Intelligent health early warning system based on cloud processing

Abstract

The invention discloses an intelligent health early warning system based on cloud processing, and relates to the field of intelligent early warning, and the system comprises the steps: comprehensively collecting the physiological, environment and behavior data of a user through a multi-mode collection module, and dynamically adjusting the collection frequency; after preprocessing and end-to-end encryption are completed on the user side edge equipment, the data are uploaded to the cloud for mixed storage and life cycle management; performing fusion analysis on the data by the system, mining internal association and extracting key feature vectors, and constructing a personalized dynamic health assessment model according to the internal association and the key feature vectors; the system calculates a comprehensive health risk score and a specific disease incidence probability based on a model, realizes graded early warning, and bidirectionally interacts and feeds back to a user and medical staff through an intelligent terminal. The method has the advantages that safety and high efficiency of data are guaranteed through edge-cloud cooperative processing and end-to-end encryption, accurate risk assessment and graded early warning are realized based on a personalized health model, and the method has early warning accuracy, use convenience and medical specialty.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of real-time early warning, in particular to an intelligent health early warning system based on cloud processing. BACKGROUND

[0002] With the acceleration of social rhythm and the high incidence of chronic diseases, the traditional passive and lagging health management mode has been difficult to meet the needs of modern disease prevention and control. At the same time, the popularity of wearable devices has generated a large amount of physiological parameters, but single and isolated data cannot effectively reveal the health risk trend.

[0003] The current health early warning method on the market is difficult to match the multi-dimensional and personalized health management needs. Data collection is mostly limited to single physiological indicators, lacking the coordinated collection of environmental and behavioral data, and the device adaptability is poor, the data format is not unified, which is easy to form fragmented data islands and cannot build a complete health portrait. Data processing mostly relies on cloud direct operation, lacking edge side preprocessing link, not only low transmission efficiency, but also privacy leakage risk of insufficient data encryption protection. The model construction mostly uses fixed threshold or general template, ignoring individual differences such as age, underlying disease, genetic history, weak generalization ability and lack of dynamic iteration mechanism, resulting in insufficient risk assessment accuracy. The early warning mechanism is mostly in the form of single prompt, not enough refined, and lacks multi-terminal adaptation ability, making it difficult to achieve differentiated and accurate touch. At the same time, most methods do not establish a closed loop of user feedback and medical end interaction, which cannot optimize the model according to the actual use, and the user compliance and medical professionalism need to be improved. SUMMARY

[0004] In order to perfect the existing system, an intelligent health early warning system based on cloud processing is provided, which collects multi-modal data, cooperates with edge-cloud, and ensures data security and efficiency through end-to-end encryption, realizes accurate risk assessment and graded early warning through personalized health model, adapts to multi-terminal interaction and supports closed loop optimization, and has early warning accuracy, convenience and medical professionalism.

[0005] To achieve the above purpose, the technical scheme adopted by the present application is:

[0006] The intelligent health early warning system based on cloud processing comprises:

[0007] A multi-modal health data collection module: the module collects physiological, environmental and behavioral health data, and the collection frequency is dynamically adjusted based on user demand;

[0008] A data preprocessing module: deployed on the user side edge device, preprocessed through outlier rejection, data normalization and data compression algorithm, and encrypted through end-to-end double encryption mechanism for data transmission;

[0009] Cloud data storage and management module: deployed in cloud server cluster, using structured and unstructured mixed storage architecture, managing data through built-in data life cycle management mechanism;

[0010] Health data fusion analysis module: through multi-modal data fusion and feature extraction, constructing personal health spatio-temporal data set and mining potential correlations among physiology, environment and behavior, forming health feature vector with significant influence on health risk assessment, batch analyzing health data to generate phased health feature data;

[0011] Health model construction module: based on user phased health feature data, constructing dynamically adaptive health assessment model, built-in basic health assessment model library, forming personalized health model through adapting user individual physiological characteristics;

[0012] Risk assessment and grading warning module: based on health feature vector and personalized health model, calculating incidence probability of common health risk diseases, combining user's past medical history and family genetic history to determine risk weight coefficients, calculating comprehensive health risk score, and dividing warning levels according to comprehensive health risk score and single risk incidence probability;

[0013] Multi-terminal warning information interaction module: user side interaction supports smart watch, smart phone, tablet computer and home smart terminal, and medical staff side interaction is through medical institution special terminal, and user feedback is through evaluation of warning information.

