in real time, and the problem of health information islands is prominent, which affects the comprehensive utilization of data and the scientific nature of risk assessment; third, the health risk model lacks the ability to capture complex time series and multi-dimensional features, resulting in low accuracy and timeliness of risk prediction; fourth, there is a lack of efficient anomaly detection and automatic early warning mechanisms, making it difficult to respond to and handle sudden health events in a timely manner; fifth, the intelligent optimization and automated processing capabilities are insufficient, the model performance is limited, and it is difficult to meet the needs of large-scale, intelligent, and personalized health management.

[0008] Therefore, how to provide a health risk assessment method and early warning system based on multi-source data analysis is an urgent problem that needs to be solved by those skilled in the art. Summary of the Invention

[0009] One purpose of the present invention is to propose a health risk assessment method and early warning system based on multi-source data analysis. The present invention combines a recursive neural network model, an improved dragonfly algorithm and multi-source health data fusion optimization technology. Through time series deep learning modeling, intelligent parameter optimization and efficient integration of heterogeneous health data, it accurately mines the dynamic correlation characteristics between individual health indicators and realizes continuous monitoring and risk trend analysis of chronic disease data. The system uses a recursive neural network to extract dynamic features from high-dimensional, multi-time series health data, and uses the improved dragonfly algorithm to perform global optimization of the model structure and parameters, further improving the accuracy and generalization ability of risk prediction. At the same time, through multi-source data fusion optimization technology, the system can integrate multi-channel health information such as physical examinations, wearable devices, and medical records in real time to build a comprehensive health portrait. In the risk assessment and early warning links, the system supports automatic graded early warning and personalized health management suggestion push, realizing early intervention and dynamic management of high-risk populations throughout the process, significantly improving the scientificity, timeliness and intelligence level of health risk prevention and control.

[0010] A health risk assessment method based on multi-source data analysis according to an embodiment of the present invention includes the following steps:

[0011] S1. Collect multi-source health data through medical testing equipment and pre-process it to build a health data set;

[0012] S2, extract key health indicators, sequence features and statistical features based on the health data set, and generate multi-source fusion feature vectors;

[0013] S3. Based on multi-source fusion feature vectors, a recursive neural network model is used to construct a health risk assessment model. The individual health status is modeled as an intelligent agent with the ability to perceive time series characteristics and predict health risks. The model predicts the evolution of individual health risks over time and outputs preliminary health risk assessment results.

[0014] S4. Based on the preliminary health risk assessment results, the improved dragonfly algorithm is used to optimize the structural parameters of the health risk assessment model. A local search and global search coordination mechanism is configured for each dragonfly. Based on the search paths and dynamic position update behaviors of all dragonflies, the structural parameters are coordinated and optimized through centralized training and distributed execution to generate the optimal parameter set.

[0015] S5. Use the optimal parameter set to evaluate the performance of the health risk assessment model and output the final health risk assessment model;

[0016] S6. Deploy the final health risk assessment model to the health risk assessment and early warning system, receive and analyze newly acquired health data in real time, and output individual health risk levels;

[0017] S7. When the health risk assessment result exceeds the preset risk threshold, an early warning message is automatically generated and pushed to doctors in relevant departments and on-site staff.

[0018] Optionally, the multi-source health data specifically includes personal basic information, physical examination data, physiological indicators, laboratory test results, medical imaging data, past medical history, family history, lifestyle information and medical testing equipment data.

[0019] Optionally, the key health indicators specifically include blood pressure, blood sugar, blood lipids, body mass index, heart rate, liver and kidney function indicators, blood oxygen saturation, urine routine, and electrocardiogram parameters.

[0020] Optionally, the S2 specifically includes:

[0021] S21. Perform format normalization, time alignment, and data integrity verification on various types of original health data in the health data set to obtain a normalized health data set;

[0022] S22. Extracting basic health indicators from the structured health data in the normalized health data set using a feature engineering method to obtain a basic health indicator set;

[0023] S23. Segment and perform statistical analysis on the time series health data in the normalized health data set by using a sliding window to extract time series features and obtain a time series feature set;

[0024] S24. Automatically extracting features from the unstructured health data in the normalized health dataset using a deep feature encoding network to obtain an unstructured feature set;

[0025] S25. Perform feature concatenation and normalization processing on the basic health indicator set, the time series feature set, and the unstructured feature set to generate a fused feature set;

[0026] S26. Perform feature screening and dimensionality reduction processing on the fused feature set, and output a multi-source fused feature vector.

[0027] Optionally, the S3 specifically includes:

[0028] S31, performing time step expansion on the multi-source fusion feature vector to generate a time series feature sequence;

[0029] S32. Initialize the structural parameters of the recursive neural network based on the time series feature sequence, configure the network structure according to the actual requirements of the health risk assessment task, and randomly initialize the weights and bias parameters of each layer to build a health risk assessment model:

[0030]

[0031] Among them, W (l)Represents the weight initialization vector of the lth layer, l represents the layer number of the recursive neural network, N represents the total amount of data used for weight initialization, i takes a value between 1 and N, and represents the i-th group number in the multi-source fusion feature vector group, M represents the feature dimension of each group of multi-source fusion feature vectors, j takes a value between 1 and M, and represents the specific j-th feature number in each group of multi-source fusion feature vectors, It represents the value of the jth feature in the i-th group of multi-source fusion feature vectors when the l-th layer weight is initialized;

[0032] S33, inputting the time series feature sequence into the health risk assessment model, performing time series feature modeling on the individual health status, and generating a dynamic health status representation;

[0033] S34. Based on the dynamic health status representation, a sequence of health risk prediction values ​​of the individual at each time step is generated through the output layer of the health risk assessment model;

[0034] S35. Summarize and process the health risk prediction value sequence and output preliminary health risk assessment results.

