An edge computing-based personalized health monitoring and early warning method

Abstract

The application discloses a kind of personalized health monitoring and early warning method based on edge computing, comprising the following steps: obtaining standardized health data sequence;Multi-dimensional feature analysis is carried out, and the corresponding health state feature set is constructed;Health state assessment is carried out using improved PanLUNA model, and the health state assessment result of user current time is obtained;Abnormal identification is carried out to the health state of user current, and abnormal identification result is obtained;Risk degree division is carried out, and the corresponding health warning information is generated;Health warning information is sent to user terminal and remote monitoring platform, and cloud health data record is formed;Improved PanLUNA model is updated, and is issued to edge computing node, and health state assessment and abnormal identification are realized, and the real-time and evaluation accuracy of health monitoring are improved.

Description

Technical Field

[0001] This invention relates to the field of health monitoring and early warning, and in particular to a personalized health monitoring and early warning method based on edge computing. Background Technology

[0002] With the continuous development of information technology, IoT technology, and smart wearable devices, health monitoring technology based on multi-source data has been widely applied in the fields of smart healthcare, telemedicine, and personal health management. Current technologies typically collect users' physiological parameters, such as heart rate, blood pressure, blood oxygen, body temperature, and exercise data, through various terminal devices. The collected data is then transmitted via network to a server or cloud platform for unified processing and analysis, thereby enabling the monitoring of the user's health status. Simultaneously, some existing technologies also incorporate data analysis models to assess the user's health status and issue warnings to the user or caregiver when abnormalities are detected, thus improving the level of intelligence in health management.

[0003] With the increasing number of health monitoring data sources and the continuous expansion of data scale, existing technologies have gradually revealed a series of shortcomings in practical applications. Firstly, regarding data processing architecture, most existing technologies adopt a centralized processing approach centered on the cloud, uploading all collected multi-source health data to the cloud for analysis and processing. This approach is easily affected by network bandwidth limitations and transmission delays when dealing with high-frequency, multi-dimensional continuous data, leading to data processing lags and difficulty in timely reflecting changes in the user's current health status, thus affecting the real-time nature and effectiveness of health alerts. Summary of the Invention

[0004] One objective of this invention is to propose a personalized health monitoring and early warning method based on edge computing. This invention fully utilizes multi-source health data acquisition technology, edge computing processing mechanism, and improved PanLUNA model to standardize user health data, perform multi-dimensional feature analysis and health status assessment, and combine anomaly identification results to achieve risk level classification and graded early warning. At the same time, the model is continuously updated and distributed through a cloud server to form a dynamic optimization mechanism of cloud-edge collaboration, which has the advantages of high real-time performance, high assessment accuracy, strong personalization and timely early warning response.

[0005] A personalized health monitoring and early warning method based on edge computing according to an embodiment of the present invention includes the following steps: Collect multi-source health data corresponding to users and send the multi-source health data to edge computing nodes. The edge computing nodes preprocess the multi-source health data to obtain a standardized health data sequence. The edge computing nodes perform multidimensional feature analysis on the standardized health data sequence to extract health status features of changes in the user's physiological state and construct the corresponding health status feature set. The health status feature set is input into the improved PanLUNA model to assess the health status and obtain the user's current health status assessment result. The edge computing nodes identify anomalies in the user's current health status based on the health status assessment results, and obtain the anomaly identification results. Based on the anomaly identification results, the user's current health status is classified into risk levels, and corresponding health warning information is generated based on different risk levels; Health warning information is sent to user terminals and remote monitoring platforms, and standardized health data sequences, health status feature sets, health status assessment results and abnormal identification results are uploaded to cloud servers for storage, forming cloud health data records; Based on cloud-based health data records, the improved PanLUNA model is updated and distributed to edge computing nodes.

[0006] Optionally, wearable devices and physiological sensors deployed at the user end can be used to continuously monitor the user's physiological state. During the data collection process, various types of physiological data are collected according to a preset sampling frequency, and each data point is timestamped using a built-in clock module to form multi-source health data. The multi-source health data is then transmitted to the edge computing node in real time via a wireless communication module. After receiving the multi-source health data, the edge computing node performs preprocessing, which includes noise reduction, outlier removal, missing value completion, and time series alignment to obtain a standardized health data sequence in a unified format.

[0007] Optionally, the construction of the health status feature set specifically includes: The standardized health data sequence is divided into windows according to time order to obtain the time window data corresponding to the current health status assessment period; Statistical features are extracted from the data in each time window to obtain a set of statistical features; Extract time-series features from the data in each time window to obtain a set of time-series features; Frequency domain features are extracted from the data of each time window to obtain a set of frequency domain features; The statistical feature set, time series feature set, and frequency domain feature set are concatenated to form the health status feature set.