[0014] Preferably, the multi-modal health data acquisition module specifically comprises:

[0015] Physiological data acquisition unit: adapting smart watch, blood glucose meter and sphygmomanometer, collecting core physiological indicators in real time through standardized interface, supporting dynamic adjustment of collection frequency, automatically caching data when device is disconnected, and supplementing transmission after reconnection;

[0016] Environmental data acquisition unit: integrating temperature and humidity, air quality sensors, collecting temperature and PM2.5 concentration parameters of user's environment in real time, and aligning collected data and physiological data by timestamp;

[0017] Behavioral data acquisition unit: associated with exercise APP and diet recording tool, providing user active input interface to obtain step count and diet structure behavior data;

[0018] Device adaptation and data coordination unit: providing standardized interface to adapt different brands and types of collection devices, and unifying data format.

[0019] Preferably, the data preprocessing module specifically comprises:

[0020] The data preprocessing unit adopts an edge computing architecture, an outlier elimination algorithm filters mis-triggered data and sensor fault data in the collection process, data normalization processing unifies the dimensions of different device indicators, and data compression processing reduces the data volume while retaining key information.

[0021] The encrypted transmission unit adopts an end-to-end double encryption mechanism, performs symmetric encryption on the preprocessed data through an AES-256 algorithm, generates an encrypted data block, performs asymmetric encryption on the encryption key, forms a key ciphertext, and synchronously transmits the encrypted data and the key ciphertext to the cloud through a security protocol.

[0022] The device coordination and state monitoring unit coordinates the working rhythm of wearable devices, home gateways, and mobile terminal edge devices, and monitors the running state of the edge devices in real time. If a device is disconnected, a data caching mechanism is triggered, and the cached data is automatically supplemented after the device is reconnected.

[0023] Preferably, the cloud data storage and management module specifically includes:

[0024] The data classification storage unit adopts a structured and unstructured hybrid architecture, stores structured data in a distributed relational database, organizes data according to a user ID-data type-collection time three-dimensional index structure, stores unstructured data in a distributed file system, and stores large files through data sharding technology.

[0025] The data life cycle management unit archives and compresses overage non-critical data through an internal data life cycle management mechanism, permanently stores core physiological data, and completes full backup through local backup and off-site disaster recovery dual strategies.

[0026] The standardized interface service unit provides a unified data calling interface, outputs standardized data for subsequent analysis modules, receives and processes data requests in real time, and realizes data interconnection through interfacing with other modules in the system.

[0027] Preferably, the health data fusion analysis module specifically includes:

[0028] The data correlation matching unit correlates and matches physiological data, environmental data, and behavior data in the same time period through a data correlation algorithm, and constructs a personal health space-time data set containing user health status and space-time scenarios.

[0029] The feature extraction and screening unit extracts features from the correlated personal health space-time data set through a CNN-LSTM hybrid model based on an attention mechanism, mines potential correlation features between physiological indicators, physiological indicators and environmental and behavior factors, screens out key features that have a significant impact on health risk assessment through a feature screening algorithm based on mutual information entropy, and forms a health feature vector.

[0030] Analysis mode management unit: support real-time analysis and batch analysis two modes, real-time analysis for core physiological indicators abnormal fluctuations in response to seconds capture health risk, batch analysis on the daily health data comprehensive analysis, generate stage health characteristics data.

[0031] Preferably, the health model construction module specifically comprises:

[0032] Basic model matching unit: management of built-in population general model library, covering different age groups, basic disease type of adaptive model. According to the age, sex, past medical history of initial information filled in when the user registers to automatically match the corresponding population general model;

[0033] Personalized model construction unit: based on the user stage health characteristics data, through the transfer learning algorithm to match the general model parameter fine-tuning, adapt to the physiological characteristics of individual users, generate personalized health model;

[0034] Model iteration and performance monitoring unit: every 7 days of health data fine-tuning parameters, every 30 days reconstruction model, real-time monitoring of model accuracy, recall rate, F1 score, when the accuracy is lower than the preset threshold, start parameter reoptimization.

[0035] Preferably, the risk assessment and grading early warning module specifically comprises:

[0036] Health risk calculation unit: the extracted health feature vector input personalized health model, calculate the incidence probability of common health risk of users, combined with the past medical history, family history to determine the risk weight coefficient, calculate the output comprehensive health risk score;

[0037] Grading early warning execution unit: based on the comprehensive health risk score is divided into four levels of early warning, one level only in the system background record, not active push, two level through the APP message push health tips, three level with APP pop-up window, short message dual push warning information, four level additional trigger voice broadcast and emergency contact person notice;

[0038] Early warning threshold configuration unit: based on the doctor's guidance to modify the early warning trigger threshold of various health risks, save the threshold configuration record, including adjustment time, adjustment basis and guidance doctor information.