[0035] Optionally, the S4 specifically includes:

[0036] S41. Using the improved dragonfly algorithm, the preliminary health risk assessment results are used as the basis for initializing the dragonfly population, setting the number of individuals in the dragonfly population, the search space range, and the initial values ​​of the structural parameters to achieve individual distribution initialization of the dragonfly algorithm;

[0037] S42. For each initialized dragonfly, based on the preliminary health risk assessment results, configure a collaborative search mechanism of local search and global search, assign a local search strategy and a global search strategy to each dragonfly, clarify the behavioral rules of each dragonfly during the search process, and output the collaborative search configuration results.

[0038] S43, using a local search mechanism, based on the current position and structural parameters of each dragonfly body, by adjacent position perturbations and parameter fine-tuning, generating a local search candidate parameter set for the dragonfly body, and outputting the local candidate parameter set;

[0039] S44, using a global search mechanism to analyze the search paths and dynamic position update behaviors of all dragonflies, globally adjusting the positions of each dragonfly based on the distribution characteristics of all dragonflies, generating a global search candidate parameter set, and outputting the global candidate parameter set;

[0040] S45. The local search candidate parameter set and the global search candidate parameter set are collaboratively integrated. Based on the performance evaluation criteria of the health risk assessment model, all candidate structural parameters are centrally trained and distributedly executed. The performance of each group of structural parameters in the health risk assessment task is evaluated to obtain the structural parameter performance evaluation results. The collaborative integration of local search and global search adopts the improved dragonfly algorithm. Based on the dragonfly algorithm, the improved dragonfly algorithm introduces an adaptive inertia weight adjustment strategy to improve the global search capability and integrates the dynamic neighborhood update mechanism to enhance the local search accuracy:

[0041]

[0042] Among them, P represents the performance evaluation result, L represents the total number of layers of the health risk assessment model, Q represents the number of candidate groups of the structural parameter candidate set, and W (l) represents the weight initialization vector of the lth layer, represents the structural parameter of the qth candidate group in the lth layer, l represents the layer number of the health risk assessment model, q represents the number of the qth candidate group in the candidate structural parameter, r represents the number of the test sample, and t represents the number of the output dimension;

[0043] S46. Filter out the structural parameters with the highest performance evaluation results and generate the optimal structural parameter set for the health risk assessment model.

[0044] Optionally, the S5 specifically includes:

[0045] S51, inputting the optimal structural parameter set into the health risk assessment model to initialize the health risk assessment model;

[0046] S52. Input the training data set into the health risk assessment model in batches, perform forward propagation, calculate the error between the output result and the true label, adjust the health risk assessment model parameters according to the error, and obtain the trained health risk assessment model parameters and health risk assessment model prediction results;

[0047] S53. Based on the trained health risk assessment model parameters, use the test data set to verify the performance of the health risk assessment model, and output the predicted output of the health risk assessment model on the test data set;

[0048] S54. Compare the predicted output of the health risk assessment model on the test data set with the real label data, and generate performance evaluation index output based on the preset performance evaluation criteria:

[0049]

[0050] Among them, U is the performance evaluation index output, R is the total number of test samples, T is the total number of output dimensions, O is the total number of test samples, r,tThe model prediction result of the r-th test sample in the t-th output dimension, Y r,t The rth test sample is the true label data of the tth output dimension, and P represents the performance evaluation result;

[0051] S55. Jointly analyze the performance evaluation index output and the optimal structural parameter set to determine whether the performance of the current health risk assessment model meets the deployment requirements, and output the final model performance evaluation results;

[0052] S56. Output the final health risk assessment model based on the final model performance evaluation results.

[0053] Optionally, the S6 specifically includes:

[0054] S61. Import the final health risk assessment model into the health risk assessment and early warning system;

[0055] S62, the health risk assessment and early warning system receives newly acquired individual health data information in real time and generates a new health data set;

[0056] S63. Input the new health data set into the deployed final health risk assessment model, perform health risk analysis calculations, and obtain individual health risk analysis results;

[0057] S64. Generate individual health risk classification data based on the health risk analysis results;

[0058] S65. Jointly store the individual health risk classification data and the new health data set to form a traceable health risk assessment record and output the final individual health risk level;

[0059] S66. Push the final individual health risk level to the front end of the health risk assessment and early warning system for query by individual users, health managers and doctors;

[0060] Optionally, the S7 specifically includes:

[0061] S71. Input the final individual health risk level into the health risk threshold determination unit of the health risk assessment and early warning system;

[0062] S72. In the health risk threshold determination unit, a preset risk threshold standard is called to perform a risk threshold comparison on the input final individual health risk level to generate a risk threshold determination result;

[0063] S73. Input the risk threshold determination result into the health risk warning generation unit. If the final individual health risk level exceeds the preset risk threshold, a warning message is automatically generated, an alarm is issued, and the on-site staff is notified. The data is forwarded to the doctor in the relevant department for diagnosis. The health manager updates the user's health management plan and promotion plan based on the doctor's advice, and regularly tracks and feeds back to the system for record.

