User health state monitoring method and system based on intelligent wearable device

Abstract

The invention discloses a user health state monitoring method and system based on intelligent wearable equipment. The method comprises the following steps: firstly, responding to a current operation mode of the intelligent wearable equipment selected by a user; determining a data acquisition strategy according to the current operation mode; collecting target physiological data of the target user through a corresponding sensor based on the data collection strategy; inputting the target physiological data into a pre-trained multi-dimensional health state index grading model to obtain a target health level corresponding to the target physiological data; acquiring a plurality of target physiological data having a sequential relationship and corresponding target health levels according to the acquisition period; and constructing a health state monitoring report for the target user according to the multiple pieces of target physiological data with the sequential relationship and the corresponding target health levels. Through the design, the health state of the user can be comprehensively and accurately monitored.

Description

Technical Field

[0001] The present invention relates to the field of smart wearable devices, and in particular to a method and system for monitoring the health status of a user based on a smart wearable device. Background Art

[0002] As people's health awareness increases, real-time monitoring of health status becomes more and more important. Smart wearable devices have become the focus of attention due to their portability and real-time monitoring functions. However, existing smart wearable devices have certain limitations in health status monitoring. For example, they can only provide a single physiological data indicator and cannot fully reflect the user's health status. Summary of the invention

[0003] The purpose of the present invention is to provide a method and system for monitoring the health status of a user based on a smart wearable device.

[0004] In a first aspect, an embodiment of the present invention provides a method for monitoring a user's health status based on a smart wearable device, comprising:

[0005] Responding to the current operating mode of the smart wearable device selected by the user;

[0006] Determining a data collection strategy for the smart wearable device according to the current operating mode;

[0007] Based on the data collection strategy, target physiological data of the target user is collected through the corresponding sensor of the smart wearable device;

[0008] Inputting the target physiological data into a pre-trained multi-dimensional health status index grading model to obtain a target health grade corresponding to the target physiological data;

[0009] Acquire a plurality of target physiological data and corresponding target health levels having a time sequence relationship according to a collection period corresponding to the data collection strategy;

[0010] A health status monitoring report for the target user is constructed based on the multiple target physiological data having a time-series relationship and the corresponding target health levels.

[0011] In a second aspect, an embodiment of the present invention provides a server system, including a server, wherein the server is used to execute the method described in the first aspect.

[0012] Compared with the prior art, the beneficial effects provided by the present invention include: using a user health status monitoring method and system based on a smart wearable device disclosed by the present invention, by responding to the current operating mode of the smart wearable device selected by the user; determining the data collection strategy according to the current operating mode; collecting the target physiological data of the target user through the corresponding sensor based on the data collection strategy; inputting the target physiological data into a pre-trained multi-dimensional health status indicator grading model to obtain the target health level corresponding to the target physiological data; obtaining multiple target physiological data with a time series relationship and the corresponding target health level according to the collection cycle; constructing a health status monitoring report for the target user according to the multiple target physiological data with a time series relationship and the corresponding target health level. Such a design can comprehensively and accurately monitor the user's health status. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only illustrate certain embodiments of the present invention and should not be regarded as limiting the scope. For those of ordinary skill in the art, other related drawings can also be obtained based on these drawings without creative work.

[0014] Figure 1 A schematic diagram of the steps of a method for monitoring the health status of a user based on a smart wearable device provided in an embodiment of the present invention;

[0015] Figure 2 A schematic block diagram of the structure of a computer device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0016] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

[0017] The specific implementation modes of the present invention are described in detail below in conjunction with the accompanying drawings.

[0018] In order to solve the technical problems in the aforementioned background technology, Figure 1 This is a flow chart of a method for monitoring the health status of a user based on a smart wearable device provided in an embodiment of the present disclosure. The method for monitoring the health status of a user based on a smart wearable device is introduced in detail below.

[0019] Step S201, responding to the current operating mode of the smart wearable device selected by the user;

[0020] Step S202, determining a data collection strategy for the smart wearable device according to the current operation mode;

[0021] Step S203, collecting target physiological data of the target user through the corresponding sensor of the smart wearable device based on the data collection strategy;

[0022] Step S204, inputting the target physiological data into a pre-trained multi-dimensional health status index grading model to obtain a target health grade corresponding to the target physiological data;

[0023] Step S205, acquiring a plurality of target physiological data and corresponding target health levels having a time sequence relationship according to a collection cycle corresponding to the data collection strategy;

[0024] Step S206: construct a health status monitoring report for the target user according to the multiple target physiological data with a time series relationship and the corresponding target health levels.

[0025] In an embodiment of the present invention, illustratively, a user can select a current operating mode, such as a sports mode, a sleep mode, a daily mode, etc., through an interface or an application of a smart wearable device. After receiving the user's selection, the server determines a subsequent data collection strategy according to the current operating mode. The server determines the best data collection strategy corresponding to the current operating mode selected by the user. For example, in the sports mode, more sports-related physiological data, such as heart rate, acceleration, etc., may need to be collected; while in the sleep mode, sleep-related physiological data, such as electroencephalogram, heart rate, etc., are mainly collected. The smart wearable device collects the target physiological data of the target user through the built-in sensor according to the determined data collection strategy. For example, in the sports mode, a smart bracelet may collect data such as acceleration and angular velocity; in the sleep mode, a smart watch may collect data such as electroencephalogram and heart rate. The server inputs the collected target physiological data into a pre-trained multi-dimensional health status indicator grading model. The model analyzes and calculates based on the input data to determine the target health level of the target user. The health level can be expressed as different levels, such as good, general, and attention-needed. The server regularly obtains multiple target physiological data and corresponding target health levels that have a temporal relationship according to the collection cycle determined by the data collection strategy. In this way, a time series data set can be formed to conduct a more comprehensive analysis and evaluation of the user's health status. The server constructs a health status monitoring report for the target user based on the multiple target physiological data and corresponding target health levels obtained with a time series relationship. The report may include the following: basic information of the target user, such as name, age, gender, etc. Target physiological data in each acquisition cycle, and the corresponding health level. Trend analysis of health status, such as the trend of health level changes over a period of time. Personalized health advice, providing targeted health advice based on target physiological data and health level, such as exercise advice, diet advice, etc. Reminder of abnormal conditions, such as timely reminder of users when obvious abnormal physiological data or health level is found. The report can be presented in a visual form, such as generating a chart, a report document, or displaying it to the user through an application program interface. Users can view the health status monitoring report at any time through smart wearable devices or related applications to understand their health status. Through the above steps, the user health status monitoring method based on smart wearable devices can monitor the user's physiological data in real time, evaluate the health level, and provide personalized health advice to help users better manage their health. At the same time, the server can continuously optimize and adjust the monitoring strategy based on user feedback and changes in health data to improve the effect of health management.

[0026] Assume that user Xiao Ming selects the exercise mode of the smart wearable device. After receiving the selection, the server controls the sensor of the smart wearable device to start collecting Xiao Ming's exercise-related physiological data, such as heart rate, acceleration, etc., according to the data collection strategy determined by the exercise mode. At the same time, the collected data is input into the pre-trained multi-dimensional health status indicator classification model to obtain Xiao Ming's target health level. According to the collection cycle, the server regularly obtains multiple target physiological data and corresponding target health levels with a time series relationship. Over a period of time, the server continues to collect Xiao Ming's exercise data, and according to the analysis results of the model, it is found that Xiao Ming's health level is gradually decreasing. Based on these data and health levels, the server constructs a health status monitoring report for Xiao Ming. The report includes Xiao Ming's basic information, physiological data and health levels in each collection cycle, trend analysis of health status, and personalized health recommendations. The server also displays the monitoring report to Xiao Ming through the application interface to remind him to pay attention to his health status. At the same time, according to the trend analysis, the server found that Xiao Ming had abnormal heart rate fluctuations in his recent exercise, which may have potential health problems. Therefore, the server provides Xiao Ming with further suggestions, such as rest, adjusting exercise intensity, or consulting a doctor. After receiving the report, Xiao Ming took corresponding measures according to the server's suggestions. He adjusted his exercise plan, increased his rest time, and made an appointment with a doctor for further examination. In this way, Xiao Ming was able to understand his health status in a timely manner and take positive measures to maintain his health. In this scenario, smart wearable devices and servers work together to achieve real-time monitoring and personalized management of users' health status. The server provides accurate health assessments and suggestions based on users' choices and physiological data to help users better manage their health. At the same time, users can view reports at any time through the application, understand their health status, and make adjustments based on the suggestions. This health status monitoring method based on smart wearable devices provides users with a convenient and efficient way to manage their health, which helps to improve users' health awareness and quality of life.