[0039] Preferably, the multi-terminal early warning information interaction module specifically comprises:

[0040] User-side multi-terminal adaptation unit: adapt to smart watches, smartphones, and home smart terminal devices, and present information according to the characteristics of the devices. The smart watch presents the warning level, key abnormal indicators, and quick operations. The smartphone APP presents complete warning details, historical records, and health charts. The home smart terminal pushes information through voice broadcasting and screen display.

[0041] Medical staff-side terminal interaction unit: provides a data interface for medical institution terminals, pushes real-time three-level and above warning information of patients, and supports medical staff to view complete health data and risk assessment results of patients.

[0042] Feedback information processing unit: provides a user feedback entry, supports users to evaluate the accuracy of the warning and the practicability of the suggestion, classifies and arranges the feedback results according to the warning types, synchronizes to the dynamic health model construction module, and serves as the basis for model iteration and optimization.

[0043] Compared with the prior art, the advantages of the present application are that:

[0044] Through multi-modal acquisition, physiological, environmental, and behavioral data are fully covered and collected. Multiple devices are adapted, and the acquisition frequency is dynamically adjusted. Edge-side preprocessing and end-to-end double encryption are combined to improve transmission efficiency while ensuring data integrity and security. The cloud hybrid storage architecture and the life cycle management mechanism realize efficient storage and safe backup of data classification, and provide reliable support for subsequent analysis. Relying on the hybrid model based on the attention mechanism and the transfer learning algorithm, the potential correlation of multi-dimensional data is accurately mined, the dynamic and iterative personalized health model is constructed, the risk weight is optimized combined with the previous medical history and genetic history, and the risk assessment is more in line with individual characteristics. The grading warning mechanism and the multi-terminal adaptation design realize the differentiated and accurate access of warning information. At the same time, the closed loop is formed through user feedback and medical staff interaction, the model performance is continuously optimized, the convenience of users and the professionalism of medical treatment are taken into account, and the accuracy, timeliness, and practicability of health risk warning are comprehensively improved. BRIEF DESCRIPTION OF DRAWINGS

[0045] Figure 1 The system schematic diagram of the present application is shown in the figure;

[0046] Figure 2 The multi-modal health data acquisition module diagram of the present application is shown in the figure;

[0047] Figure 3 The data preprocessing module diagram of the present application is shown in the figure;

[0048] Figure 4 The cloud data storage and management module diagram of the present application is shown in the figure;

[0049] Figure 5 The health data fusion analysis module diagram of the present application is shown in the figure;

[0050] Figure 6 The health model construction module diagram proposed in the present application;

[0051] Figure 7 The risk assessment and grading early warning module diagram proposed in the present application;

[0052] Figure 8 The multi-terminal early warning information interaction module diagram proposed in the present application. DETAILED DESCRIPTION

[0053] The following description is used to disclose the present application so that those skilled in the art can implement the present application. The preferred embodiments in the following description are only as examples, and other obvious modifications can be thought of by those skilled in the art.

[0054] Referring to Figure 1 As shown, the intelligent health early warning system based on cloud processing includes:

[0055] Multi-modal health data acquisition module: the module collects three types of health data of physiology, environment and behavior with full coverage, and the collection frequency is dynamically adjusted based on user demand;

[0056] Data preprocessing module: deployed on user side edge device, preprocessed through outlier rejection, data normalization and data compression algorithm, and encrypted data transmission by using end-to-end double encryption mechanism;

[0057] Cloud data storage and management module: deployed in cloud server cluster, using structured and unstructured hybrid storage architecture, and managing data through built-in data life cycle management mechanism;

[0058] Health data fusion analysis module: through multi-modal data fusion and feature extraction, constructing personal health spatio-temporal data set and mining potential correlation among physiology, environment and behavior, forming health feature vector which has significant influence on health risk assessment, and generating phased health feature data by batch analyzing health data;

[0059] Health model construction module: based on user phased health feature data, constructing dynamically adaptive health assessment model, built-in basic health assessment model library, forming personalized health model by adapting user individual physiological characteristics;

[0060] Risk assessment and grading early warning module: based on health feature vector and personalized health model, calculating the incidence probability of common health risk diseases of the user, determining the risk weight coefficient of each item combined with the user's past medical history and family genetic history, calculating the comprehensive health risk score, and dividing the early warning level according to the comprehensive health risk score and the incidence probability of single risk;

[0061] Multi-terminal early warning information interaction module: user side interaction supports four types of devices, including smart watches, smart phones, tablet computers and home intelligent terminals, and the medical staff side interaction is through a medical institution special terminal, and the user feeds back through evaluation of the early warning information.