[0008] Optionally, obtaining the health status assessment results specifically includes: The improved PanLUNA model is used to assess health status by inputting a set of health status features. The improved PanLUNA model includes an individual baseline anchoring module, a time-frequency coupling mapping module, an edge constraint gating module, and a dual-reference offset evaluation module. The improvements of the improved PanLUNA model are as follows: the traditional PanLUNA model uses a fixed threshold or a single time series analysis method, while the improved PanLUNA model establishes health status assessment based on the user's individual historical features by introducing an individual baseline anchoring module. The time-frequency coupling mapping module performs cross-correlation modeling of statistical features, time series features, and frequency domain features. The edge constraint gating module incorporates data arrival delay, computational load status, and the completeness of health data into the calculation process, performs resource-aware path pruning on the feature transmission path, and performs joint offset calculation on the individual health baseline representation and the stable health reference representation through the dual-reference offset evaluation module. Under a unified evaluation framework, it comprehensively represents the degree of deviation of the user's current health status from the individual's normal state and the degree of deviation from the stable health level. In the individual baseline anchoring module, the historical health status features corresponding to the user's historical health data are retrieved, an individual health baseline representation corresponding to the current user is constructed, and baseline alignment processing is performed on each health status feature in the health status feature set to obtain the baseline-aligned health status feature set. In the time-frequency coupling mapping module, the health status feature set after baseline alignment is cross-correlated. Based on the synchronous change relationship between statistical features and time-series features, the fluctuation transmission relationship between time-series features and frequency domain features, and the distribution response relationship between statistical features and frequency domain features, a coupled state representation that characterizes the user's current physiological state change relationship is generated. In the edge constraint gating module, the data arrival delay, computing load status and health data integrity of the edge computing node are obtained, path gating control quantity is generated, and resource-aware path pruning processing is performed on the feature transmission path in the coupled state representation according to the path gating control quantity. Feature transmission paths that match the current edge resource state are retained, and feature transmission paths that do not match the current edge resource state are suppressed to obtain the edge adaptive state table. In the dual-reference offset evaluation module, the edge adaptive state representation is offset against the individual health baseline representation to obtain the individual offset and the stable offset. The individual offset and the stable offset are then jointly processed to generate the current health status evaluation value, thus forming the user's current health status evaluation result.

[0009] Optionally, obtaining the anomaly identification result specifically includes: The health status assessment results of consecutive moments in chronological order are compiled to form a health status assessment sequence. Extracting features of health status changes based on health status assessment sequences; Extracting health status fluctuation characteristics based on health status assessment sequences; By combining the characteristics of changes in health status with the characteristics of fluctuations in health status, an abnormality judgment index is formed. The anomaly detection index is used to identify anomalies in the user's current health status, and the anomaly detection results are obtained.

[0010] Optionally, the generation of the health warning information specifically includes: An anomaly identification sequence is formed based on the anomaly identification results at the current time and in consecutive time steps; Extracting persistent anomaly features based on anomaly identification sequences; Extracting anomaly frequency features based on anomaly identification sequences; Obtain the current health status assessment result, and fuse the health status assessment result with the abnormality persistence characteristics and abnormality frequency characteristics to obtain the corresponding risk level value; The risk level value is compared with the preset risk classification threshold to determine the risk level corresponding to the user's current health status and generate corresponding health warning information.

[0011] Optionally, the formation of the cloud-based health data record specifically includes: Health warning information is sent to user terminals and remote monitoring platforms. Standardized health data sequences, health status feature sets, health status assessment results and anomaly identification results are encapsulated and processed to form corresponding upload data packets. Generate data identification information for the uploaded data packets; The uploaded data packet is combined with the data identification information to form a set of data to be uploaded; Through the communication link between the edge computing node and the cloud server, the data set to be uploaded is uploaded to the cloud server, and the cloud server receives, parses and stores the data set to be uploaded, forming a cloud health data record associated with data identification information.

[0012] Optionally, the updating and distribution of the improved PanLUNA model specifically includes: The cloud-based health data records are organized and divided to form a training dataset for model updates and a validation dataset for model evaluation. The improved PanLUNA model is updated based on the training dataset. During the update process, the health status feature set in the training dataset is read one by one and input into the improved PanLUNA model to obtain the corresponding health status assessment results. The health status assessment results are compared with the existing health status assessment results in the training dataset. The parameters of the improved PanLUNA model are adjusted according to the comparison results to complete the model update process. After completing the model update process, the updated improved PanLUNA model is validated based on the validation dataset to obtain the updated health status assessment results, which are then compared with the corresponding health status assessment results to obtain the model evaluation results. The updated and improved PanLUNA model is judged based on the model evaluation results. When the difference between the updated health status assessment result and the corresponding health status assessment result is smaller than the difference before the update, the updated and improved PanLUNA model is determined as the target model, and the corresponding model update file is generated. The model update file is distributed to the edge computing node through the communication link between the cloud server and the edge computing node.

[0013] The beneficial effects of this invention are: This invention preprocesses and standardizes multi-source health data at edge computing nodes, enabling health data from different sources and formats to form a unified, standardized health data sequence. This effectively reduces the impact of data noise and inconsistencies, improving the stability and reliability of subsequent data analysis. Compared to existing technologies that rely on centralized cloud processing, this invention moves key processing steps to the edge, significantly reducing data transmission latency and improving the real-time performance of health status analysis. This allows users' current health status to be perceived and processed promptly, thereby improving the overall system's response efficiency.

[0014] In terms of health status analysis, this invention constructs a health status feature set by performing multi-dimensional feature analysis on standardized health data sequences, and combines this with an improved PanLUNA model for health status assessment, thus more comprehensively reflecting changes in the user's physiological state. Compared with existing technologies that use single features or general models, this invention has stronger adaptability in both feature representation and model evaluation, resulting in more refined and personalized assessment results, thereby effectively improving the accuracy of health status assessment.

[0015] This invention, building upon anomaly identification, introduces a mechanism for analyzing persistent and frequency features based on anomaly identification sequences to comprehensively characterize abnormal user health status at the temporal level. By comprehensively analyzing the duration and frequency of anomalies, it achieves deep identification of abnormal states, effectively avoiding the misjudgments or omissions caused by relying solely on single-moment judgments in existing technologies, thereby improving the reliability and stability of anomaly identification. Furthermore, by fusing health status assessment results with anomaly features, it achieves a fine-grained classification of risk levels and generates corresponding health warning information based on these risk levels. This makes the warning results more targeted and hierarchical, enhancing the practical value of the warning information.

[0016] This invention enhances the timeliness and collaboration of health management by promptly sending health warning information to user terminals and remote monitoring platforms, thereby providing simultaneous reminders to users and relevant monitoring personnel. Simultaneously, standardized health data sequences, health status feature sets, health status assessment results, and anomaly identification results are uniformly uploaded to a cloud server for associated storage, forming structured cloud-based health data records. This provides comprehensive data support for subsequent data analysis and model optimization.