[0064] S74. If the final individual health risk level does not exceed the preset risk threshold and the user belongs to the sub-healthy group, the health manager will develop a complete health management plan and promotion plan for the user.

[0065] A health risk assessment and early warning system based on multi-source data analysis according to an embodiment of the present invention includes the following modules:

[0066] Health data acquisition module, used to collect multi-source health data;

[0067] The health feature fusion module performs feature concatenation and normalization on the basic health indicator set, time series feature set, and unstructured feature set to generate a fused feature set;

[0068] The health risk assessment module summarizes and processes the health risk prediction value sequence and outputs the preliminary health risk assessment results;

[0069] Improve the Dragonfly algorithm module to perform coordinated optimization of structural parameters through centralized training and distributed execution to generate the optimal parameter set;

[0070] The health risk assessment model optimization module outputs the final health risk assessment model based on the final model performance evaluation results;

[0071] The health risk assessment module receives and analyzes newly acquired health data in real time and outputs individual health risk levels;

[0072] The risk warning module is used to push warning information to doctors and health managers.

[0073] The beneficial effects of the present invention are:

[0074] The present invention significantly improves the scientificity, accuracy and intelligence of health risk management by integrating multi-source health data, introducing advanced feature extraction and time series modeling methods, combining intelligent optimization algorithms, and realizing real-time risk assessment and automatic early warning. Compared with the existing technology, the system can automatically collect and integrate health data from multiple sources such as physical examinations, wearable devices, medical archives, etc., breaking through the limitations of data silos and information fragmentation, and effectively supporting the dynamic health management of individuals throughout their life cycle. Based on the application of deep learning models such as recursive neural networks, it not only achieves efficient mining of high-dimensional complex health data and accurate modeling of dynamic time series features, but also improves the automatic tuning capability of model structure parameters through improved intelligent optimization algorithms, ensuring high accuracy and good generalization performance of risk assessment results.

[0075] In addition, the present invention has built a complete health risk grading and early warning mechanism, which can conduct real-time dynamic monitoring of individual health status and determine the risk level. When encountering abnormal or high-risk situations, the system can automatically trigger an early warning and push it to relevant medical service personnel, so as to achieve early detection and early intervention of health risks, greatly improving the timeliness and pertinence of emergency response. The system's automated processing and intelligent feedback capabilities significantly reduce the data processing and manual decision-making burden of medical staff and health managers, and improve the efficiency and quality of health management services. At the same time, the platform has good scalability and compatibility, and can be widely used in multiple scenarios such as chronic disease management, health examinations, disease prevention, insurance assessments, etc. It has important practical significance and social value in promoting the development of smart medical care and improving the health level of the whole people. BRIEF DESCRIPTION OF THE DRAWINGS

[0076] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0077] Figure 1 This is a flow chart of a health risk assessment method based on multi-source data analysis proposed by the present invention;

[0078] Figure 2 This is a schematic diagram of a health risk assessment method based on multi-source data analysis proposed by the present invention;

[0079] Figure 3 This is a data flow diagram of a health risk assessment and early warning system based on multi-source data analysis proposed by the present invention. DETAILED DESCRIPTION

[0080] The present invention will now be described in further detail with reference to the accompanying drawings, which are simplified schematic diagrams that illustrate the basic structure of the present invention in a schematic manner.

[0081] refer to Figure 1-2 , a health risk assessment method based on multi-source data analysis, comprising the following steps:

[0082] S1. Collect multi-source health data through medical testing equipment and pre-process it to build a health data set;

[0083] S2, extract key health indicators, sequence features and statistical features based on the health data set, and generate multi-source fusion feature vectors;

[0084] S3. Based on multi-source fusion feature vectors, a recursive neural network model is used to construct a health risk assessment model. The individual health status is modeled as an intelligent agent with the ability to perceive time series characteristics and predict health risks. The model predicts the evolution of individual health risks over time and outputs preliminary health risk assessment results.

[0085] S4. Based on the preliminary health risk assessment results, the improved dragonfly algorithm is used to optimize the structural parameters of the health risk assessment model. A local search and global search coordination mechanism is configured for each dragonfly. Based on the search paths and dynamic position update behaviors of all dragonflies, the structural parameters are coordinated and optimized through centralized training and distributed execution to generate the optimal parameter set.

[0086] S5. Use the optimal parameter set to evaluate the performance of the health risk assessment model and output the final health risk assessment model;

[0087] S6. Deploy the final health risk assessment model to the health risk assessment and early warning system, receive and analyze newly acquired health data in real time, and output individual health risk levels;

[0088] S7. When the health risk assessment result exceeds the preset risk threshold, an early warning message is automatically generated and pushed to doctors in relevant departments and on-site staff.

[0089] This invention uses a recursive neural network and an improved dragonfly algorithm to collaboratively optimize the model structure, enabling dynamic time-series feature extraction and accurate health risk prediction from multi-source health data. Through a distributed collaborative optimization mechanism, the efficiency of optimizing model parameters and the accuracy of risk assessment are significantly improved. The system integrates real-time early warning and automatic notification capabilities, enabling timely identification of high-risk individuals and providing feedback to physicians, enabling efficient and intelligent dynamic health risk management.