[0027] In an embodiment of the present invention, the multi-dimensional health status index grading model is obtained in the following manner.

[0028] Through the multi-dimensional health status index classification model, feature recognition is performed on sample physiological data instances and multiple health statuses respectively to obtain corresponding instance features and multiple health status features;

[0029] For each of the health status features, performing multi-source data integration on the health status feature, the remaining health status features in the plurality of health status features, and the instance feature to obtain an integrated feature;

[0030] Wherein, in the multi-source data integration execution stage, the influence coefficient of the health status feature is higher than the influence coefficient of the remaining health status features;

[0031] Performing an integration operation on the integration feature corresponding to each of the health status features and the multiple health status features to obtain a target integration feature;

[0032] Through the multidimensional health status indicator grading model, the sample physiological data instance is graded into multiple health states according to the target integration characteristics to obtain a health level, and according to the health level and the instance type of the sample physiological data instance, the network parameters of the multidimensional health status indicator grading model are adjusted until a trained multidimensional health status indicator grading model is obtained.

[0033] In an embodiment of the present invention, illustratively, data collection: collect a large number of sample physiological data instances and corresponding multiple health states. These data can come from various sources, such as medical institutions, health monitoring equipment, etc. Feature extraction: use a multidimensional health state indicator grading model to perform feature recognition on sample physiological data instances and multiple health states respectively. This can be achieved by using a deep learning algorithm or other feature extraction methods. Multi-source data integration: for each health state feature, the health state feature is integrated with the remaining health state features and instance features in multiple health state features for multi-source data integration. In the multi-source data integration execution stage, the influence coefficient of the health state feature is higher than the influence coefficient of the remaining health state features. This means that the health state feature has a greater weight in the integration process. Integrated feature generation: perform an integration operation on the corresponding integrated features of each health state feature and multiple health state features to obtain a target integrated feature. Model training: use the target integrated feature and the corresponding health state to grade the sample physiological data instance into multiple health states to obtain a health grade. According to the health grade and the instance type of the sample physiological data instance, adjust the network parameters of the multidimensional health state indicator grading model until a trained multidimensional health state indicator grading model is obtained. Model evaluation: Use the test set to evaluate the trained multi-dimensional health status indicator classification model to determine its accuracy and performance. Model optimization: According to the evaluation results, optimize the multi-dimensional health status indicator classification model, such as adjusting network parameters, increasing training data, etc. Model deployment: Deploy the trained multi-dimensional health status indicator classification model to the server to monitor the user's health status in real time.

[0034] Assume that the server receives a batch of sample physiological data instances and corresponding multiple health states. These data include the user's physiological indicators, such as heart rate, blood pressure, blood sugar, etc., and corresponding health states, such as normal, hypertension, diabetes, etc. The server collects a large number of sample physiological data instances and corresponding health states from multiple data sources. These data are stored in a database for subsequent processing. The server uses a deep learning algorithm to extract features from the sample physiological data instances. The algorithm automatically learns the patterns and features in the physiological data to better understand the data. For each health state feature, the server integrates the health state feature with the remaining health state features and instance features in multiple health state features. For example, for the heart rate health state feature, the server integrates the heart rate feature with other health state features (such as blood pressure, blood sugar, etc.) and instance features (such as user age, gender, etc.). During the integration process, the server assigns different weights to each feature according to the importance of the health state feature. The integration operation is performed on the corresponding integrated features of each health state feature and multiple health state features to obtain the target integrated features. These target integrated features combine multiple health state features and instance features to provide more comprehensive information. The server uses the target integrated features and the corresponding health status to classify the sample physiological data instance into multiple health states to obtain a health level. For example, the target integrated features are input into a classification model, and the model predicts the health state level to which the sample physiological data instance belongs. The server uses a test set to evaluate the trained multidimensional health status indicator grading model. Evaluation indicators may include accuracy, recall, F1 value, etc. Based on the evaluation results, the server optimizes the multidimensional health status indicator grading model. For example, the server can adjust the network parameters of the model, increase the quantity or quality of training data, or use more advanced algorithms. The server deploys the trained multidimensional health status indicator grading model to its own server to monitor the user's health status in real time. When the user wears a smart wearable device and uploads physiological data, the server receives the data and uses the trained model to analyze and evaluate it.

[0035] Through the above steps, the server can use the multi-dimensional health status indicator classification model to monitor and evaluate the user's health status in real time, and provide corresponding health advice and intervention measures based on the evaluation results to help users maintain their health. At the same time, the server can also continuously optimize and improve the model to improve its accuracy and performance, and provide users with better health services.

[0036] In the embodiment of the present invention, for each of the health status features, multi-source data integration is performed on the health status feature, the remaining health status features in the multiple health status features and the instance features to obtain an integrated feature, which can be implemented through the following examples.

[0037] Performing a masking operation on the remaining health status features among the multiple health status features to obtain masked remaining health status features;

[0038] Perform multi-source data integration on the masked remaining health state feature and the health state feature to obtain a multi-health state integrated feature, and perform multi-source data integration on the health state feature and the instance feature to obtain a first multi-source integrated feature;

[0039] An integrated feature corresponding to the health state is obtained according to the multiple health state integrated features and the first multi-source integrated features.

[0040] In an embodiment of the present invention, exemplarily, a masking operation is performed on the remaining health status features except the current health status feature in multiple health status features to obtain the remaining health status features after masking. The masking operation may be to randomly select part of the remaining health status features for hiding, or to select part of the features for hiding according to a certain rule (such as a fixed ratio or based on feature importance). The remaining health status features after masking are integrated with the current health status features through multi-source data to obtain multi-health status integrated features. Multi-source data integration can be performed in a variety of ways, such as element-level addition, multiplication, summation and other operations. At the same time, the current health status features are integrated with the instance features through multi-source data to obtain a first multi-source integrated feature. According to the multi-health status integrated features and the first multi-source integrated features, the integrated features corresponding to the health status are obtained. This can be obtained by combining, fusing or other appropriate operations on the multi-health status integrated features and the first multi-source integrated features.

[0041] Assume that the server receives a batch of sample physiological data, which includes multiple health status features and corresponding instance features. The server needs to perform multi-source data integration on these data to obtain the corresponding integrated features of the health status. The server randomly selects a part of the remaining health status features to hide, and obtains the remaining health status features after masking. The server adds the remaining health status features after masking to the current health status features at the element level to obtain the multi-health status integrated features. At the same time, the server multiplies the current health status features with the instance features at the element level to obtain the first multi-source integrated features. The server sums the multi-health status integrated features and the first multi-source integrated features to obtain the corresponding integrated features of the health status. Through the above steps, the server performs multi-source data integration on each health status feature and obtains the corresponding integrated features. These integrated features contain information on multiple health status features and instance features, which provides a basis for subsequent health status classification.

[0042] In the embodiment of the present invention, obtaining the integrated feature corresponding to the health state according to the multi-health state integrated feature and the first multi-source integrated feature can be implemented through the following examples.

[0043] Perform a combination operation on the multi-health state integrated feature and the first multi-source integrated feature to obtain an integrated feature corresponding to the health state; or

[0044] The multi-health state integrated feature is multiplied by a preset multi-health state integrated feature influence coefficient to obtain a first intermediate feature; the first intermediate feature and the first multi-source integrated feature are combined to obtain an integrated feature corresponding to the health state, wherein the multi-health state integrated feature influence coefficient is not lower than a first threshold value, and the multi-health state integrated feature influence coefficient is not higher than a second threshold value.