[0062] Referring to Figure 2 As shown in the figure, the multi-modal health data acquisition module specifically includes:

[0063] Physiological data acquisition unit: adapt to smart watches, blood glucose meters and sphygmomanometers, real-time acquisition of core physiological indicators through standardized interfaces, support dynamic adjustment of acquisition frequency, automatic caching of data when the device is disconnected, and data transmission after reconnection;

[0064] Environmental data acquisition unit: integrate temperature and humidity, air quality sensors, real-time acquisition of temperature and PM2.5 concentration parameters of the environment where the user is located, and alignment of collected data and physiological data according to time stamp;

[0065] Behavioral data acquisition unit: associated with exercise APPs and diet recording tools, provides a user-initiated input interface to obtain step count and diet structure behavior data;

[0066] Device adaptation and data coordination unit: provides standardized interface adaptation for different brands and types of collection devices, and unifies data format.

[0067] Specifically, physiological data acquisition: connected wearable devices and home medical devices monitor physiological indicators at a preset frequency, and after monitoring is completed, raw data is transmitted to the module in real time through Bluetooth or Wi-Fi; for devices that need to be operated manually, the device automatically uploads data to the module after the user completes the measurement; after the module receives the physiological data, it checks whether the data is missing, has format errors, etc.; if the data is complete and the format is correct, it directly enters the next step of temporary storage; if there is a missing, mark the missing time period and record; if the format is incorrect, send a prompt to the user that the data format of a certain device is abnormal, please re-measure and upload;

[0068] Environmental data acquisition: the module automatically activates temperature and humidity, air quality and ultraviolet sensors integrated in the home gateway or mobile terminal, the sensors collect surrounding environmental data at a preset frequency, and real-time judgment is made during the collection process to determine whether the sensors are working normally, and an environmental sensor abnormality reminder is triggered when a fault occurs; after collecting environmental data, the module automatically adds a time stamp and a geographic location marker to the data to ensure that each piece of environmental data corresponds to the specific time and location where the user is located;

[0069] Behavioral data acquisition: automatic data synchronization: the module establishes a real-time data channel with the associated exercise APPs and smart watches, and the step count, exercise duration, exercise type and other data monitored by the APPs or watches are synchronized to the module every 1 minute.

[0070] Referring to Figure 3 As shown in the data preprocessing module specifically includes:

[0071] Data preprocessing unit: using edge computing architecture, outlier rejection algorithm to filter the false trigger data, sensor fault data in the collection process, through data normalization processing to unify the dimension of different equipment indicators, through data compression processing, under the premise of retaining key information, reduce the data volume;

[0072] Encryption transmission unit: using end-to-end double encryption mechanism, through AES-256 algorithm to symmetrically encrypt the preprocessed data, generate encrypted data block, asymmetrically encrypt the encryption key, form the key ciphertext, through the security protocol to synchronize the encrypted data and the key ciphertext to the cloud;

[0073] Device coordination and state monitoring unit: coordinates the work rhythm of wearable devices, home gateways, mobile terminal edge devices, and monitors the running state of edge devices in real time. If the device is disconnected, the data caching mechanism is triggered. After the device is reconnected, the cached data is automatically supplemented.

[0074] Specifically, the module encapsulates each batch of compressed data as an independent data block, uses the AES-256 encryption algorithm to encrypt the data block, automatically generates a random 256-bit encryption key, and performs confusion and permutation operations on the data block through the key to generate an encrypted data block that cannot be directly interpreted. At the same time, the correspondence between the key and the data block is recorded to ensure accurate matching during subsequent decryption;

[0075] To prevent the leakage of AES keys during transmission, the module uses the RSA-2048 asymmetric encryption algorithm to encrypt the AES keys twice, pre-stores the public key of the cloud server, encrypts the AES keys with the public key, and generates key ciphertext that can only be decrypted by the cloud private key. After encryption, each batch of data corresponds to the combination of encrypted data block + key ciphertext, ensuring that data and keys are transmitted separately to improve security;

[0076] The module detects the current network state in real time, and preferentially selects 5G network with fast speed and high stability. If the 5G signal is weak, it automatically switches to Wi-Fi network. If it is in a network-free environment, it switches to Bluetooth transmission. The data transmission is not interrupted during the switching process, and the breakpoint resume mechanism is used to continue sending from the last transmission interruption position, avoiding data duplication or loss;

[0077] The module starts the confirmation-retransmission mechanism after data transmission, sends data to the cloud, and waits for the cloud to return the successful reception confirmation signal. If no confirmation is received within 5 seconds, the batch of data is automatically resent, with a maximum of 3 retries. If all 3 retries fail, the data is temporarily stored in the local cache area of the edge device, and the data is batched and supplemented in time sequence when the network is restored.