[0090] In this embodiment, the multi-source health data specifically includes personal basic information, physical examination data, physiological indicators, laboratory test results, medical imaging data, past medical history, family history, lifestyle information and medical testing equipment data.

[0091] This technology integrates basic personal information, physical examination data, physiological indicators, laboratory test results, medical imaging, past medical history, family history, lifestyle, and medical device data to achieve comprehensive integration of multidimensional health information. Through collaborative analysis of multi-source data, it effectively improves the integrity and scientific nature of health risk assessment, significantly enhances the accuracy of individual health status identification and the foresight of risk prediction, and provides a solid data foundation for precise health management and early warning.

[0092] In this embodiment, the key health indicators specifically include blood pressure, blood sugar, blood lipids, body mass index, heart rate, liver and kidney function indicators, blood oxygen saturation, urine routine, and electrocardiogram parameters.

[0093] This invention comprehensively collects and analyzes key health indicators, including blood pressure, blood sugar, blood lipids, body mass index, heart rate, liver and kidney function, blood oxygen saturation, urine routine, and electrocardiogram parameters, to accurately depict multidimensional health status. Through multi-indicator joint modeling, the sensitivity and specificity of health risk identification are improved, effectively supporting dynamic risk assessment and personalized health management in complex health conditions, ensuring the scientific nature and applicability of the assessment results.

[0094] In this embodiment, S2 specifically includes:

[0095] S21. Perform format normalization, time alignment, and data integrity verification on various types of original health data in the health data set to obtain a normalized health data set;

[0096] S22. Extracting basic health indicators from the structured health data in the normalized health data set using a feature engineering method to obtain a basic health indicator set;

[0097] S23. Segment and perform statistical analysis on the time series health data in the normalized health data set by using a sliding window to extract time series features and obtain a time series feature set;

[0098] S24. Automatically extracting features from the unstructured health data in the normalized health dataset using a deep feature encoding network to obtain an unstructured feature set;

[0099] S25. Perform feature concatenation and normalization processing on the basic health indicator set, the time series feature set, and the unstructured feature set to generate a fused feature set;

[0100] S26. Perform feature screening and dimensionality reduction processing on the fused feature set, and output a multi-source fused feature vector.

[0101] This method ensures high-quality data input by formatting, time-aligning, and integrity-verifying health datasets. Combining feature engineering, sliding windows, and a deep feature encoding network, it comprehensively extracts structured, temporal, and unstructured features. It then concatenates, normalizes, and reduces the dimensionality of multi-source features to achieve efficient information fusion. This method significantly improves the expressive power and feature relevance of multidimensional health data, providing a solid data foundation for the construction of subsequent health risk assessment models and accurate predictions.

[0102] In this embodiment, S3 specifically includes:

[0103] S31, performing time step expansion on the multi-source fusion feature vector to generate a time series feature sequence;

[0104] S32. Initialize the structural parameters of the recursive neural network based on the time series feature sequence, configure the network structure according to the actual requirements of the health risk assessment task, and randomly initialize the weights and bias parameters of each layer to build a health risk assessment model:

[0105]

[0106] Among them, W (l) Represents the weight initialization vector of the lth layer, l represents the layer number of the recursive neural network, N represents the total amount of data used for weight initialization, i takes a value between 1 and N, and represents the i-th group number in the multi-source fusion feature vector group, M represents the feature dimension of each group of multi-source fusion feature vectors, j takes a value between 1 and M, and represents the specific j-th feature number in each group of multi-source fusion feature vectors, It represents the value of the jth feature in the i-th group of multi-source fusion feature vectors when the l-th layer weight is initialized;

[0107] S33, inputting the time series feature sequence into the health risk assessment model, performing time series feature modeling on the individual health status, and generating a dynamic health status representation;

[0108] S34. Based on the dynamic health status representation, a sequence of health risk prediction values ​​of the individual at each time step is generated through the output layer of the health risk assessment model;

[0109] S35. Summarize and process the health risk prediction value sequence and output preliminary health risk assessment results.

[0110] This method achieves dynamic time-series modeling of individual health data by expanding the time series of multi-source fused feature vectors and initializing a recursive neural network. Combining random initialization of multi-layer network parameters with feature sequence input, the system accurately captures the time-varying characteristics of health status and automatically outputs a health risk prediction sequence for each time step. This method improves the timeliness and predictive accuracy of health risk assessment, providing intelligent decision-making support for individualized dynamic risk management and early intervention.

[0111] In this embodiment, the S4 specifically includes:

[0112] S41. Using the improved dragonfly algorithm, the preliminary health risk assessment results are used as the basis for initializing the dragonfly population, setting the number of individuals in the dragonfly population, the search space range, and the initial values ​​of the structural parameters to achieve individual distribution initialization of the dragonfly algorithm;

[0113] S42. For each initialized dragonfly, based on the preliminary health risk assessment results, configure a collaborative search mechanism of local search and global search, assign a local search strategy and a global search strategy to each dragonfly, clarify the behavioral rules of each dragonfly during the search process, and output the collaborative search configuration results.