[0017] Regarding model updates, this invention constructs training and validation datasets based on cloud-based health data records, continuously updates and validates the improved PanLUNA model, and then distributes the updated model to edge computing nodes, enabling dynamic iterative optimization of the model at the edge. Compared to existing technologies that rely on manual or offline updates, this invention can continuously optimize model performance based on the latest data, ensuring the model remains highly adaptable to the user's health status, thereby further improving the accuracy and intelligence of the overall health monitoring and early warning system. Attached Figure Description

[0018] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is an overall flowchart of a personalized health monitoring and early warning method based on edge computing proposed in this invention; Figure 2 This is a schematic diagram of the structure of the improved PanLUNA model, which is a personalized health monitoring and early warning method based on edge computing proposed in this invention. Figure 3 This is a schematic diagram illustrating the construction of health warning information for a personalized health monitoring and early warning method based on edge computing proposed in this invention. Detailed Implementation

[0019] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0020] refer to Figures 1-3 A personalized health monitoring and early warning method based on edge computing includes the following steps: Collect multi-source health data corresponding to users and send the multi-source health data to edge computing nodes. The edge computing nodes preprocess the multi-source health data to obtain a standardized health data sequence. The edge computing nodes perform multidimensional feature analysis on the standardized health data sequence to extract health status features of changes in the user's physiological state and construct the corresponding health status feature set. The health status feature set is input into the improved PanLUNA model to assess the health status and obtain the user's current health status assessment result. The edge computing nodes identify anomalies in the user's current health status based on the health status assessment results, and obtain the anomaly identification results. Based on the anomaly identification results, the user's current health status is classified into risk levels, and corresponding health warning information is generated based on different risk levels; Health warning information is sent to user terminals and remote monitoring platforms, and standardized health data sequences, health status feature sets, health status assessment results and abnormal identification results are uploaded to cloud servers for storage, forming cloud health data records; Based on cloud-based health data records, the improved PanLUNA model is updated and distributed to edge computing nodes.

[0021] In this embodiment, the user's physiological state is continuously monitored by wearable devices and physiological sensors deployed at the user end. Wearable devices include smart bracelets, smartwatches, or portable health monitoring terminals. Physiological sensors are used to collect the user's heart rate signal, blood oxygen saturation signal, body temperature signal, and exercise status signal. During the data collection process, various physiological data are collected according to a preset sampling frequency, and each data is timestamped by a built-in clock module to form multi-source health data. The multi-source health data is sent to the edge computing node in real time through a wireless communication module, which includes a Bluetooth module, a WiFi module, or a cellular communication module. After receiving the multi-source health data, the edge computing node performs preprocessing, which includes noise reduction, outlier removal, missing value completion, and time series alignment to obtain a standardized health data sequence in a unified format.

[0022] In this embodiment, the construction of the health status feature set specifically includes: The standardized health data sequence is divided into windows according to time order to obtain the time window data corresponding to the current health status assessment period; The process of obtaining time window data is as follows: the standardized health data sequence is arranged continuously in chronological order, and the window length and window sliding step are determined according to the time span of the current health status assessment. Data segments of continuous time periods are sequentially extracted from the standardized health data sequence using a sliding window method. Each time window corresponds to a segment of multi-source health data within a continuous time period, and the multi-source health data are kept time-aligned within the same time window. During the window division process, the window length is controlled to cover the complete health status assessment cycle, thus obtaining the time window data corresponding to the current health status assessment cycle. Statistical features are extracted from the data in each time window to obtain a set of statistical features; The process of obtaining the statistical feature set specifically includes: for each time window of data, traversing and processing each health data in the time window in chronological order, calculating the average level of each health data in the time window and the degree of change relative to the average level. The average level is obtained by summing all health data in the time window and dividing by the total number of health data. The degree of change is obtained by calculating the distribution of the difference between each health data and the average level, thus forming a statistical feature set of the overall distribution status of health data in the time window. Extract time-series features from the data in each time window to obtain a set of time-series features; The process of obtaining the time series feature set is as follows: For each time window of data, the health data is compared point by point in chronological order, the data changes between adjacent time points are calculated, the change trend and fluctuation state of each health data in the time dimension are obtained, and the health data within the time window is summarized based on the change trend and fluctuation state to form a time series feature set of the health data change pattern over time. Frequency domain features are extracted from the data of each time window to obtain a set of frequency domain features; The process of obtaining the frequency domain feature set is as follows: For each time window of data, the health data arranged in chronological order within the time window are subjected to frequency decomposition, transforming the original health data that changes over time into a data representation composed of different frequency components; during the frequency decomposition process, the distribution of each health data at different frequency changes is extracted to obtain the feature information of each health data in the low-frequency and high-frequency change parts. The low-frequency change part represents the overall trend of health data change, while the high-frequency change part represents the rapid fluctuation of health data; the intensity distribution of each frequency component is statistically analyzed to obtain the frequency domain feature set of the periodic change characteristics of health data within the time window; The statistical feature set, time series feature set, and frequency domain feature set are concatenated to form the health status feature set.