[0045] In an embodiment of the present invention, exemplarily, the step of obtaining the integrated feature corresponding to the health state according to the multi-health state integrated feature and the first multi-source integrated feature can be implemented in the following two ways: Combination operation: The server combines the multi-health state integrated feature with the first multi-source integrated feature to obtain the integrated feature corresponding to the health state. This combination operation can be an element-level addition, multiplication, summation and other operations, or a more complex combination function. For example, the server can add the multi-health state integrated feature and the corresponding elements of the first multi-source integrated feature to obtain the integrated feature. Multiplication operation and combination operation: The server first multiplies the multi-health state integrated feature with the preset multi-health state integrated feature influence coefficient to obtain the first intermediate feature. The multi-health state integrated feature influence coefficient is used to control the influence of the multi-health state integrated feature on the integrated feature. The value range of the influence coefficient can be set according to the actual situation, for example, not less than the first threshold and not higher than the second threshold. Then, the server combines the first intermediate feature with the first multi-source integrated feature to obtain the integrated feature corresponding to the health state. Similarly, this combination operation can be an element-level addition, multiplication, summation and other operations, or a more complex combination function.

[0046] Assume that the server receives the multi-health state integration feature 'MHCS' and the first multi-source integration feature 'TDIF', and needs to obtain the corresponding integration feature 'IC' of the health state based on these two features.

[0047] Combined operation method: The server adds the multi-health status integration feature 'MHCS' and the first multi-source integration feature 'TDIF' at the element level to obtain the integrated feature 'IC'. IC = MHCS + TDIF. This method is simple and direct. The corresponding elements of the two features are added to obtain an integrated feature that includes multiple health states and the first multi-source integration information.

[0048] Multiplication and combination operation mode: The server first multiplies the multi-health state integration feature 'MHCS' with the preset multi-health state integration feature influence coefficient 'ICC' to obtain the first intermediate feature 'IMF'. IMF = MHCS * ICC. The value of the multi-health state integration feature influence coefficient 'ICC' can be adjusted according to the actual situation, for example, it is set to 0.5. Then, the server adds the first intermediate feature 'IMF' and the first multi-source integration feature 'TDIF' element-wise to obtain the corresponding integration feature 'IC' of the health state. IC = IMF + TDIF. This method uses multiplication and combination operations to limit the influence of the multi-health state integration feature on the integration feature, while combining the information of the first multi-source integration feature.

[0049] Through the above two methods, the server can select the appropriate method to obtain the corresponding integrated features of the health status according to specific needs and data characteristics. These integrated features will provide an important basis for subsequent health status classification.

[0050] In an embodiment of the present invention, the number of instance features is multiple. After the sample physiological data instances and multiple health states are respectively feature identified through the multidimensional health status indicator grading model to obtain corresponding instance features and multiple health status features, the following implementation method is also provided.

[0051] For each of the instance features, performing multi-source data integration on the instance feature and the plurality of health status features to obtain a sample integration feature;

[0052] The step of grading the sample physiological data instance into a plurality of health states according to the target integration feature to obtain a health level includes:

[0053] The sample physiological data instance is graded into a plurality of health states according to the sample integration feature and the target integration feature to obtain a health grade.

[0054] In an embodiment of the present invention, exemplarily, in the above method, after the sample physiological data instance and multiple health states are respectively identified by the multidimensional health state indicator grading model to obtain the corresponding instance features and multiple health state features, it also includes a step of integrating the instance features with the multiple health state features for each instance feature to obtain the sample integration feature. Then, the sample physiological data instance is graded into multiple health states according to the sample integration feature and the target integration feature to obtain the health level. Assume that the server receives a batch of sample physiological data and needs to perform multidimensional health state indicator grading on these data. Feature identification and acquisition: The server uses the multidimensional health state indicator grading model to perform feature identification on the sample physiological data. The model analyzes the sample physiological data and multiple health states respectively to obtain the corresponding instance features and multiple health state features. Multi-source data integration: For each instance feature, the server performs multi-source data integration with multiple health state features to obtain the sample integration feature. Multi-source data integration can be performed in a variety of ways, such as element-level addition, multiplication, summation and other operations. For example, for a certain instance feature and a certain health state feature, the server can add their corresponding elements to obtain an element in the sample integration feature. The server classifies the sample physiological data instance into multiple health states according to the sample integration feature and the target integration feature to obtain a health level. The classification process can be judged based on the similarity or difference between the sample integration feature and the target integration feature. For example, the server can set a threshold, and when the similarity between the sample integration feature and a health state feature exceeds the threshold, the sample physiological data instance is classified into the health state. The server determines the health level of the sample physiological data instance according to the result of the health state classification. The health level can be expressed in numbers or words, such as "healthy", "sub-healthy", "disease", etc. The server generates a health status monitoring report, which includes basic information of the sample physiological data instance, health status classification results, and related suggestions and guidance. The report can be presented to the user in a visual form, such as displayed on a web page or sent to the user through a mobile application. Through the above steps, the server can comprehensively consider multiple features of the sample physiological data and their relationship with the health state, classify multi-dimensional health status indicators, and provide users with detailed health status reports and suggestions. In this way, users can better understand their health status and take corresponding measures to maintain their health.

[0055] In the embodiment of the present invention, each health state has a corresponding target integration feature, and the sample physiological data instance is graded into multiple health states according to the sample integration feature and the target integration feature to obtain a health level, which can be implemented through the following examples.

[0056] Perform multi-source data integration on the target integrated feature corresponding to the health state, the target integrated features corresponding to the remaining health states in the multiple health states, and the sample integrated feature to obtain the integrated feature to be used;

[0057] The health status of the sample physiological data instance is graded according to the integrated feature to be used to obtain a health grade corresponding to the health status.

[0058] In an embodiment of the present invention, illustratively, in the above method, each health state has a corresponding target integration feature. With the server as the execution body, the sample physiological data instance is graded into multiple health states according to the sample integration feature and the target integration feature, and the specific steps of obtaining the health level are as follows: the server obtains the target integration feature corresponding to each health state. For each health state, the target integration feature represents the feature representation of the health state. At the same time, the target integration features corresponding to the remaining health states other than the current health state in the multiple health states are obtained. The target integration features of the remaining health states are used to provide comparison and reference with the current health state. In addition, the sample integration feature is obtained. The sample integration feature is a comprehensive feature representation of the sample physiological data instance. Next, the target integration feature corresponding to the health state, the target integration feature corresponding to the remaining health state and the sample integration feature are integrated with multi-source data to obtain the integration feature to be used. Multi-source data integration can be performed in a variety of ways, such as element-level addition, multiplication, summation and other operations. In this way, the integration feature to be used integrates the feature information of the health state, the remaining health state and the sample. The server grades the health state of the sample physiological data instance to be tested according to the integration feature to be used. Specifically, the health level can be determined based on the similarity or difference between the integrated feature to be used and the target integrated feature of different health states. For example, a threshold can be set or a classifier can be used to determine which health state the integrated feature to be used belongs to. Assuming that the integrated feature to be used is most similar to the target integrated feature of health state 1 and exceeds the set threshold, the sample physiological data instance is classified as health state 1. The server determines the health level of the sample physiological data instance based on the result of the health state classification. The health level can be expressed in numbers or words, for example, 1 for health, 2 for sub-health, 3 for disease, etc. The server stores or provides the determined health level to the user together with the relevant information of the sample physiological data instance. The following is an example of a specific scenario, which takes the server as the execution subject and describes in detail the execution process of the above steps:

[0059] Assume that the server has received a batch of sample physiological data and needs to classify the health status of these data. Feature acquisition: The server obtains the physiological data of each sample from the database. At the same time, the target integrated features corresponding to each health state are obtained, and these features can be extracted from the physiological data through the feature extraction algorithm. For example, for health state 1, the target integrated feature may be the average value of physiological indicators such as heart rate, blood pressure, and blood sugar. In addition, the target integrated features corresponding to the remaining health states in multiple health states and the sample integrated features are obtained. Multi-source data integration: The server adds the target integrated features of health state 1 and the target integrated features of the remaining health states element-wise to obtain the integrated feature 1 to be used. The integrated feature 1 to be used integrates the feature information of health state 1 and other health states. Then, the integrated feature 1 to be used is element-wise multiplied with the sample integrated feature to obtain the integrated feature 2 to be used. The integrated feature 2 to be used takes into account the influence of the sample features on the health status classification. Health status classification: The server uses the integrated feature 2 to be used to classify the health status of the physiological data of the sample to be tested. The server compares the integrated feature 2 to be used with the target integrated feature of health state 1 and calculates the similarity. If the similarity exceeds the set threshold, the sample physiological data instance is determined to belong to health state 1. Health level determination: The server determines the health level of the sample physiological data instance as 1 and stores this information in the database together with other relevant information of the sample. For example, the server can record the time, location, individual information, etc. of the sample.