[0078] Referring to Figure 4 As shown in the figure, the cloud data storage and management module specifically includes:

[0079] Data classification storage unit: using a structured and unstructured mixed architecture, structured data is stored in a distributed relational database, and data is organized according to a user ID-data type-collection time three-dimensional index structure, unstructured data is stored in a distributed file system, and large files are stored by data sharding technology;

[0080] Data life cycle management unit: through the built-in data life cycle management mechanism, the overage non-critical data is archived and compressed, the core physiological data is permanently stored, and the full backup is completed by using the local backup and off-site disaster recovery double strategies;

[0081] Standardized interface service unit: providing a unified data calling interface, outputting standardized data for subsequent analysis modules, receiving and processing data requests in real time, and realizing data interconnection through interfacing with other modules in the system.

[0082] Specifically, the decrypted health data is classified: numerical physiological indicators such as heart rate and blood pressure, step count, and dietary calorie statistical behavior data, as well as warning records, service appointment information, and the like are marked as structured data; electrocardiogram waveform files, sleep monitoring original curve data, and health report PDF documents are marked as unstructured data; after classification, the corresponding storage architecture is automatically matched, structured data is directed to a MySQL distributed cluster, and unstructured data is allocated to a HDFS distributed file system;

[0083] Structured data storage establishes an index for structured data according to a user ID-data type-collection time three-dimensional structure, and the index information is stored synchronously to the cluster index node, ensuring that the data storage location can be quickly located during subsequent queries; according to the hash value of the user ID, the data is evenly distributed to different nodes of the MySQL cluster, avoiding excessive storage pressure on a single node; when storing, the data is organized according to daily partition rules, a new data partition is automatically created at 0 o'clock every day, and the read and write permissions of the historical data partition are retained;

[0084] Unstructured data storage divides large-volume unstructured data into multiple data shards according to a specification of 16MB-64MB, generates a unique identifier for each shard, and records the number and order of the shards; the shards are evenly distributed to different data nodes of the HDFS cluster, and the storage location and association of the shards are recorded in the NameNode node;

[0085] The stored data is classified, and core physiological data such as heart rate and blood pressure are marked as permanent storage level, environmental temperature and humidity and non-critical behavior data are marked as regular storage level, and temporary generated health data statistical report is marked as temporary storage level; for the data of the regular storage level, if the storage time exceeds 1 year, it is automatically compressed into ZIP format and archived to a low-cost storage node; for the data of the temporary storage level, it is automatically deleted after 30 days; for the data of the permanent storage level, the original format and read-write permission are kept, and there is no time limit for retention.

[0086] Referring to Figure 5 As shown in the figure, the health data fusion analysis module specifically includes:

[0087] The data correlation matching unit: through the data correlation algorithm, the physiological data, environmental data and behavior data in the same time period are correlated and matched, and a personal health space-time data set containing the user's health state and space-time scene is constructed;

[0088] The feature extraction and screening unit: through the CNN-LSTM hybrid model based on attention mechanism, the correlated personal health space-time data set is subjected to feature extraction, the potential correlation features between physiological indicators, physiological indicators and environmental and behavior factors are mined, the key features having significant influence on health risk assessment are screened out through the feature screening algorithm based on mutual information entropy, and a health feature vector is formed;

[0089] The analysis mode management unit: supports two modes of real-time analysis and batch analysis, the real-time analysis is for capturing health risks in seconds in response to abnormal fluctuations of core physiological indicators, and the batch analysis is for comprehensive analysis of daily health data to generate periodic health feature data.

[0090] Specifically, the module calls the built-in attention mechanism CNN-LSTM hybrid model, first extracts features from the single-mode data in the data set, for physiological data, captures the trend feature of heart rate over time through the LSTM model; for environmental data, extracts the fluctuation feature of environmental parameters through the CNN model; for behavior data, extracts behavior event features to form a single-mode feature set;

[0091] On the basis of single features, the model focuses on the potential correlation of the three types of data through the attention mechanism, for example, analyzes the coordinated changes of heart rate and blood pressure in high temperature environment, such as whether the heart rate increases by 10 times / min and the blood pressure fluctuates by 5mmHg when the temperature rises by 5℃, mines the correlation between the behavior of staying up late and the next day's fasting blood glucose, and captures the extreme scene correlation, and converts these cross-modal correlation rules into quantifiable correlation features to supplement the feature set.