[0023] In this embodiment, obtaining the health status assessment results specifically includes: The improved PanLUNA model is used to assess health status by inputting a set of health status features. The improved PanLUNA model includes an individual baseline anchoring module, a time-frequency coupling mapping module, an edge constraint gating module, and a dual-reference offset evaluation module. The improvements of the improved PanLUNA model are as follows: the traditional PanLUNA model uses a fixed threshold or a single time series analysis method, while the improved PanLUNA model establishes health status assessment based on the user's individual historical features by introducing an individual baseline anchoring module. The time-frequency coupling mapping module performs cross-correlation modeling of statistical features, time series features, and frequency domain features. The edge constraint gating module incorporates data arrival delay, computational load status, and the completeness of health data into the calculation process, performs resource-aware path pruning on the feature transmission path, and performs joint offset calculation on the individual health baseline representation and the stable health reference representation through the dual-reference offset evaluation module. Under a unified evaluation framework, it comprehensively represents the degree of deviation of the user's current health status from the individual's normal state and the degree of deviation from the stable health level. After the individual baseline anchoring layer performs baseline alignment on the health status feature set, it outputs the baseline-aligned health status feature set to the time-frequency coupling mapping layer and passes the individual health baseline representation as reference information to the dual-reference offset evaluation layer. After the time-frequency coupling mapping layer performs cross-correlation modeling on the baseline-aligned features, it outputs the generated coupled state representation to the edge constraint gating layer. The edge constraint gating layer selectively controls the feature transmission path in the coupled state representation according to the running status of the edge computing nodes and outputs the processed edge adaptive state representation to the dual-reference offset evaluation layer. The dual-reference offset evaluation layer simultaneously receives the edge adaptive state representation, the individual health baseline representation, and the stable health reference representation, performs joint offset calculation on the three, and outputs the current health status evaluation result. In the individual baseline anchoring module, the historical health status features corresponding to the user's historical health data are retrieved, an individual health baseline representation corresponding to the current user is constructed, and baseline alignment processing is performed on each health status feature in the health status feature set to obtain the baseline-aligned health status feature set. The process of obtaining the baseline-aligned health status feature set is as follows: Each health status feature in the health status feature set is acquired, and historical health status features for the corresponding time period are retrieved from the user's historical health data. These historical health status features are then aggregated to form an individual health baseline representation. Each health status feature in the health status feature set is matched item by item with its corresponding individual health baseline representation. The offset between the current health status feature and the corresponding baseline value is calculated, and the health status features are corrected. Finally, the corrected health status features are recombined to form the baseline-aligned health status feature set. In the time-frequency coupling mapping module, the health status feature set after baseline alignment is cross-correlated. Based on the synchronous change relationship between statistical features and time-series features, the fluctuation transmission relationship between time-series features and frequency domain features, and the distribution response relationship between statistical features and frequency domain features, a coupled state representation that characterizes the user's current physiological state change relationship is generated. The specific process of generating the coupled state representation is as follows: The baseline-aligned health state feature set is grouped according to feature type, classifying statistical features, temporal features, and frequency domain features separately; based on time sequence, a time-by-time correspondence analysis is performed between statistical features and temporal features, extracting synchronous change relationships by comparing the direction and magnitude of change of the two types of features at the same time position; an association mapping is performed between temporal features and frequency domain features, matching the change trends in temporal features with the frequency distribution in frequency domain features to identify the response of different change trends in each frequency component, obtaining the fluctuation transmission relationship; a distribution alignment process is performed between statistical features and frequency domain features, extracting the distribution response relationship by analyzing the correspondence between the distribution of statistical features in different numerical intervals and the distribution of frequency domain features in different frequency ranges; the synchronous change relationship, fluctuation transmission relationship, and distribution response relationship are fused to form a unified feature association representation, and the feature association representation is structured to generate a coupled state representation of the user's current physiological state change relationship; In the edge constraint gating module, the data arrival delay, computing load status and health data integrity of the edge computing node are obtained, path gating control quantity is generated, and resource-aware path pruning processing is performed on the feature transmission path in the coupled state representation according to the path gating control quantity. Feature transmission paths that match the current edge resource state are retained, and feature transmission paths that do not match the current edge resource state are suppressed to obtain the edge adaptive state table. The process of acquiring data arrival delay, computing load status, and health data integrity is as follows: In the edge computing node, the received multi-source health data is time-stamped, and the data arrival delay is calculated based on the time difference between the data transmission time and the data arrival time; by monitoring the current processing task occupancy of the edge computing node, the computing load status, which characterizes the utilization of computing resources, is obtained. The computing load status includes the utilization level of processing units and task queuing status; the integrity of the received health data is checked; and by statistically analyzing the correspondence between the actual amount of data received and the amount of data that should have been received within a predetermined time range, the integrity of the health data is determined. The process of obtaining the edge adaptive state representation is as follows: A comprehensive analysis is performed on each feature transmission path in the coupled state representation, and the contribution of each feature transmission path is evaluated based on the path gating control quantity. The path gating control quantity is then correlated with data arrival delay, computational load status, and health data integrity to characterize the current edge computing node's operating state. Based on the path gating control quantity and the current edge computing node's operating state, resource-aware path pruning is performed on each feature transmission path. Feature transmission paths that are compatible with the current edge computing node's operating state are retained, while those that are incompatible are suppressed. During resource-aware path pruning, the feature information corresponding to the suppressed feature transmission paths is weakened or removed, while the feature information corresponding to the retained feature transmission paths is strengthened. Finally, the feature information corresponding to each feature transmission path after resource-aware path pruning is integrated to form an edge adaptive state representation adapted to the current edge computing node's operating state. In the dual-reference offset evaluation module, the edge adaptive state representation is offset against the individual health baseline representation to obtain the individual offset and the stable offset. The individual offset and the stable offset are then jointly processed to generate the current health status evaluation value, thus forming the user's current health status evaluation result. The formation process of the current health status assessment value is as follows: Each feature in the edge-adaptive state representation is compared item by item with the corresponding feature in the individual health baseline representation to calculate the individual offset, reflecting the degree of deviation of the user's current state from their individual normal state; each feature in the edge-adaptive state representation is compared item by item with the corresponding feature in the pre-constructed stable health reference representation to calculate the stable offset, reflecting the degree of deviation of the user's current state from the stable health level; the individual offset and stable offset are normalized, and based on the relative magnitude and trend of their changes, they are jointly weighted. When both the individual offset and stable offset increase simultaneously, their contribution to the current health status assessment value is enhanced; when their trends are inconsistent, their respective weights are adjusted according to their relative magnitude. The result of the joint weighting is then output as the current health status assessment value.