[0114] S43, using a local search mechanism, based on the current position and structural parameters of each dragonfly body, by adjacent position perturbations and parameter fine-tuning, generating a local search candidate parameter set for the dragonfly body, and outputting the local candidate parameter set;

[0115] S44, using a global search mechanism to analyze the search paths and dynamic position update behaviors of all dragonflies, globally adjusting the positions of each dragonfly based on the distribution characteristics of all dragonflies, generating a global search candidate parameter set, and outputting the global candidate parameter set;

[0116] S45. The local search candidate parameter set and the global search candidate parameter set are collaboratively integrated. Based on the performance evaluation criteria of the health risk assessment model, all candidate structural parameters are centrally trained and distributedly executed. The performance of each group of structural parameters in the health risk assessment task is evaluated to obtain the structural parameter performance evaluation results. The collaborative integration of local search and global search adopts the improved dragonfly algorithm. Based on the dragonfly algorithm, the improved dragonfly algorithm introduces an adaptive inertia weight adjustment strategy to improve the global search capability and integrates the dynamic neighborhood update mechanism to enhance the local search accuracy:

[0117]

[0118] Among them, P represents the performance evaluation result, L represents the total number of layers of the health risk assessment model, Q represents the number of candidate groups of the structural parameter candidate set, and W (l) represents the weight initialization vector of the lth layer, represents the structural parameter of the qth candidate group in the lth layer, l represents the layer number of the health risk assessment model, q represents the number of the qth candidate group in the candidate structural parameter, r represents the number of the test sample, and t represents the number of the output dimension;

[0119] S46. Filter out the structural parameters with the highest performance evaluation results and generate the optimal structural parameter set for the health risk assessment model.

[0120] This paper utilizes an improved dragonfly algorithm, combined with adaptive inertia weights and a dynamic neighborhood mechanism, to achieve global and local coordinated optimization of the structural parameters of health risk assessment models. By allocating local and global search strategies within the dragonfly model, the efficiency and accuracy of parameter optimization are improved. Centralized training and distributed execution ensure optimal parameter selection, significantly enhancing model performance and health risk prediction accuracy, providing efficient and scientific algorithmic support for intelligent health management.

[0121] In this embodiment, the S5 specifically includes:

[0122] S51, inputting the optimal structural parameter set into the health risk assessment model to initialize the health risk assessment model;

[0123] S52. Input the training data set into the health risk assessment model in batches, perform forward propagation, calculate the error between the output result and the true label, adjust the health risk assessment model parameters according to the error, and obtain the trained health risk assessment model parameters and health risk assessment model prediction results;

[0124] S53. Based on the trained health risk assessment model parameters, use the test data set to verify the performance of the health risk assessment model, and output the predicted output of the health risk assessment model on the test data set;

[0125] S54. Compare the predicted output of the health risk assessment model on the test data set with the real label data, and generate performance evaluation index output based on the preset performance evaluation criteria:

[0126]

[0127] Among them, U is the performance evaluation index output, R is the total number of test samples, T is the total number of output dimensions, O is the total number of test samples, r,t The model prediction result of the r-th test sample in the t-th output dimension, Y r,t The rth test sample is the true label data of the tth output dimension, and P represents the performance evaluation result;

[0128] S55. Jointly analyze the performance evaluation index output and the optimal structural parameter set to determine whether the performance of the current health risk assessment model meets the deployment requirements, and output the final model performance evaluation results;

[0129] S56. Output the final health risk assessment model based on the final model performance evaluation results.

[0130] This invention applies an optimal set of structural parameters to a health risk assessment model, combining batch training, forward propagation, and error feedback to achieve efficient optimization of model parameters. Based on test set performance verification and comparison with real-world labels, it quantifies and outputs multi-dimensional performance evaluation metrics, effectively ensuring the scientificity and accuracy of model predictions. Jointly analyzing performance and parameters ensures that the final model meets actual deployment requirements, significantly improving the reliability and application value of health risk assessment.

[0131] In this embodiment, S6 specifically includes:

[0132] S61. Import the final health risk assessment model into the health risk assessment and early warning system;

[0133] S62, the health risk assessment and early warning system receives newly acquired individual health data information in real time and generates a new health data set;

[0134] S63, inputting the new health data set into the deployed final health risk assessment model, performing health risk analysis calculations, and obtaining individual health risk analysis results;

[0135] S64. Generate individual health risk classification data based on the health risk analysis results;

[0136] S65. Jointly store the individual health risk classification data and the new health data set to form a traceable health risk assessment record and output the final individual health risk level;

[0137] S66. Push the final individual health risk level to the front end of the health risk assessment and early warning system for query by individual users, health managers and doctors;

[0138] This invention implements real-time analysis and risk grading of new health data by deploying the final health risk assessment model within the health risk assessment and early warning system. Combining individual health risk grading with data traceability and storage ensures the accuracy and traceability of health risk assessment results. The system supports front-end push notifications, allowing individual users, health managers, and doctors to access and query health risk levels in real time, improving the intelligence and responsiveness of health management.

[0139] In this embodiment, the S7 specifically includes:

[0140] S71. Input the final individual health risk level into the health risk threshold determination unit of the health risk assessment and early warning system;

[0141] S72. In the health risk threshold determination unit, a preset risk threshold standard is called to perform a risk threshold comparison on the input final individual health risk level to generate a risk threshold determination result;

[0142] S73. Input the risk threshold determination result into the health risk warning generation unit. If the final individual health risk level exceeds the preset risk threshold, a warning message is automatically generated, an alarm is issued, and the on-site staff is notified. The data is forwarded to the doctor in the relevant department for diagnosis. The health manager updates the user's health management plan and promotion plan based on the doctor's advice, and regularly tracks and feeds back to the system for record.