[0092] Referring to Figure 6 As shown in the figure, the health model construction module specifically includes:

[0093] Basic model matching unit: manage the built-in population general model library, covering different age groups, basic disease type adaptation model. According to the age, sex, past medical history of the user filled in the initial information automatically matched the corresponding population general model;

[0094] Personalized model construction unit: based on the user's stage health feature data, the matched general model is fine-tuned through transfer learning algorithm, which adapts to the user's individual physiological characteristics and generates personalized health model;

[0095] Model iteration and performance monitoring unit: fine-tune the parameters every 7 days of health data accumulation, and reconstruct the model every 30 days. Real-time monitoring of model accuracy, recall rate, F1 score, when the accuracy is lower than the preset threshold, start parameter re-optimization.

[0096] Specifically, the module calls the built-in basic health assessment model library, which contains more than 10 types of population general models. According to the user's basic information, the module matches the general model, and then adjusts the core parameters of the model according to the user's 3-day basic data. After the parameter adjustment is completed, the model is officially put into use and enters the real-time health assessment state, while recording the initial model parameters and the start time;

[0097] After the general model is enabled, the module receives the personal health feature vector output by the multi-dimensional health data fusion analysis module in real time, which includes heart rate trend, blood pressure fluctuation, diet and rest characteristics. The feature data is integrated by day to form the user's daily health feature set, and is associated with the user's daily warning record and feedback information to enrich the model training data dimension. Every 7 days of daily health feature set accumulation, the module starts the transfer learning fine-tuning process, taking the general model as the basic framework, and optimizing the model parameters dimension by dimension with 7-day personalized feature data: for example, by analyzing the user's blood sugar fluctuation after staying up late, the influence weight of rest behavior on metabolic indicators in the model is adjusted; according to the synergistic change of user's heart rate and blood pressure in high temperature environment, the environmental factor related parameters are optimized; During the fine-tuning process, the core framework of the model is retained, and only the parameters that do not match the individual characteristics of the user are adjusted to avoid the interruption of evaluation caused by model reconstruction, and finally the personal exclusive health model that fits the user is formed.

[0098] Referring to Figure 7 As shown in the figure, the risk assessment and grading warning module specifically includes:

[0099] Health risk calculation unit: input the extracted health feature vector into the personalized health model, calculate the user's incidence probability of common health risks, and determine the risk weight coefficient combined with the user's past medical history and family genetic history to calculate and output the comprehensive health risk score;

[0100] Grade pre-warning execution unit: based on the comprehensive health risk score, four levels of pre-warning are divided, level one is only recorded in the background of the system, not actively pushed, level two pushes health tips through APP messages, level three pushes pre-warning information through APP pop-up windows and short messages, and level four additionally triggers voice broadcast and emergency contact person notification;

[0101] Pre-warning threshold configuration unit: based on the doctor's guidance, the pre-warning trigger threshold of various health risks is modified, and the threshold configuration record is saved, including adjustment time, adjustment basis and guiding doctor information.

[0102] Specifically, the module inputs the health feature vector into the user-specific health model, calculates the risk probability one by one according to the disease type, and matches the corresponding evaluation logic in the model for common health risks such as cardiovascular diseases, metabolic diseases and sleep disorders. For example, when calculating the cardiovascular risk, the characteristics of sudden heart rate increase frequency, blood pressure peak value, and heart rate recovery speed after exercise are analyzed, combined with whether the user has a history of hypertension, the cardiovascular risk probability in the next 7 days is output. When calculating the metabolic disease risk, focus on the characteristics of blood glucose fluctuation amplitude, excessive times of dietary heat, body weight change trend, etc., combined with family history of diabetes, the metabolic abnormality risk probability in the next 30 days is output.

[0103] The single risk probability is weighted, and the weight coefficient is dynamically adjusted based on the user's medical history and health focus. If the user has a history of hypertension, the cardiovascular risk weight is increased from the default 20% to 35%. If the user is a teenager, the sleep disorder risk weight is increased from 15% to 25%. After weighting, the preliminary risk value is generated by risk score accumulation;

[0104] The module converts the weighted preliminary risk value into a standard score of 0-100 points, with no risk characteristics + no medical history corresponding to 0 points, and multiple abnormal characteristics + high-risk medical history corresponding to 100 points, and the middle part is linearly mapped according to the risk severity.