[0060] Through the above steps, the server can comprehensively consider the characteristics of different health states and the overall characteristics of the sample, accurately classify the health status of the sample physiological data, and provide personalized health advice and management for users.

[0061] In an embodiment of the present invention, the health status feature includes a first health status feature and a second health status feature;

[0062] For each of the health status features, multi-source data integration is performed on the health status feature, the remaining health status features in the multiple health status features, and the instance features to obtain an integrated feature, which can be implemented through the following examples.

[0063] For each of the health status features, a multiplication result of a second health status feature of the remaining health status features in the multiple health status features and a preset first multi-source integration feature influence coefficient is obtained, and a product feature is obtained according to the first health status feature of the health status feature and the multiplication result, wherein the first multi-source integration feature influence coefficient is not higher than a first threshold value;

[0064] For each of the health status features, performing multi-source data integration on the first health status feature of the health status feature and the health status feature to obtain a health status integration sub-feature, and performing multi-source data integration on the first health status feature of the health status feature and the instance feature to obtain an instance integration sub-feature;

[0065] The integrated feature corresponding to the health state is obtained according to the product feature, the health state integrated sub-feature and the instance integrated sub-feature.

[0066] In an embodiment of the present invention, exemplarily, in the above method, the health state feature includes a first health state feature and a second health state feature. With the server as the execution subject, multi-source data integration is performed for each health state feature, and the specific steps of obtaining the integrated feature are as follows: Feature extraction: The server extracts the first health state feature and the second health state feature from the sample physiological data. The first health state feature and the second health state feature can be physiological indicators or feature vectors related to a specific health state. Multi-source data integration: For each health state feature, the server obtains the second health state feature of the remaining health state features in multiple health state features. At the same time, the server presets the first multi-source integration feature influence coefficient. Next, the server multiplies the second health state feature of the remaining health state feature with the preset first multi-source integration feature influence coefficient to obtain a product feature. The product feature reflects the degree of influence of the remaining health state feature on the current health state feature. Then, the server performs multi-source data integration on the first health state feature of the health state feature and the health state feature to obtain a health state integration sub-feature. Multi-source data integration can be achieved through operations such as element-level addition, multiplication, and summation. Similarly, the server integrates the first health status feature of the health status feature with the instance feature through multi-source data integration to obtain the instance integration sub-feature. Finally, the server obtains the corresponding integrated feature of the health status based on the product feature, the health status integration sub-feature and the instance integration sub-feature. The integrated feature combines the information of the health status feature, the remaining health status features and the instance feature. The following is a specific scenario example, with the server as the execution subject, which describes the execution process of the above steps in detail:

[0067] Assume that the server has received a batch of sample physiological data and needs to perform a health status assessment on these data. Feature extraction: The server extracts a first health status feature and a second health status feature from the sample physiological data. The first health status feature may be heart rate, and the second health status feature may be blood pressure. Multi-source data integration: For the heart rate health status feature, the server obtains the second health status feature of the remaining health status features in other health status features (such as blood pressure, blood sugar, etc.). At the same time, the server presets the influence coefficient of the first multi-source integration feature to be 0.5. Then, the server multiplies the second health status feature (such as blood pressure) of the remaining health status features with the preset first multi-source integration feature influence coefficient of 0.5 to obtain a product feature. Assuming that the second health status feature of blood pressure is 120 and the influence coefficient of the first multi-source integration feature is 0.5, the product feature is 60. Next, the server performs multi-source data integration on the first health status feature (such as heart rate) of the heart rate health status feature and the heart rate health status feature to obtain a health status integration sub-feature. Assuming that the first health state feature of the heart rate health state feature is 70, and the heart rate health state feature is 80, the health state integration sub-feature is 150. Similarly, the server integrates the first health state feature of the heart rate health state feature with the instance feature to obtain the instance integration sub-feature. Assuming that the instance feature is age, the first health state feature of the heart rate health state feature is 70, and the instance feature is 40, the instance integration sub-feature is 280. Finally, the server obtains the corresponding integrated feature of the heart rate health state based on the product feature, the health state integration sub-feature and the instance integration sub-feature. Assuming that the product feature is 60, the health state integration sub-feature is 150, and the instance integration sub-feature is 280, the corresponding integrated feature of the heart rate health state is 500. Through the above steps, the server integrates multi-source data for each health state feature and obtains the corresponding integrated features. These integrated features integrate the information of the health state feature, the remaining health state features and the instance features, providing a more comprehensive reference for subsequent health state assessment.

[0068] In the embodiment of the present invention, the obtaining of the integrated feature corresponding to the health status according to the product feature, the health status integrated sub-feature and the instance integrated sub-feature can be implemented through the following example.

[0069] Perform a combination operation on the product feature, the health status integrated sub-feature and the instance integrated sub-feature to obtain an integrated feature corresponding to the health status; or,

[0070] The product feature is multiplied by a preset product feature influence coefficient to obtain a second intermediate feature; the second intermediate feature, the health status integration sub-feature and the instance integration sub-feature are combined to obtain an integrated feature corresponding to the health status, wherein the product feature influence coefficient is not lower than a first threshold value, and the product feature influence coefficient is not higher than a second threshold value.

[0071] In the embodiment of the present invention, in the above method, the server is used as the execution subject, and the step of obtaining the integrated feature corresponding to the health state according to the product feature, the health state integrated sub-feature and the instance integrated sub-feature can be implemented in the following two ways: Combination operation: The server performs a combination operation on the product feature, the health state integrated sub-feature and the instance integrated sub-feature to obtain the integrated feature corresponding to the health state. This combination operation can be an element-level addition, multiplication, summation and other operations, or a more complex combination function. For example, the server can add the corresponding elements of the product feature, the health state integrated sub-feature and the instance integrated sub-feature to obtain the integrated feature. Multiplication operation and combination operation: The server first multiplies the product feature with the preset product feature influence coefficient to obtain the second intermediate feature. The preset product feature influence coefficient is used to control the degree of influence of the product feature on the integrated feature. The value range of the influence coefficient can be set according to the actual situation, for example, not less than the first threshold and not higher than the second threshold. Then, the server performs a combination operation on the second intermediate feature, the health state integrated sub-feature and the instance integrated sub-feature to obtain the integrated feature corresponding to the health state. Similarly, this combination operation can be element-wise addition, multiplication, summation, etc., or more complex combination functions.

[0072] The following is a specific scenario example, taking the server as the execution subject, and describes in detail the execution process of the above steps:

[0073] Assume that the server receives the product feature 'PC', the health status integration sub-feature 'HCS' and the instance integration sub-feature 'ICS', and needs to obtain the corresponding integration feature 'IC' of the health status based on these three features.

[0074] Combination operation method: The server adds the product feature 'PC', the health status integration sub-feature 'HCS' and the instance integration sub-feature 'ICS' at the element level to obtain the integrated feature 'IC'. IC=PC+HCS+ICS'. This method is simple and direct. The corresponding elements of the three features are added to obtain an integrated feature that contains the product feature, the health status integration sub-feature and the instance integration sub-feature information.