[0143] S74. If the final individual health risk level does not exceed the preset risk threshold and the user belongs to the sub-healthy group, the health manager will develop a complete health management plan and promotion plan for the user.

[0144] This invention utilizes a health risk threshold determination unit to automatically compare thresholds and manage individual health risk levels. The system intelligently generates early warning information based on risk levels, enabling timely diagnosis by doctors, dynamic development and updating of management plans by health managers, and comprehensive tracking and feedback. This approach improves the timeliness and relevance of health risk warnings, facilitates personalized health interventions and scientific decision-making, and significantly enhances the intelligence of health management services.

[0145] refer to Figure 3 , a health risk assessment and early warning system based on multi-source data analysis, including the following modules:

[0146] Health data acquisition module, used to collect multi-source health data;

[0147] The health feature fusion module performs feature concatenation and normalization on the basic health indicator set, time series feature set, and unstructured feature set to generate a fused feature set;

[0148] The health risk assessment module summarizes and processes the health risk prediction value sequence and outputs the preliminary health risk assessment results;

[0149] Improve the Dragonfly algorithm module to perform coordinated optimization of structural parameters through centralized training and distributed execution to generate the optimal parameter set;

[0150] The health risk assessment model optimization module outputs the final health risk assessment model based on the final model performance evaluation results;

[0151] The health risk assessment module receives and analyzes newly acquired health data in real time and outputs individual health risk levels;

[0152] The risk warning module is used to push warning information to doctors and health managers.

[0153] This invention utilizes multi-module collaboration to achieve multi-source collection, feature fusion, and normalization of health data, enhancing the comprehensiveness and accuracy of health risk assessments. The improved Dragonfly algorithm module enables efficient collaborative optimization of structural parameters and dynamically outputs the optimal model. The system supports real-time analysis of new data and risk warnings, automatically disseminating warning information to doctors and health managers, strengthening the intelligence and responsiveness of health management and improving individual health risk prevention and control capabilities.

[0154] Example 1:

[0155] In order to verify the feasibility of the present invention in implementation, the present invention was applied to the dynamic health risk assessment and intelligent early warning services for people at high risk of cardiovascular chronic diseases in the health management center of a large tertiary hospital throughout 2024. The center has long served residents with physical examinations, outpatient follow-up and chronic disease management needs, and is particularly concerned with the early identification and intervention of cardiovascular health risks in middle-aged and elderly people. The traditional health management process mainly relies on routine physical examinations and manual follow-up once a year, which makes it difficult to detect sudden risks such as fluctuations in health indicators and sudden death in a timely manner. To this end, the hospital cooperated with the team of the present invention to establish a dynamic health risk intelligent monitoring system for nearly 2,000 community residents with a higher risk of cardiovascular disease in the health management center. The data collection time range is from January to December 2024, covering all kinds of seasonal health fluctuations in spring, summer, autumn and winter throughout the year.

[0156] In actual implementation, managed individuals are equipped with mainstream smart wearable devices. The system automatically collects continuous physiological data such as heart rate, blood pressure, step count, sleep quality, and body temperature. This data is then combined with historical physical examination data from the hospital's electronic health records, blood biochemical indicators, past medical history, family genetic information, and daily living habits. All data is uploaded to a central database in real time, where the system automatically performs data cleaning, missing information completion, and anomaly detection. It integrates existing physical examination results with dynamic monitoring data to construct a multi-dimensional, time-series data profile of the individual.

[0157] In this scenario, the system uses the recursive neural network time series modeling algorithm proposed in this invention to perform dynamic feature extraction and risk trend analysis on individual health data. The model automatically learns the changing patterns of different health indicators in the time dimension, and can effectively capture small abnormal fluctuations in important physiological parameters such as heart rate and blood pressure. Through intelligent optimization algorithms, the system continuously adjusts model parameters based on historical data and current status to ensure high accuracy of risk assessment. The system automatically pushes warning information based on different risk levels, generates personalized health management recommendations for high-risk individuals, and simultaneously pushes them to responsible doctors and health managers to achieve closed-loop management of health risks.

[0158] Throughout 2024, a total of 1,843 high-risk residents were included in this intelligent health management system, with a male-to-female ratio of approximately 1:1, an age distribution between 43 and 78 years old, and an average age of 59 years old. The system automatically collects more than 150,000 pieces of data every day, and the total amount of health data collected throughout the year reached more than 54 million pieces. Through the dynamic data fusion and time series risk analysis of the present invention, a total of 189 high-risk health event warnings were intelligently identified throughout the year, including 83 warnings for abnormal heart rate fluctuations, 52 warnings for a sharp increase in blood pressure, 21 warnings for a sudden drop in sleep quality, 10 warnings for abnormal body temperature, and 23 warnings for abnormal ECG parameters. All warnings are automatically pushed to the responsible doctor and the person within 30 minutes. After receiving the warning, the doctor can intervene by phone as soon as possible to guide the patient to return for an offline visit or conduct personalized conditioning.