[0105] Referring to Figure 8 As shown in the figure, the multi-terminal pre-warning information interaction module specifically includes:

[0106] User-side multi-terminal adaptation unit: adapt to smart watches, smartphones, and home smart terminal devices, and display information according to device characteristics, smart watches present pre-warning levels, key abnormal indicators and quick operations, smartphone APP displays complete pre-warning details, history records and health charts, and home smart terminals push information through voice broadcast and screen display;

[0107] Medical staff side terminal interaction unit: provides a data interface for medical institution special terminals, and pushes real-time level three and above pre-warning information of the patients under the management, and supports medical staff to view complete health data and risk assessment results of the patients;

[0108] The feedback information processing unit provides a user feedback entrance, supports the user to evaluate the warning accuracy and the suggestion practicability, classifies and arranges according to the warning type-feedback result, synchronizes to the dynamic health model construction module, and serves as the basis for model iteration optimization.

[0109] It should be noted that the above-mentioned sequence of the embodiments of the present application is only for description, and does not represent the advantages and disadvantages of the embodiments. The above describes a specific embodiment of the present application. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multi-task processing and parallel processing are also possible or can be advantageous.

[0110] Each embodiment in the specification is described in a progressive manner, and the same or similar parts between each embodiment can be referred to each other. Each embodiment focuses on the difference from other embodiments.

[0111] The above only describes the preferred embodiments of the present application, and does not limit the present application. Any modification, equivalent replacement, improvement, etc. within the principles of the present application shall be included in the protection scope of the present application.

Claims

1. A cloud-based intelligent health early warning system, characterized in that, include: Multimodal health data acquisition module: This module collects comprehensive health data in three categories: physiological, environmental, and behavioral, with the acquisition frequency dynamically adjusted based on user needs; Data preprocessing module: Deployed on the user-side edge device, it performs preprocessing through outlier removal, data normalization, and data compression algorithms, and uses an end-to-end dual encryption mechanism to encrypt data transmission; Cloud data storage and management module: Deployed on a cloud server cluster, it adopts a hybrid structured and unstructured storage architecture and manages data through a built-in data lifecycle management mechanism; Health data fusion and analysis module: Through multimodal data fusion and feature extraction, it constructs a personal health spatiotemporal dataset and explores the potential correlations between physiology, environment and behavior, forming a health feature vector that has a significant impact on health risk assessment, and batch analyzes health data to generate phased health feature data; Health Model Building Module: Builds dynamically adapted health assessment models based on users' periodic health characteristic data. It has a built-in basic health assessment model library and forms personalized health models by adapting to individual user physiological characteristics. Risk assessment and graded early warning module: Based on health feature vectors and personalized health models, calculate the incidence probability of common health risk diseases of users, combine the user's past medical history and family genetic history to determine the risk weight coefficients of each risk, calculate the comprehensive health risk score, and classify the early warning level according to the comprehensive health risk score and the incidence probability of individual risks. Multi-terminal early warning information interaction module: User-side interaction supports four types of devices: smartwatches, smartphones, tablets, and home smart terminals. Medical staff-side interaction is conducted through a dedicated terminal in medical institutions, where users provide feedback by evaluating the early warning information.

2. The intelligent health early warning system based on cloud processing according to claim 1, characterized in that, The multimodal health data acquisition module specifically includes: Physiological data acquisition unit: Adapted to smartwatches, blood glucose meters, and blood pressure monitors, it collects core physiological indicators in real time through a standardized interface, supports dynamic adjustment of the acquisition frequency, automatically caches data when the device is disconnected, and retransmits it after reconnection. Environmental data acquisition unit: integrates temperature, humidity, and air quality sensors to collect real-time parameters such as temperature and PM2.5 concentration in the user's environment, and aligns the collected data with physiological data by timestamp; Behavioral data collection unit: Connects to sports apps and diet tracking tools, provides a user-initiated input interface, and obtains behavioral data such as steps and dietary structure; Equipment Adaptation and Data Coordination Unit: Provides standardized interfaces to adapt to different brands and types of acquisition devices, and unifies data formats.

3. The cloud-based intelligent health early warning system according to claim 1, characterized in that, The data preprocessing module specifically includes: Data preprocessing unit: Adopting an edge computing architecture, the outlier removal algorithm filters out falsely triggered data and sensor fault data during the acquisition process, unifies the units of different device indicators through data normalization, and reduces data volume while retaining key information through data compression. Encrypted transmission unit: It adopts an end-to-end dual encryption mechanism. It uses the AES-256 algorithm to symmetrically encrypt the preprocessed data to generate encrypted data blocks, and performs asymmetric encryption on the encryption key to form key ciphertext. The encrypted data and key ciphertext are then transmitted to the cloud synchronously through a security protocol. Device Coordination and Status Monitoring Unit: Coordinates the working rhythm of wearable devices, home gateways, and mobile terminal edge devices, monitors the operating status of edge devices in real time, and triggers a data caching mechanism if a device disconnects, automatically re-uploading cached data after the device reconnects.