[0024] In this embodiment, obtaining the anomaly identification result specifically includes: The health status assessment results of consecutive moments in chronological order are compiled to form a health status assessment sequence. Extracting features of health status changes based on health status assessment sequences; The specific process of extracting health status change characteristics is as follows: the health status assessment results in the health status assessment sequence are compared point by point in chronological order to obtain the changes in health status assessment results between adjacent time points, and the changes in consecutive adjacent time points are summarized to characterize the trend of health status assessment results over time. The trend of change is summarized and organized to obtain the change characteristics of the health status from stable to changing or from changing to stable, thus forming health status change characteristics. Extracting health status fluctuation characteristics based on health status assessment sequences; The extraction process of health status fluctuation characteristics is as follows: the health status assessment results in the health status assessment sequence are analyzed as a whole within a predetermined time range to obtain the deviation of each health status assessment result from the overall level within the time range, and the deviation is summarized to characterize the degree of fluctuation of the health status assessment results within the time range; the degree of fluctuation is summarized and organized to obtain the characteristic information of the stability or intensity of the change of health status within the predetermined time range, thus forming the health status fluctuation characteristics. By combining the characteristics of changes in health status with the characteristics of fluctuations in health status, an abnormality judgment index is formed. The formation of the anomaly determination index is specifically as follows: Health status change characteristics and health status fluctuation characteristics are uniformly processed; based on the ability of health status change characteristics to represent the continuous changes in the health status assessment sequence and the ability of health status fluctuation characteristics to represent the intensity of fluctuations in the health status assessment sequence, the health status change characteristics and health status fluctuation characteristics are combined to jointly represent the degree of abnormal change in the health status assessment sequence; during the combined processing, when the health status change characteristics increase, their influence on the anomaly determination index is enhanced; when the health status fluctuation characteristics increase, their influence on the anomaly determination index is also enhanced; after completing the combined processing, the processing results are uniformly organized to form an anomaly determination index representing the degree of abnormality in the user's current health status. The user's current health status is identified as abnormal based on the anomaly detection indicators, and the anomaly identification results are obtained. The anomaly identification results are obtained as follows: the changes of anomaly judgment indicators within the time range corresponding to the health status assessment sequence are analyzed; the stability of the user's current health status is judged based on the numerical trend and magnitude of the anomaly judgment indicators; when the anomaly judgment indicators show a continuous increase or are at a high level over consecutive time, it is determined that the user's current health status has an abnormal change; when the anomaly judgment indicators remain stable or are at a low level over consecutive time, it is determined that the user's current health status has not an abnormal change; and the corresponding anomaly identification results are formed based on the judgment results.

[0025] In this embodiment, the generation of health warning information specifically includes: An anomaly identification sequence is formed based on the anomaly identification results at the current time and in consecutive time steps; The specific process of forming the anomaly identification sequence is as follows: taking the current time as the benchmark, trace back a preset number of consecutive time moments, obtain the anomaly identification results corresponding to each time moment in sequence, and arrange them in chronological order. Place the anomaly identification result of the earliest time moment at the beginning of the sequence and the anomaly identification result of the current time moment at the end of the sequence to obtain an anomaly identification sequence that reflects the relationship of continuous time change. Extracting persistent anomaly features based on anomaly identification sequences; The extraction process of the abnormal persistence feature is as follows: In the abnormal identification sequence, starting from the abnormal identification result corresponding to the current time, the abnormal identification results of each previous time are traversed in reverse chronological order to determine whether each time is in an abnormal state. When the abnormal identification results of consecutive time are all in an abnormal state, the count is accumulated. When the first non-abnormal state appears, the traversal is terminated, and the number of consecutive time in an abnormal state from the current time to the termination time is determined as the abnormal persistence feature. Extracting anomaly frequency features based on anomaly identification sequences; The specific process of extracting the anomaly frequency feature is as follows: within the time window corresponding to the anomaly identification sequence, the anomaly identification results at each moment are traversed and counted one by one, the identification results in the abnormal state are accumulated and counted, and the total number of moments in the time window is recorded. After completing the traversal of all moments, the ratio of the total number of times the abnormal state occurs to the total number of moments in the time window is calculated to obtain the anomaly frequency feature. Obtain the current health status assessment result, and fuse the health status assessment result with the abnormality persistence characteristics and abnormality frequency characteristics to obtain the corresponding risk level value; The process of obtaining the risk level value is as follows: The abnormal persistence feature is normalized and proportionally converted to the time window length to obtain the proportion of the abnormal persistence feature within the time window; the abnormal frequency feature and the corresponding health status assessment result at the current moment are obtained; fixed weight coefficients are assigned to the proportion of the abnormal persistence feature, the abnormal frequency feature, and the health status assessment result, and weighted summation is performed according to preset weights, where each weight coefficient is a preset constant and the sum is 1; the weighted results are then accumulated to obtain the risk level value representing the user's current health risk level. The risk level value is compared with the preset risk classification threshold to determine the risk level corresponding to the user's current health status and generate corresponding health warning information. The process of forming health warning information is as follows: Risk level values ​​are compared sequentially with preset risk grading thresholds. The threshold range of the risk level value is determined in ascending order. When the risk level value is less than the preset risk grading threshold, it is classified as a low-risk level. When the risk level value is greater than or equal to the preset risk grading threshold corresponding to the previous risk level and less than the preset risk grading threshold corresponding to the next risk level, it is classified as a medium-risk level. When the risk level value is greater than or equal to the preset risk grading threshold, it is classified as a high-risk level. After determining the risk level, this risk level is used as the core identifier and combined with the current anomaly identification results and health status assessment results in a structured combination to generate the corresponding health warning information. The health warning information includes the risk level identifier, the corresponding risk level value, and the abnormal state information that triggered the risk level.