[0075] Multiplication and combination operation mode: The server first multiplies the product feature 'PC' with the preset product feature influence coefficient 'ICP' to obtain the second intermediate feature 'IMF'. IMF = PC * ICP. The value of the preset product feature influence coefficient 'ICP' can be adjusted according to the actual situation, for example, set to 0.5. Then, the server adds the second intermediate feature 'IMF', the health status integration sub-feature 'HCS' and the instance integration sub-feature 'ICS' element-wise to obtain the corresponding integration feature 'IC' for the health status. IC = IMF + HCS + ICS. This method uses multiplication and combination operations to limit the influence of the product feature on the integration feature by the preset product feature influence coefficient, and combines the information of the health status integration sub-feature and the instance integration sub-feature.

[0076] Through the above two methods, the server can select the appropriate method to obtain the corresponding integrated features of the health status according to specific needs and data characteristics. These integrated features will provide an important basis for subsequent health status classification.

[0077] In the embodiment of the present invention, the integration operation is performed on the integration feature corresponding to each of the health status features and the multiple health status features to obtain the target integration feature, which can be implemented through the following examples.

[0078] For each of the health status characteristics, integrating or combining the integration characteristics corresponding to the health status characteristics with the health status characteristics multi-source data to obtain the target integration characteristics corresponding to the health status characteristics;

[0079] The step of grading the sample physiological data instance into a plurality of health states according to the target integration feature to obtain a health level includes:

[0080] For each of the health states, the sample physiological data instances are graded into the health states according to the target integration features corresponding to the health states to obtain the health grade corresponding to the health states.

[0081] In an embodiment of the present invention, exemplarily, in the above method, the integration operation is performed on the integrated features corresponding to each health status feature and multiple health status features, and the step of obtaining the target integrated features can be implemented in the following ways: Feature selection: The server selects the current health status feature from multiple health status features. The selected health status feature can be determined according to specific needs or algorithm settings. Multi-source data integration or combination: For the selected health status feature, the server performs multi-source data integration or combination of its corresponding integrated features with the health status feature. Multi-source data integration may include but is not limited to operations such as element-level addition, multiplication, and summation to comprehensively consider information from different data sources. For example, for the health status feature A and its corresponding integrated feature A', the server can perform the operation of A'+A to obtain the integrated feature A". Target integrated feature generation: After multi-source data integration or combination, the server obtains the target integrated features corresponding to the health status features. These target integrated features integrate the information of the health status features and their integrated features to more comprehensively reflect the health status.

[0082] After obtaining the target integration feature, the sample physiological data instance can be graded into multiple health states according to the feature, and the step of obtaining the health level can be implemented in the following ways: Health state classification: The server determines the number of health states that need to be graded. For example, it may be necessary to divide the health state into different levels such as normal, abnormal 1, and abnormal 2. Target integration feature analysis: The server analyzes the sample physiological data instance for each health state according to its corresponding target integration feature. Classification algorithms, clustering algorithms, or other suitable methods can be used to determine which health state the sample physiological data instance belongs to. Health level determination: The server determines the health state level to which the sample physiological data instance belongs based on the analysis results of the target integration feature. For example, if the target integration feature meets a specific threshold or pattern, the sample physiological data instance is classified into the corresponding health level. Health level report: The server generates a health level report, which contains health state level information for each sample physiological data instance. The report can be presented to the user in a visual form, such as displayed on a web page or sent to the user through a mobile application.

[0083] The following is a specific scenario example, taking the server as the execution subject, and describes in detail the execution process of the above steps:

[0084] Assume that the server receives a batch of sample physiological data and needs to classify the health status of these data. Feature selection and multi-source data integration: The server selects heart rate, blood pressure and blood sugar as the current health status features from multiple health status features. For the corresponding integrated feature HR' of the heart rate feature and the corresponding integrated feature BP' of the blood pressure feature, the server performs the HR'+BP' operation to obtain the comprehensive integrated feature HBP' of the heart rate and blood pressure. Health status classification and analysis: The server determines that the health status needs to be divided into three levels: normal, slightly abnormal and severely abnormal. For the comprehensive integrated feature HBP', the server uses a classification algorithm based on support vector machine to determine which health status the sample physiological data instance belongs to. Health level determination: According to the output of the classification algorithm, the server classifies the sample physiological data instance into three levels: normal, slightly abnormal and severely abnormal. For example, if the comprehensive integrated feature HBP' falls within the threshold range of the normal state, the sample physiological data instance is classified as the normal health level. Health level report: The server generates a health level report, which contains the health status level information of each sample physiological data instance. The report is displayed in a visual way on the user interface, and the user can intuitively understand the health status of each sample. Through the above steps, the server can accurately grade the health status according to the sample physiological data and multi-source data integration features, and provide a detailed health level report to help users better understand their health status.

[0085] In an embodiment of the present invention, the number of the health status characteristics is not less than three, and for each of the health status characteristics, the health status characteristic is integrated with the remaining health status characteristics in the multiple health status characteristics and the instance characteristics through multi-source data to obtain an integrated feature, which can be implemented through the following examples.

[0086] For each of the health status characteristics, selecting a preset number of remaining health status characteristics from a plurality of remaining health status characteristics in the plurality of health status characteristics as first remaining health status characteristics, and selecting the remaining remaining health status characteristics in the plurality of remaining health status characteristics as second remaining health status characteristics;

[0087] performing a masking operation on the first remaining health state feature to obtain a masked first remaining health state feature;

[0088] Perform multi-source data integration on the health state feature, the masked first remaining health state feature, and the second remaining health state feature to obtain a health state mask feature, and perform multi-source data integration on the health state feature and the instance feature to obtain a second multi-source integrated feature;

[0089] According to the health status masking feature and the second multi-source integrated feature, an integrated feature corresponding to the health status is obtained.

[0090] In an embodiment of the present invention, illustratively, in the above method, the number of health status features is not less than three. With the server as the execution body, multi-source data integration is performed for each health status feature, and the specific steps of obtaining the integrated feature are as follows: Feature selection: The server selects a preset number of remaining health status features from multiple health status features as the first remaining health status feature. These remaining health status features can be selected from the health status feature set according to certain rules or algorithms. For example, the first three health status features are selected as the first remaining health status features. At the same time, the remaining health status features in the multiple remaining health status features are used as the second remaining health status features. The second remaining health status features are other health status features other than the first remaining health status features. Masking operation: A masking operation is performed on the first remaining health status feature to obtain the masked first remaining health status feature. The masking operation can be to randomly select part of the first remaining health status feature to hide, or to select part of the feature to hide according to certain rules (such as a fixed ratio or based on feature importance). Multi-source data integration: The health status feature is multi-source data integrated with the masked first remaining health status feature and the second remaining health status feature to obtain the health status mask feature. Multi-source data integration can be performed in a variety of ways, such as element-level addition, multiplication, summation, and other operations. At the same time, the health status feature and the instance feature are multi-source data integrated to obtain a second multi-source integrated feature. Integrated feature generation: According to the health status mask feature and the second multi-source integrated feature, the integrated feature corresponding to the health status is obtained. This can be obtained by combining, fusing, or other appropriate operations on the health status mask feature and the second multi-source integrated feature.

[0091] The following is a specific scenario example, taking the server as the execution subject, and describes in detail the execution process of the above steps:

[0092] Assume that the server receives a batch of sample physiological data and needs to integrate these data from multiple sources to obtain the corresponding integrated features of the health status.

[0093] Feature selection: The server selects heart rate, blood pressure, blood sugar and blood lipids as health status features from multiple health status features. The first three health status features are selected as the first remaining health status features, namely heart rate, blood pressure and blood sugar. The remaining health status feature blood lipids are used as the second remaining health status features. Masking operation: Masking operation is performed on the heart rate and blood pressure in the first remaining health status features to obtain the masked first remaining health status features. For example, part of the heart rate and blood pressure data is randomly selected for hiding. Multi-source data integration: The health status features (such as blood sugar) are element-wise added to the masked first remaining health status features (such as heart rate and blood pressure) to obtain the health status masked features. At the same time, the health status features are element-wise multiplied with the instance features (such as age and gender) to obtain the second multi-source integrated features. Integrated feature generation: According to the health status masked features and the second multi-source integrated features, the corresponding integrated features of the health status are obtained by summing operations. For example, the sum of the corresponding elements of the health status masked features and the second multi-source integrated features is calculated to obtain the integrated features. Through the above steps, the server integrates multi-source data for each health status feature and obtains the corresponding integrated features. These integrated features combine the information of health status features, remaining health status features and instance features, providing a more comprehensive reference for subsequent health status classification.