4. The intelligent health early warning system based on cloud processing according to claim 1, characterized in that, The cloud data storage and management module specifically includes: Data classification and storage unit: It adopts a hybrid architecture of structured and unstructured data. Structured data is stored in a distributed relational database and organized by a three-dimensional index structure of user ID-data type-collection time. Unstructured data is stored in a distributed file system and large files are divided and stored through data sharding technology. Data lifecycle management unit: Archives and compresses overdue non-critical data through the built-in data lifecycle management mechanism, permanently stores core physiological data, and completes full backup using a dual strategy of local backup and off-site disaster recovery; Standardized Interface Service Unit: Provides a unified data call interface, outputs standardized data for subsequent analysis modules, receives and processes data requests in real time, and achieves data interoperability by connecting with other modules within the system.

5. The intelligent health early warning system based on cloud processing according to claim 1, characterized in that, The health data fusion and analysis module specifically includes: Data association and matching unit: Through data association algorithms, physiological data, environmental data, and behavioral data within the same time period are associated and matched to construct a personal health spatiotemporal dataset that includes the user's health status and spatiotemporal context; Feature extraction and screening unit: The unit extracts features from the associated personal health spatiotemporal dataset using a CNN-LSTM hybrid model based on an attention mechanism, and mines potential correlation features between physiological indicators, and between physiological indicators and environmental and behavioral factors. The unit then uses a feature screening algorithm based on mutual information entropy to screen out key features that have a significant impact on health risk assessment, forming a health feature vector. Analysis mode management unit: Supports two modes: real-time analysis and batch analysis. Real-time analysis responds within seconds to capture health risks by detecting abnormal fluctuations in core physiological indicators. Batch analysis performs comprehensive analysis on daily health data and generates periodic health characteristic data.

6. The intelligent health early warning system based on cloud processing according to claim 1, characterized in that, The health model construction module specifically includes: Basic Model Matching Unit: Manages the built-in general-purpose model library, covering suitable models for different age groups and underlying disease types. It automatically matches the corresponding general-purpose model based on the user's initial information (age, gender, and medical history) entered during registration. Personalized model building unit: Based on the user's phased health characteristic data, the parameters of the matched general model are fine-tuned through transfer learning algorithm to adapt to the user's individual physiological characteristics and generate a personalized health model; Model Iteration and Performance Monitoring Unit: Fine-tunes parameters every 7 days of accumulated health data, reconstructs the model every 30 days, and monitors model accuracy, recall, and F1 score in real time. When the accuracy falls below a preset threshold, it initiates parameter re-optimization.

7. The intelligent health early warning system based on cloud processing according to claim 1, characterized in that, The risk assessment and graded early warning module specifically includes: Health risk calculation unit: Input the extracted health feature vector into the personalized health model, calculate the probability of the user's occurrence of common health risks, determine the weight coefficient of each risk by combining the user's past medical history and family genetic history, and calculate and output a comprehensive health risk score; Tiered early warning execution unit: Based on the comprehensive health risk score, four levels of early warning are divided. Level 1 is only recorded in the system background and is not actively pushed. Level 2 pushes health tips through APP messages. Level 3 pushes early warning information through both APP pop-up and SMS. Level 4 additionally triggers voice broadcast and emergency contact notification. Warning threshold configuration unit: Based on doctor's guidance, modify the warning trigger thresholds for various health risks, and save the threshold configuration records, including the adjustment time, adjustment basis and guiding doctor information.

8. The intelligent health early warning system based on cloud processing according to claim 1, characterized in that, The multi-terminal early warning information interaction module specifically includes: User-side multi-terminal adaptation unit: Adapts to smartwatches, smartphones, and home smart terminal devices, and displays information differently according to device characteristics. Smartwatches display warning levels, key abnormal indicators, and quick operations. Smartphone apps display complete warning details, historical records, and health charts. Home smart terminals push information through voice broadcasts and screen displays. Medical staff side terminal interaction unit: provides a data interface for medical institution dedicated terminals, pushes level 3 and above early warning information of patients under management in real time, and supports medical staff to view complete health data and risk assessment results of patients; Feedback Information Processing Unit: Provides a user feedback entry point, supports user evaluation of the accuracy of early warnings and the practicality of suggestions, categorizes and organizes feedback results by early warning type, and synchronizes them to the dynamic health model construction module as a basis for model iteration and optimization.

Images