[0159] Table 1 Comparison of health risk assessment and early warning effects based on multi-source data analysis

[0160]

[0161]

[0162] Table 1 shows that 68.5% of individuals adjusted their medication or lifestyle promptly after receiving the warning, and their health indicators returned to a safe range within a week. Seven patients with suspected myocardial infarction were promptly sent to the hospital after receiving multiple consecutive warnings from the system, avoiding serious consequences. Compared with the control group, which only managed its health through manual follow-up and routine physical examinations during the same period in 2023, the underreporting rate of high-risk events dropped from 21.6% to 5.4%, and the early intervention rate for health events increased nearly threefold. At the same time, the system significantly reduced the follow-up pressure on medical staff, reducing the average number of manual telephone follow-up visits per person per month from 2.7 to 1.1, saving significant manpower and material resources. Furthermore, user satisfaction surveys show that over 90% of managed residents believe that the system can promptly detect health risks, enhancing their sense of security and proactive health management.

[0163] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. A health risk assessment method and early warning system based on multi-source data analysis, characterized in that: The steps include: S1. Collect multi-source health data through medical testing equipment and pre-process it to build a health data set; S2, extract key health indicators, sequence features and statistical features based on the health data set, and generate multi-source fusion feature vectors; S3. Based on multi-source fusion feature vectors, a recursive neural network model is used to construct a health risk assessment model. The individual health status is modeled as an intelligent agent with the ability to perceive time series characteristics and predict health risks. The model predicts the evolution of individual health risks over time and outputs preliminary health risk assessment results. S4. Based on the preliminary health risk assessment results, the improved dragonfly algorithm is used to optimize the structural parameters of the health risk assessment model. A local search and global search coordination mechanism is configured for each dragonfly. Based on the search paths and dynamic position update behaviors of all dragonflies, the structural parameters are coordinated and optimized through centralized training and distributed execution to generate the optimal parameter set. S5. Use the optimal parameter set to evaluate the performance of the health risk assessment model and output the final health risk assessment model; S6. Deploy the final health risk assessment model to the health risk assessment and early warning system, receive and analyze newly acquired health data in real time, and output individual health risk levels; S7. When the health risk assessment result exceeds the preset risk threshold, an early warning message is automatically generated and pushed to doctors in relevant departments and on-site staff.

2. A health risk assessment method and early warning system based on multi-source data analysis according to claim 1, characterized in that: The multi-source health data specifically includes personal basic information, physical examination data, physiological indicators, laboratory test results, medical imaging data, past medical history, family history, lifestyle information and medical testing equipment data.

3. A health risk assessment method and early warning system based on multi-source data analysis according to claim 1, characterized in that: The key health indicators specifically include blood pressure, blood sugar, blood lipids, body mass index, heart rate, liver and kidney function indicators, blood oxygen saturation, urine routine, and electrocardiogram parameters.

4. A health risk assessment method and early warning system based on multi-source data analysis according to claim 1, characterized in that: The S2 specifically includes: S21. Perform format normalization, time alignment, and data integrity verification on various types of raw health data in the health data set to obtain a normalized health data set; S22. Extracting basic health indicators from the structured health data in the normalized health data set using a feature engineering method to obtain a basic health indicator set; S23. Segment and perform statistical analysis on the time series health data in the normalized health data set by using a sliding window to extract time series features and obtain a time series feature set; S24. Automatically extracting features from the unstructured health data in the normalized health dataset using a deep feature encoding network to obtain an unstructured feature set; S25. Perform feature concatenation and normalization processing on the basic health indicator set, the time series feature set, and the unstructured feature set to generate a fused feature set; S26. Perform feature screening and dimensionality reduction processing on the fused feature set, and output a multi-source fused feature vector.

5. A health risk assessment method and early warning system based on multi-source data analysis according to claim 1, characterized in that: The S3 specifically includes: S31, performing time step expansion on the multi-source fusion feature vector to generate a time series feature sequence; S32. Initialize the structural parameters of the recursive neural network based on the time series feature sequence, configure the network structure according to the actual requirements of the health risk assessment task, and randomly initialize the weights and bias parameters of each layer to build a health risk assessment model: Among them, W (l) Represents the weight initialization vector of the lth layer, l represents the layer number of the recursive neural network, N represents the total amount of data used for weight initialization, i takes a value between 1 and N, and represents the i-th group number in the multi-source fusion feature vector group, M represents the feature dimension of each group of multi-source fusion feature vectors, j takes a value between 1 and M, and represents the specific j-th feature number in each group of multi-source fusion feature vectors, It represents the value of the jth feature in the i-th group of multi-source fusion feature vectors when the l-th layer weight is initialized; S33, inputting the time series feature sequence into the health risk assessment model, performing time series feature modeling on the individual health status, and generating a dynamic health status representation; S34. Based on the dynamic health status representation, a sequence of health risk prediction values ​​of the individual at each time step is generated through the output layer of the health risk assessment model; S35. Summarize and process the health risk prediction value sequence and output preliminary health risk assessment results.