[0026] In this embodiment, the formation of cloud-based health data records specifically includes: Health warning information is sent to user terminals and remote monitoring platforms. Standardized health data sequences, health status feature sets, health status assessment results and anomaly identification results are encapsulated and processed to form corresponding upload data packets. The upload data packet is formed as follows: After obtaining standardized health data sequences, health status feature sets, health status assessment results, and anomaly identification results, the data is divided into fields and formatted uniformly. The data is then classified and organized according to data type and written sequentially into the same data carrier. During the writing process, corresponding data field identifiers and data length information are set for each type of data to distinguish different data contents. After all data is written, the data carrier is encapsulated as a whole to form an upload data packet containing multi-source health data and processing results. Generate data identification information for the uploaded data packets; The specific process of generating data identification information is as follows: After the upload data packet is encapsulated, the user identity information corresponding to the current data is retrieved from the edge computing node to obtain the user identifier, the device number or network identifier of the edge computing node executing the data processing task is read as the edge computing node identifier, and the current system time is obtained as the timestamp information; the user identifier, edge computing node identifier, and timestamp information are combined and encoded in a field-based manner to form data identification information that corresponds one-to-one with the upload data packet; The uploaded data packet is combined with the data identification information to form a set of data to be uploaded; The combination process is as follows: after generating the upload data packet and the corresponding data identification information, the data identification information is embedded into the upload data packet as index information, and the data identification information is bound to the upload data packet one-to-one through field association; after the embedding or attachment is completed, the combined data is organized into a unified structure to form a set of data to be uploaded containing identification information and business data. Through the communication link between the edge computing node and the cloud server, the data set to be uploaded is uploaded to the cloud server, and the cloud server receives, parses and stores the data set to be uploaded, forming a cloud health data record associated with the data identification information; The process of forming cloud-based health data records is as follows: After receiving the data set to be uploaded, the cloud server unpacks the data set, separates the data identification information from the upload data packet, and classifies and marks the data according to the user identifier, edge computing node identifier, and timestamp information in the data identification information; it parses the standardized health data sequence, health status feature set, health status assessment results, and anomaly identification results in the upload data packet, and extracts the corresponding content according to the data type; after completing the data parsing, the data identification information is used as an index field, associated and stored with the various types of health data obtained from the parsing, and written into the cloud database in chronological order to form a cloud-based health data record corresponding to the user at a specific time.

[0027] In this embodiment, the improvement of the PanLUNA model update and distribution specifically includes: The cloud-based health data records are organized and divided to form a training dataset for model updates and a validation dataset for model evaluation. The specific implementation method for data processing and partitioning is as follows: After acquiring cloud-based health data records, each cloud-based health data record is standardized and organized according to a unified data structure. The standardized health data sequence, health status feature set, health status assessment results, and anomaly identification results are aligned and their completeness is verified. Data records with missing key fields or anomalies are removed. After completing the data processing, the cloud-based health data records are sorted according to time order, and the data records within consecutive time periods are divided into two parts based on time sequence. The first part is used as the training dataset for model updates, and the second part is used as the validation dataset for model evaluation. The improved PanLUNA model is updated based on the training dataset. During the update process, the health status feature set in the training dataset is read one by one and input into the improved PanLUNA model to obtain the corresponding health status assessment results. The health status assessment results are compared with the existing health status assessment results in the training dataset. The parameters of the improved PanLUNA model are adjusted according to the comparison results to complete the model update process. The specific process of parameter tuning for the improved PanLUNA model is as follows: After reading the health status feature set from the training dataset one by one and inputting it into the improved PanLUNA model, the health status assessment result output by the model is obtained and compared item by item with the corresponding health status assessment result in the training dataset. Based on the difference between the two, the corresponding loss function value is constructed. Based on the magnitude and trend of the loss function value, the influence degree of each parameter in the improved PanLUNA model is determined, and each parameter is adjusted by increasing or decreasing in the direction that reduces the loss function value, so that the health status assessment result obtained when inputting the same health status feature set in the future gradually approaches the corresponding health status assessment result in the training dataset. After completing one round of processing of all training datasets, the above process of calculating and adjusting parameters based on the loss function is repeated until the loss function value corresponding to each data record converges to a stable range, thus completing the parameter tuning of the improved PanLUNA model. After completing the model update process, the updated improved PanLUNA model is validated based on the validation dataset to obtain the updated health status assessment results, which are then compared with the corresponding health status assessment results to obtain the model evaluation results. The process of obtaining the model evaluation results is as follows: After completing the model update process, the health status feature sets in the validation dataset are read one by one and input into the updated improved PanLUNA model to obtain the corresponding health status evaluation results. The health status evaluation results corresponding to each health status feature set in the validation dataset are retrieved. The two types of health status evaluation results for the same data record are compared item by item, the difference value corresponding to each data record is calculated, and the difference values ​​of all data records are summarized to obtain the overall difference level. Based on the distribution of the difference values, whether the difference in each data record is within the allowable range is statistically analyzed to form the difference statistics result. The overall difference level and the difference statistics result are combined as the model evaluation result. The updated and improved PanLUNA model is judged based on the model evaluation results. When the difference between the updated health status assessment result and the corresponding health status assessment result is smaller than the difference before the update, the updated and improved PanLUNA model is determined as the target model, and the corresponding model update file is generated. The model update file is distributed to the edge computing node through the communication link between the cloud server and the edge computing node.