[0094] In the embodiment of the present invention, obtaining the integrated feature corresponding to the health status according to the health status masking feature and the second multi-source integrated feature can be implemented through the following examples.

[0095] Perform a combination operation on the health status masking feature and the second multi-source integrated feature to obtain an integrated feature corresponding to the health status; or,

[0096] The health status masking feature is multiplied by the influence coefficient of the preset health status integration sub-feature to obtain a third intermediate feature; the third intermediate feature is combined with the second multi-source integration feature to obtain an integration feature corresponding to the health status;

[0097] The health status integration sub-feature influence coefficient is not lower than a first threshold, and the health status integration sub-feature influence coefficient is not higher than a second threshold.

[0098] In an embodiment of the present invention, exemplarily, in the above method, the server is used as the execution subject, and the step of obtaining the integrated feature corresponding to the health state according to the health state masking feature and the second multi-source integrated feature can be implemented in the following two ways: combination operation: the server combines the health state masking feature with the second multi-source integrated feature to obtain the integrated feature corresponding to the health state. This combination operation can be an element-level addition, multiplication, summation and other operations, or a more complex combination function. For example, the server can add the corresponding elements of the health state masking feature and the second multi-source integrated feature to obtain the integrated feature. Multiplication operation and combination operation: the server first multiplies the health state masking feature with the preset health state integrated sub-feature influence coefficient to obtain the third intermediate feature. The preset health state integrated sub-feature influence coefficient is used to control the degree of influence of the health state masking feature on the integrated feature. The value range of the influence coefficient can be set according to the actual situation, for example, not less than the first threshold and not higher than the second threshold. Then, the server combines the third intermediate feature and the second multi-source integrated feature to obtain the integrated feature corresponding to the health state. Similarly, this combination operation can be element-wise addition, multiplication, summation, etc., or more complex combination functions.

[0099] The following is a specific scenario example, taking the server as the execution subject, and describes in detail the execution process of the above steps:

[0100] Assume that the server receives the health status mask feature 'HSF' and the second multi-source integration feature 'TDIF', and needs to obtain the corresponding integration feature 'IC' of the health status based on these two features.

[0101] Combined operation mode:

[0102] The server performs element-wise addition of the health status masking feature 'HSF' and the second multi-source integrated feature 'TDIF' to obtain the integrated feature 'IC'. IC=HSF+TDIF' is a simple and direct method that adds the corresponding elements of the two features to obtain an integrated feature that includes the health status masking feature and the second multi-source integrated feature information.

[0103] Multiplication and combination operations:

[0104] The server first multiplies the health status masking feature 'HSF' with the preset health status integration sub-feature influence coefficient 'ICS' to obtain the third intermediate feature 'IMF'. IMF = HSF * ICS. The value of the preset health status integration sub-feature influence coefficient 'ICS' can be adjusted according to the actual situation, for example, set to 0.5. Then, the server adds the third intermediate feature 'IMF' and the second multi-source integration feature 'TDIF' element-wise to obtain the corresponding integration feature 'IC' of the health status. IC = IMF + TDIF. This method uses multiplication and combination operations to limit the degree of influence of the health status masking feature on the integration feature by the preset health status integration sub-feature influence coefficient, while combining the information of the second multi-source integration feature. Through the above two methods, the server can select an appropriate method to obtain the corresponding integration feature of the health status according to specific needs and data characteristics. These integration features will provide an important basis for subsequent health status classification.

[0105] In the embodiments of the present invention, the following implementation modes are also provided.

[0106] Obtain corresponding noisy health index collection values ​​sent by multiple test smart wearable devices to obtain a noisy health index collection value set; each of the noisy health index collection values ​​is the data after the original health index collection value of the corresponding test smart wearable device is noisy according to the noisy strategy;

[0107] Acquire a test data set, the test data set comprising a plurality of test data, each test data comprising any preset health indicator value of any preset health indicator in a preset health indicator set, and any health status type in a preset health status type set; the noisy health indicator collection value and the original health indicator collection value are in the form of a preset matrix, and the element value in the preset matrix indicates whether the key data collected by the corresponding test smart wearable device exists in the test data;

[0108] Based on the noise-added health indicator collection value set and the noise-added strategy, determining the inferred proportion of each test data in the test data set in the original health indicator collection value set; the original health indicator collection value set includes the original health indicator collection values ​​of each of the multiple test smart wearable devices;

[0109] Determining the critical impact of each preset health indicator in the preset health indicator set on the target health level based on the inferred proportion of each of the test data;

[0110] Based on the key impact degree of each preset health indicator in the preset health indicator set, health status characteristics are extracted from the preset health indicator set; the health status characteristics are used to train a multidimensional health status indicator grading model for the target health level.

[0111] In an embodiment of the present invention, exemplarily, the server receives corresponding noise-added health index collection values ​​from multiple test smart wearable devices. The noise-added health index collection value is obtained by applying a noise-added strategy to the original health index collection value. For example, the original health index collection value may be a heart rate value, and noise can be added to it through a noise-added strategy to obtain a noise-added health index collection value. The server obtains a test data set, which contains multiple test data. Each test data includes a preset health index value of any preset health index in a preset health index set, and any health state type in a preset health state type set. The noise-added health index collection value and the original health index collection value are represented in the form of a preset matrix, and the element value in the preset matrix indicates whether the key data collected by the corresponding test smart wearable device exists in the test data. The server determines the inference proportion of each test data in the test data set in the original health index collection value set based on the noise-added health index collection value set and the noise-added strategy. The inference proportion reflects the degree of matching between the noise-added health index collection value and the original health index collection value. The inference proportion can be calculated by comparing the similarity between the noise-added health index collection value and the original health index collection value. The server determines the critical influence degree of each preset health indicator in the preset health indicator set on the target health level based on the inferred proportion of each test data. The critical influence degree indicates the importance of the preset health indicator to the health status classification. The critical influence degree can be determined according to the size of the inferred proportion. The server extracts health status features from the preset health indicator set based on the critical influence degree of each preset health indicator in the preset health indicator set. The health status features are key preset health indicators related to the target health level. The health status features can be selected according to the threshold of the critical influence degree or other rules. The server uses the extracted health status features to train the multidimensional health status indicator grading model. The training process may include steps such as adjusting the parameters of the model and optimizing the performance of the model. The trained multidimensional health status indicator grading model can be used to accurately grade the user's health status.

[0112] The following is a specific scenario example, taking the server as the execution subject, and describes in detail the execution process of the above steps:

[0113] Assume that the server has received noisy health indicator collection values ​​from multiple test smart wearable devices, and needs to process and analyze these data for the training of a multi-dimensional health status indicator grading model. Obtain a set of noisy health indicator collection values: The server receives the noisy health indicator collection value 'A1' from test smart wearable device 1, and receives the noisy health indicator collection value 'A2' from test smart wearable device 2, and so on. The noisy health indicator collection value 'A1' is obtained by applying the noise addition strategy to the original health indicator collection value 'H1'. Obtain a test data set: The server obtains a test data set 'D', which contains multiple test data. The test data 'd1' includes the preset health indicator value 'H1' in the preset health indicator set 'P' and the health status type 'T1' in the preset health status type set 'S'. The noisy health indicator collection value 'A1' and the original health indicator collection value 'H1' are represented in the form of a preset matrix, where the element value '1' indicates that the key data collected by the corresponding device exists in the test data, and the element value '0' indicates that it does not exist. Determine the inference ratio: The server analyzes the similarity between the noisy health indicator collection value 'A1' and the original health indicator collection value 'H1'. The similarity can be measured using indicators such as correlation coefficient and mean square error. According to the similarity result, the inferred proportion of the test data 'd1' in the original health indicator collection value set is determined as 'p1'. Determine the critical impact degree**: The server determines the critical impact degree of the preset health indicator 'H1' in the preset health indicator set 'P' on the target health level as 'c1' based on the inferred proportion 'p1' and other factors. The critical impact degree can be calculated according to the preset rules or models. Extract health status features: The server extracts the health status feature 'F1' from the preset health indicator set 'P' based on the critical impact degree 'c1', such as the preset health indicator 'H1'. Multi-dimensional health status indicator grading model training: The server uses the health status feature 'F1' to train the multi-dimensional health status indicator grading model. During the training process, the server adjusts the parameters of the model and optimizes the performance of the model to improve the accuracy and generalization ability of the model. After the training is completed, the multi-dimensional health status indicator grading model can be used to grade the health status of new health indicator data.