6. A health risk assessment method and early warning system based on multi-source data analysis according to claim 1, characterized in that: The S4 specifically includes: S41. Using the improved dragonfly algorithm, the preliminary health risk assessment results are used as the basis for initializing the dragonfly population, setting the number of individuals in the dragonfly population, the search space range, and the initial values ​​of the structural parameters to achieve individual distribution initialization of the dragonfly algorithm; S42. For each initialized dragonfly, based on the preliminary health risk assessment results, configure a collaborative search mechanism of local search and global search, assign a local search strategy and a global search strategy to each dragonfly, clarify the behavioral rules of each dragonfly during the search process, and output the collaborative search configuration results. S43, using a local search mechanism, based on the current position and structural parameters of each dragonfly body, by adjacent position perturbations and parameter fine-tuning, generating a local search candidate parameter set for the dragonfly body, and outputting the local candidate parameter set; S44, using a global search mechanism to analyze the search paths and dynamic position update behaviors of all dragonflies, globally adjusting the positions of each dragonfly based on the distribution characteristics of all dragonflies, generating a global search candidate parameter set, and outputting the global candidate parameter set; S45. The local search candidate parameter set and the global search candidate parameter set are collaboratively integrated. Based on the performance evaluation criteria of the health risk assessment model, all candidate structural parameters are centrally trained and distributedly executed. The performance of each group of structural parameters in the health risk assessment task is evaluated to obtain the structural parameter performance evaluation results. The collaborative integration of local search and global search adopts the improved dragonfly algorithm. Based on the dragonfly algorithm, the improved dragonfly algorithm introduces an adaptive inertia weight adjustment strategy to improve the global search capability and integrates the dynamic neighborhood update mechanism to enhance the local search accuracy: Among them, P represents the performance evaluation result, L represents the total number of layers of the health risk assessment model, Q represents the number of candidate groups of the structural parameter candidate set, and W (l) represents the weight initialization vector of the lth layer, represents the structural parameter of the qth candidate group in the lth layer, l represents the layer number of the health risk assessment model, q represents the number of the qth candidate group in the candidate structural parameter, r represents the number of the test sample, and t represents the number of the output dimension; S46. Filter out the structural parameters with the highest performance evaluation results and generate the optimal structural parameter set for the health risk assessment model.

7. A health risk assessment method and early warning system based on multi-source data analysis according to claim 1, characterized in that: The S5 specifically includes: S51, inputting the optimal structural parameter set into the health risk assessment model to initialize the health risk assessment model; S52. Input the training data set into the health risk assessment model in batches, perform forward propagation, calculate the error between the output result and the true label, adjust the health risk assessment model parameters according to the error, and obtain the trained health risk assessment model parameters and health risk assessment model prediction results; S53. Based on the trained health risk assessment model parameters, use the test data set to verify the performance of the health risk assessment model, and output the predicted output of the health risk assessment model on the test data set; S54. Compare the predicted output of the health risk assessment model on the test data set with the real label data, and generate performance evaluation index output based on the preset performance evaluation criteria: Among them, U is the performance evaluation index output, R is the total number of test samples, T is the total number of output dimensions, O is the total number of test samples, r,t The model prediction result of the r-th test sample in the t-th output dimension, Y r,t The rth test sample is the true label data of the tth output dimension, and P represents the performance evaluation result; S55. Jointly analyze the performance evaluation index output and the optimal structural parameter set to determine whether the performance of the current health risk assessment model meets the deployment requirements, and output the final model performance evaluation results; S56. Output the final health risk assessment model based on the final model performance evaluation results.

8. The health risk assessment method and early warning system based on multi-source data analysis according to claim 1, characterized in that: The S6 specifically includes: S61. Import the final health risk assessment model into the health risk assessment and early warning system; S62, the health risk assessment and early warning system receives newly acquired individual health data information in real time and generates a new health data set; S63. Input the new health data set into the deployed final health risk assessment model, perform health risk analysis calculations, and obtain individual health risk analysis results; S64. Generate individual health risk classification data based on the health risk analysis results; S65. Jointly store the individual health risk classification data and the new health data set to form a traceable health risk assessment record and output the final individual health risk level; S66. Push the final individual health risk level to the front end of the health risk assessment and early warning system for individual users, health managers and doctors to query.

9. The health risk assessment method and early warning system based on multi-source data analysis according to claim 1, characterized in that: The S7 specifically includes: S71. Input the final individual health risk level into the health risk threshold determination unit of the health risk assessment and early warning system; S72. In the health risk threshold determination unit, a preset risk threshold standard is called to perform a risk threshold comparison on the input final individual health risk level to generate a risk threshold determination result; S73. Input the risk threshold determination result into the health risk warning generation unit. If the final individual health risk level exceeds the preset risk threshold, a warning message is automatically generated, an alarm is issued, and the on-site staff is notified. The data is forwarded to the doctor in the relevant department for diagnosis. The health manager updates the user's health management plan and promotion plan based on the doctor's advice, and regularly tracks and feeds back to the system for record. S74. If the final individual health risk level does not exceed the preset risk threshold and the user belongs to the sub-healthy group, the health manager will develop a complete health management plan and promotion plan for the user.

10. A health risk assessment and early warning system based on multi-source data analysis, which implements the health risk assessment method based on multi-source data analysis according to any one of claims 1 to 9, characterized in that: Includes the following modules: Health data acquisition module, used to collect multi-source health data; The health feature fusion module performs feature concatenation and normalization on the basic health indicator set, time series feature set, and unstructured feature set to generate a fused feature set; The health risk assessment module summarizes and processes the health risk prediction value sequence and outputs the preliminary health risk assessment results; Improve the Dragonfly algorithm module to perform coordinated optimization of structural parameters through centralized training and distributed execution to generate the optimal parameter set; The health risk assessment model optimization module outputs the final health risk assessment model based on the final model performance evaluation results; The health risk assessment module receives and analyzes newly acquired health data in real time and outputs individual health risk levels; The risk warning module is used to push warning information to doctors and health managers.

Images