[0028] Example 1: In a chronic disease management scenario in a city community, the community health service center provides continuous health monitoring services for people with hypertension and cardiovascular risk. The participants are 120 residents aged 45 to 75. Each resident wears a wearable device with heart rate, blood pressure, blood oxygen, and activity monitoring functions, and also uses a home blood pressure monitor for regular measurements. Traditionally, the data collected by these devices is mainly uploaded to a cloud platform via mobile network for centralized analysis. Community doctors need to view the analysis results in the backend system and make manual interventions. However, in actual operation, there are problems such as significant data upload delays, untimely anomaly identification, and a high false alarm rate. Especially during unstable networks or peak periods, some data cannot be uploaded in time, resulting in delayed health warnings and affecting the effectiveness of interventions.

[0029] In this embodiment, a personalized health monitoring and early warning method based on edge computing is introduced, deploying edge computing nodes in community gateway devices. Multi-source health data collected by the user's wearable device is first transmitted to the edge computing node, where data cleaning, outlier removal, and unified format conversion are performed locally to form a standardized health data sequence. Subsequently, a multi-dimensional feature analysis module comprehensively analyzes the user's heart rate fluctuations, blood pressure trends, and blood oxygen stability over continuous time periods, extracting health status features reflecting physiological changes and constructing a health status feature set.

[0030] In the health status assessment phase, the health status feature set is input into the improved PanLUNA model for processing, yielding the user's current health status assessment result. Compared to traditional threshold-based judgment methods, this model can incorporate user historical data features to achieve more individualized assessment results. During anomaly identification, the system not only identifies anomalies at the current moment but also forms anomaly identification sequences based on anomalies over consecutive time periods, further extracting anomaly persistence and frequency features. For example, if a user experiences consecutive heart rate abnormalities within 30 minutes, the system can identify it as a persistent anomaly, while occasional single anomalies are differentiated and processed accordingly.

[0031] In the risk level classification process, the health status assessment results are fused with the characteristics of abnormal persistence and abnormal frequency to obtain a risk level value, which is then compared with a risk grading threshold to determine the risk level. Corresponding health warning information is generated based on different risk levels and sent in real time to the user's mobile phone and the community doctor's monitoring platform via edge computing nodes. For example, when the system detects that a user experiences multiple episodes of elevated blood pressure over a short period of time, which lasts for an extended period, it will determine the risk level as high and immediately push a warning message to the user to rest and notify a doctor for intervention.

[0032] Meanwhile, the system encapsulates standardized health data sequences, health status feature sets, health status assessment results, and anomaly identification results, and adds user identifiers, device identifiers, and time information before uploading them to a cloud server for unified storage. The cloud continuously updates the improved PanLUNA model based on accumulated data records and distributes the updated model to various edge computing nodes, thereby achieving continuous optimization of model performance.

[0033] After three months of continuous operation, compared with traditional cloud-based centralized processing solutions, the method of this invention shows significant advantages in terms of real-time performance, accuracy, and effectiveness of early warning. Specific data is shown in Table 1 below.

[0034] Table 1. Comparison of experimental data between edge computing-based health monitoring methods and traditional methods.

[0035] As shown in Table 1, the method of this invention significantly reduces the average data processing latency, from 8.6 seconds in the traditional method to 1.9 seconds, indicating that the introduction of edge computing nodes effectively improves the real-time performance of data processing. Regarding anomaly identification accuracy, it increases from 82.3% to 93.7%, while both false positive and false negative rates decrease significantly, demonstrating that the introduction of anomaly persistence and frequency characteristics makes anomaly identification more accurate and reliable. In terms of early warning response time, it is shortened from 10.2 seconds to 2.4 seconds, enabling users to obtain health warning information more promptly, thereby improving the intervention effect.

[0036] Furthermore, the timeliness rate of user intervention increased from 76.8% to 91.5%, indicating that the tiered early warning mechanism can effectively guide users and monitoring personnel to take timely measures. Regarding model updates, the method of this invention, through a cloud-edge collaboration mechanism, shortens the model update cycle from 30 days to 7 days, enabling the model to adapt more quickly to changes in user health status, thereby continuously improving assessment accuracy.

[0037] In summary, this embodiment fully verifies the significant advantages of the present invention in improving the real-time performance of health monitoring, enhancing the accuracy of anomaly identification, and improving the effectiveness of early warning. It effectively solves the problems of large data processing delays, inaccurate anomaly identification, and untimely early warning in the prior art.

[0038] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A personalized health monitoring and early warning method based on edge computing, characterized in that, Includes the following steps: Collect multi-source health data corresponding to users and send the multi-source health data to edge computing nodes. The edge computing nodes preprocess the multi-source health data to obtain a standardized health data sequence. The edge computing nodes perform multidimensional feature analysis on the standardized health data sequence to extract health status features of changes in the user's physiological state and construct the corresponding health status feature set. The health status feature set is input into the improved PanLUNA model to assess the health status and obtain the user's current health status assessment result. The edge computing nodes identify anomalies in the user's current health status based on the health status assessment results, and obtain the anomaly identification results. Based on the anomaly identification results, the user's current health status is classified into risk levels, and corresponding health warning information is generated based on different risk levels; Health warning information is sent to user terminals and remote monitoring platforms, and standardized health data sequences, health status feature sets, health status assessment results and abnormal identification results are uploaded to cloud servers for storage, forming cloud health data records; Based on cloud-based health data records, the improved PanLUNA model is updated and distributed to edge computing nodes.

2. The personalized health monitoring and early warning method based on edge computing according to claim 1, characterized in that, Wearable devices and physiological sensors deployed at the user end continuously monitor the user's physiological state. During the data collection process, various types of physiological data are collected according to a preset sampling frequency, and each data point is timestamped by a built-in clock module to form multi-source health data. The multi-source health data is sent to the edge computing node in real time through a wireless communication module. After receiving the multi-source health data, the edge computing node performs preprocessing, which includes noise reduction, outlier removal, missing value completion, and time series alignment, to obtain a standardized health data sequence in a unified format.