[0114] Through the above steps, the server can use the noisy health indicator collection values ​​and test data sets to determine the key impact of the preset health indicators on the target health level, and extract health status features for training the multi-dimensional health status indicator classification model. This can improve the accuracy and adaptability of the model and better monitor and evaluate the user's health status.

[0115] In an embodiment of the present invention, the determination of the inferred proportion of each test data in the test data set in the original health indicator collection value set based on the noisy health indicator collection value set and the noisy strategy can be implemented through the following examples.

[0116] For each test data in the test data set, extract the associated element value corresponding to the test data from each noisy health indicator collection value in the noisy health indicator collection value set; the position arrangement of each associated element value in the corresponding noisy health indicator collection value corresponds to the position arrangement of the test data in the test data set;

[0117] For each of the multiple associated element values, determining whether the test data indicated by the targeted associated element value is the targeted test data, and obtaining a recognition result corresponding to the targeted associated element value;

[0118] Based on the corresponding identification results of each of the associated element values ​​and the noise adding strategy, the inferred proportion of the targeted test data in the original health indicator collection value set is determined.

[0119] In an embodiment of the present invention, exemplarily, in the above method, the server is used as the execution subject, and based on the noisy health indicator collection value set and the noise adding strategy, the step of determining the inference proportion of each test data in the test data set in the original health indicator collection value set can be implemented in the following manner: Associated element value extraction: The server extracts the associated element value corresponding to the test data from the noisy health indicator collection value set for each test data in the test data set. The associated element value refers to the element value in the noisy health indicator collection value set that corresponds to the position arrangement of the test data in the test data set. Identification result determination: The server analyzes each element value in a plurality of associated element values ​​to determine whether the test data indicated by the element value is the targeted test data. It can be determined whether the associated element value matches the test data according to a preset rule or algorithm. Obtain the identification result corresponding to the associated element value, wherein the identification result can be 1 for match and 0 for mismatch. Inference proportion calculation: The server determines the inference proportion of the targeted test data in the original health indicator collection value set based on the corresponding identification results and noise adding strategy of each associated element value. The inference proportion reflects the degree of match between the noisy health indicator collection value and the original health indicator collection value. The inference ratio can be determined by calculating the proportion of associated element values ​​with an identification result of 1. The following is a specific scenario example, with the server as the execution subject, and describes in detail the execution process of the above steps: Assume that the server receives the noise-added health indicator collection value set 'GHC' and the noise-added strategy 'GS', and needs to determine the inference ratio of each test data in the test data set in the original health indicator collection value set 'OHC'. Associated element value extraction: The server extracts the associated element value 'g1' from the noise-added health indicator collection value set 'GHC' for the test data 'td1' in the test data set. The position arrangement of the associated element value 'g1' in the noise-added health indicator collection value set 'GHC' corresponds to the position arrangement of the test data 'td1' in the test data set. Identification result determination: The server analyzes the associated element value 'g1' to determine whether it matches the test data 'td1'. According to the preset rules, determine whether the associated element value 'g1' is equal to the original health indicator collection value of the test data 'td1'. Get the recognition result 'er1' corresponding to the associated element value 'g1'. Assume that the recognition result 'er1' is 1, indicating a match. Inference ratio calculation: Based on the recognition result 'er1' and the noise addition strategy 'GS', the server determines the inference ratio 'ip1' of the test data 'td1' in the original health indicator collection value set 'OHC'. According to the noise addition strategy 'GS', it can be determined how much it affects the inference ratio when the associated element value is 1. Assume that in this case, the inference ratio increases by 10%.Then the inferred proportion 'ip1' of the test data 'td1' in the original health indicator collection value set 'OHC' is the original proportion plus the increased proportion, that is, 'ip1=or1+0.1*or1`, where 'or1' is the original inferred proportion of the test data 'td1' in the original health indicator collection value set 'OHC'. Through the above steps, the server can determine the inferred proportion of each test data in the test data set in the original health indicator collection value set according to the noisy health indicator collection value set and the noisy strategy, thereby reflecting the matching degree between the noisy health indicator collection value and the original health indicator collection value. This is helpful for the subsequent multi-dimensional health status indicator grading model training and health status assessment.

[0120] In an embodiment of the present invention, the noise adding strategy includes a first noise adding strategy and a second noise adding strategy; the first noise adding strategy is that the element value in the original health index collection value is maintained in its original state after noise adding based on a first preset probability; the second noise adding strategy is that the element value in the original health index collection value is changed to a preset value after noise adding based on a second preset probability; the preset value is a value indicating that the key data collected by the corresponding test smart wearable device has the value of the targeted test data;

[0121] The determination of the inferred proportion of the targeted test data in the original health indicator collection value set based on the corresponding identification results of each of the associated element values ​​and the noise adding strategy can be implemented through the following examples.

[0122] Obtaining a first transition parameter based on the corresponding identification results of each of the associated element values, the first noise adding strategy, and the data scale of the noisy health indicator collection values ​​in the noisy health indicator collection value set;

[0123] Obtaining a second transition parameter based on a subtraction result of the first noise adding strategy and the second noise adding strategy;

[0124] Based on the result of the division operation of the first transition parameter and the second transition parameter, the inferred proportion of the targeted test data in the key data set is obtained.

[0125] In an embodiment of the present invention, exemplarily, in the above method, the server is used as the execution subject, and based on the corresponding recognition results and noise adding strategies of each associated element value, the step of determining the inference proportion of the targeted test data in the original health indicator collection value set can be implemented in the following manner: First transition parameter calculation: The server determines the number of associated element values ​​with the recognition result of 1 according to the corresponding recognition results of each associated element value. At the same time, the server obtains the data scale of the noisy health indicator collection value in the noisy health indicator collection value set. Based on the recognition result and the data scale, the server can calculate the first transition parameter. Second transition parameter calculation: The server obtains the second transition parameter according to the subtraction operation result of the first noise adding strategy and the second noise adding strategy. The first noise adding strategy indicates that the element value in the original health indicator collection value remains the same after noise adding based on the first preset probability, and the second noise adding strategy indicates that the element value in the original health indicator collection value is changed to a preset value after noise adding based on the second preset probability. The subtraction operation result reflects the difference between the two noise adding strategies. Inference proportion calculation: The server obtains the inference proportion of the targeted test data in the key data set based on the division operation result of the first transition parameter and the second transition parameter. The division operation result represents the relative proportional relationship between the first transition parameter and the second transition parameter, thereby reflecting the relative importance of the targeted test data in the key data set.

[0126] The following is a specific scenario example, taking the server as the execution subject, and describes in detail the execution process of the above steps:

[0127] Assume that the server receives a noisy health indicator collection value set 'GHC', including noisy health indicator collection values ​​of multiple test data, and an original health indicator collection value set 'OHC', including original health indicator collection values ​​of multiple test data.

[0128] First transition parameter calculation:

[0129] The server analyzes each noise-added health indicator collection value in the noise-added health indicator collection value set 'GHC' to determine the associated element value. The associated element value indicates whether the key data collected by the corresponding test smart wearable device exists in the targeted test data. The server counts the number of associated element values ​​with an identification result of 1, recorded as 'tp1'. At the same time, the server obtains the data scale of the noise-added health indicator collection value in the noise-added health indicator collection value set 'GHC', recorded as 'sz1'. Based on the identification result and data scale, the server calculates the first transition parameter 't1`, and the formula is 't1=tp1 / sz1`

[0130] Second transition parameter calculation:

[0131] The server calculates the second transition parameter 't2' according to the definition of the first noise adding strategy and the second noise adding strategy.