3. The personalized health monitoring and early warning method based on edge computing according to claim 1, characterized in that, The construction of the health status feature set specifically includes: The standardized health data sequence is divided into windows according to time order to obtain the time window data corresponding to the current health status assessment period; Statistical features are extracted from the data in each time window to obtain a set of statistical features; Extract time-series features from the data in each time window to obtain a set of time-series features; Frequency domain features are extracted from the data of each time window to obtain a set of frequency domain features; The statistical feature set, time series feature set, and frequency domain feature set are concatenated to form the health status feature set.

4. The personalized health monitoring and early warning method based on edge computing according to claim 1, characterized in that, The results of the health status assessment are obtained specifically in the following ways: The improved PanLUNA model is used to assess health status by inputting a set of health status features. The improved PanLUNA model includes an individual baseline anchoring module, a time-frequency coupling mapping module, an edge constraint gating module, and a dual-reference offset evaluation module. The improvements of the improved PanLUNA model are as follows: the traditional PanLUNA model uses a fixed threshold or a single time series analysis method, while the improved PanLUNA model establishes health status assessment based on the user's individual historical features by introducing an individual baseline anchoring module. The time-frequency coupling mapping module performs cross-correlation modeling of statistical features, time series features, and frequency domain features. The edge constraint gating module incorporates data arrival delay, computational load status, and the completeness of health data into the calculation process, performs resource-aware path pruning on the feature transmission path, and performs joint offset calculation on the individual health baseline representation and the stable health reference representation through the dual-reference offset evaluation module. Under a unified evaluation framework, it comprehensively represents the degree of deviation of the user's current health status from the individual's normal state and the degree of deviation from the stable health level. In the individual baseline anchoring module, the historical health status features corresponding to the user's historical health data are retrieved, an individual health baseline representation corresponding to the current user is constructed, and baseline alignment processing is performed on each health status feature in the health status feature set to obtain the baseline-aligned health status feature set. In the time-frequency coupling mapping module, the health status feature set after baseline alignment is cross-correlated. Based on the synchronous change relationship between statistical features and time-series features, the fluctuation transmission relationship between time-series features and frequency domain features, and the distribution response relationship between statistical features and frequency domain features, a coupled state representation that characterizes the user's current physiological state change relationship is generated. In the edge constraint gating module, the data arrival delay, computing load status and health data integrity of the edge computing node are obtained, path gating control quantity is generated, and resource-aware path pruning processing is performed on the feature transmission path in the coupled state representation according to the path gating control quantity. Feature transmission paths that match the current edge resource state are retained, and feature transmission paths that do not match the current edge resource state are suppressed to obtain the edge adaptive state table. In the dual-reference offset evaluation module, the edge adaptive state representation is offset against the individual health baseline representation to obtain the individual offset and the stable offset. The individual offset and the stable offset are then jointly processed to generate the current health status evaluation value, thus forming the user's current health status evaluation result.

5. A personalized health monitoring and early warning method based on edge computing according to claim 1, characterized in that, The anomaly identification results are obtained specifically in the following ways: The health status assessment results of consecutive moments in chronological order are compiled to form a health status assessment sequence. Extracting features of health status changes based on health status assessment sequences; Extracting health status fluctuation characteristics based on health status assessment sequences; By combining the characteristics of changes in health status with the characteristics of fluctuations in health status, an abnormality judgment index is formed. The anomaly detection index is used to identify anomalies in the user's current health status, and the anomaly detection results are obtained.

6. The personalized health monitoring and early warning method based on edge computing according to claim 1, characterized in that, The generation of the health warning information specifically includes: An anomaly identification sequence is formed based on the anomaly identification results at the current time and in consecutive time steps; Extracting persistent anomaly features based on anomaly identification sequences; Extracting anomaly frequency features based on anomaly identification sequences; Obtain the current health status assessment result, and fuse the health status assessment result with the abnormality persistence characteristics and abnormality frequency characteristics to obtain the corresponding risk level value; The risk level value is compared with the preset risk classification threshold to determine the risk level corresponding to the user's current health status and generate corresponding health warning information.

7. The personalized health monitoring and early warning method based on edge computing according to claim 1, characterized in that, The formation of the cloud-based health data records specifically includes: Health warning information is sent to user terminals and remote monitoring platforms. Standardized health data sequences, health status feature sets, health status assessment results and anomaly identification results are encapsulated and processed to form corresponding upload data packets. Generate data identification information for the uploaded data packets; The uploaded data packet is combined with the data identification information to form a set of data to be uploaded; Through the communication link between the edge computing node and the cloud server, the data set to be uploaded is uploaded to the cloud server, and the cloud server receives, parses and stores the data set to be uploaded, forming a cloud health data record associated with data identification information.

8. The personalized health monitoring and early warning method based on edge computing according to claim 1, characterized in that, The update and distribution of the improved PanLUNA model specifically includes: The cloud-based health data records are organized and divided to form a training dataset for model updates and a validation dataset for model evaluation. The improved PanLUNA model is updated based on the training dataset. During the update process, the health status feature set in the training dataset is read one by one and input into the improved PanLUNA model to obtain the corresponding health status assessment results. The health status assessment results are compared with the existing health status assessment results in the training dataset. The parameters of the improved PanLUNA model are adjusted according to the comparison results to complete the model update process. After completing the model update process, the updated improved PanLUNA model is validated based on the validation dataset to obtain the updated health status assessment results, which are then compared with the corresponding health status assessment results to obtain the model evaluation results. The updated and improved PanLUNA model is judged based on the model evaluation results. When the difference between the updated health status assessment result and the corresponding health status assessment result is smaller than the difference before the update, the updated and improved PanLUNA model is determined as the target model, and the corresponding model update file is generated. The model update file is distributed to the edge computing node through the communication link between the cloud server and the edge computing node.

Images