[0132] Assume that the first noise adding strategy indicates that the element value in the original health indicator collection value has an 80% probability of remaining the same after noise adding, and the second noise adding strategy indicates that the element value in the original health indicator collection value has a 20% probability of changing to the preset value 100 after noise adding.

[0133] Then the second transition parameter 't2' can be obtained by calculating the difference between the first noise adding strategy and the second noise adding strategy, that is, 't2 = 0.8-0.2 = 0.6'

[0134] Inferred proportion calculation:

[0135] The server divides the first transition parameter 't1' and the second transition parameter 't2' to obtain the inferred proportion 'ip' of the targeted test data in the key data set.

[0136] The specific calculation formula is 'ip = t1 / t2 = tp1 / sz1 / (0.8-0.2) = tp1 / sz1 / 0.6'

[0137] The inference ratio indicates the relative ratio of the targeted test data in the key data set, reflecting the impact of the matching degree between the noisy health indicator collection value and the original health indicator collection value on the inference ratio.

[0138] Through the above steps, the server can calculate the inferred proportion of the targeted test data in the original health indicator collection value set based on the recognition results of the associated element values, the noise addition strategy and the information of the noisy health indicator collection value set, thereby providing an important basis for the training of the multidimensional health status indicator grading model.

[0139] In an embodiment of the present invention, exemplarily, in the above method, the server is used as the execution subject, and based on the corresponding recognition results and noise adding strategies of each associated element value, the step of determining the inference proportion of the targeted test data in the original health indicator collection value set can be implemented in the following manner: First transition parameter calculation: The server determines the number of associated element values ​​with the recognition result of 1 according to the corresponding recognition results of each associated element value. At the same time, the server obtains the data scale of the noisy health indicator collection value in the noisy health indicator collection value set. Based on the recognition result and the data scale, the server can calculate the first transition parameter. Second transition parameter calculation: The server obtains the second transition parameter according to the subtraction operation result of the first noise adding strategy and the second noise adding strategy. The first noise adding strategy indicates that the element value in the original health indicator collection value remains the same after noise adding based on the first preset probability, and the second noise adding strategy indicates that the element value in the original health indicator collection value is changed to a preset value after noise adding based on the second preset probability. The subtraction operation result reflects the difference between the two noise adding strategies. Inference proportion calculation: The server obtains the inference proportion of the targeted test data in the key data set based on the division operation result of the first transition parameter and the second transition parameter. The result of the division operation represents the relative proportional relationship between the first transition parameter and the second transition parameter, thereby reflecting the relative importance of the targeted test data in the key data set. The following is an example of a specific scenario, which takes the server as the execution subject and describes in detail the execution process of the above steps: Assume that the server receives the noise-added health indicator collection value set 'GHC', including the noise-added health indicator collection values ​​of multiple test data, and the original health indicator collection value set 'OHC', including the original health indicator collection values ​​of multiple test data. First transition parameter calculation: The server analyzes each noise-added health indicator collection value in the noise-added health indicator collection value set 'GHC' to determine the associated element value. The associated element value indicates whether the key data collected by the corresponding test smart wearable device exists in the targeted test data. The server counts the number of associated element values ​​whose identification result is 1, recorded as 'tp1'. At the same time, the server obtains the data scale of the noisy health indicator collection value in the noisy health indicator collection value set 'GHC', recorded as 'sz1'. Based on the identification result and the data scale, the server calculates the first transition parameter 't1', and the formula is 't1=tp1 / sz1'. The second transition parameter calculation: The server calculates the second transition parameter 't2' according to the definitions of the first noisy strategy and the second noisy strategy. It is assumed that the first noisy strategy indicates that the element value in the original health indicator collection value has an 80% probability of remaining the same after noisy, and the second noisy strategy indicates that the element value in the original health indicator collection value has a 20% probability of changing to the preset value 100 after noisy.Then the second transition parameter 't2' can be obtained by calculating the difference between the first noise adding strategy and the second noise adding strategy, that is, 't2 = 0.8-0.2 = 0.6' Inference ratio calculation: the server divides the first transition parameter 't1' and the second transition parameter 't2' to obtain the inference ratio 'ip' of the targeted test data in the key data set. The specific calculation formula is 'ip = t1 / t2 = tp1 / sz1 / (0.8-0.2) = tp1 / sz1 / 0.6' The inference ratio represents the relative ratio of the targeted test data in the key data set, reflecting the influence of the matching degree between the noise added health indicator collection value and the original health indicator collection value on the inference ratio. Through the above steps, the server can calculate the inference ratio of the targeted test data in the original health indicator collection value set according to the recognition results of the associated element values, the noise adding strategy and the information of the noise added health indicator collection value set, thereby providing an important basis for the training of the multi-dimensional health status indicator grading model.

[0140] In an embodiment of the present invention, the preset health status type set includes target health level types; the preset health indicator set includes target preset health indicators; the inferred proportion based on each of the test data is used to determine the critical impact of each preset health indicator in the preset health indicator set on the target health level, which can be implemented through the following examples.

[0141] Based on the inferred proportion of each of the test data, determining the number of types of the target health level type existing in the key data set;

[0142] Based on the inferred proportion of each of the test data, determine the target preset health indicator A k The number of indicators existing in the key data set for each preset health indicator value below;

[0143] Based on the number of each type, the number of each indicator and the overall parameters of the noise-added health indicator collection value in the noise-added health indicator collection value set, the critical impact degree of each preset health indicator in the preset health indicator set on the target health level is determined.

[0144] In an embodiment of the present invention, exemplarily, in the above method, the server is used as the execution subject, and based on the inferred proportion of each test data, the step of determining the critical impact of each preset health indicator in the preset health indicator set on the target health level can be achieved in the following manner: Determination of the number of target health level types: The server determines the number of types of the target health level type existing in the key data set based on the inferred proportion of each test data. The number of types existing indicates the number of times the target health level type appears in the key data set. For example, if the target health level type is "normal", and the inferred proportion of "normal" in the test data appearing in the key data set is 80%, it can be determined that the number of types of "normal" existing in the key data set is 80%. Determination of the number of preset health indicator values: The server determines the number of indicators of each preset health indicator value under the target preset health indicator existing in the key data set based on the inferred proportion of each test data. The number of indicators existing indicates the number of times the target preset health indicator value appears in the key data set. For example, if the target preset health indicator is "blood pressure", and in the key data set, the preset health indicator value of "blood pressure" is 120mmHg, and in the test data where "blood pressure" appears, the inference ratio of "120mmHg" is 60%, then it can be determined that the number of indicators of "120mmHg" existing in the key data set is 60%. Determination of critical impact degree: The server determines the critical impact degree of each preset health indicator in the preset health indicator set on the target health level based on the number of each type, the number of each indicator, and the overall parameters of the noisy health indicator collection value in the noisy health indicator collection value set. The critical impact degree reflects the importance of the preset health indicator to the target health level. The critical impact degree can be determined by calculating the correlation coefficient, variance analysis, and other methods. For example, the correlation coefficient between the number of target health level types and the number of target preset health indicator values ​​can be calculated, or the difference in the critical impact degree of different preset health indicators can be compared.

[0145] The following is a specific scenario example, taking the server as the execution subject, and describes in detail the execution process of the above steps:

[0146] Assume that the server receives a batch of test data, including a preset health indicator set and a preset health status type set. The preset health indicator set includes the preset health indicators "blood pressure" and "blood sugar", and the preset health status type set includes the target health level types "normal" and "abnormal". The number of target health level types existing is determined: the server analyzes the inference ratio of each test data. It is found that the inference ratio of "normal" is 80%, and the inference ratio of "abnormal" is 20%. Therefore, it can be determined that the number of types existing in the key data set for "normal" is 80%, and the number of types existing in the key data set for "abnormal" is 20%. The number of target preset health indicator values ​​existing is determined: the server further analyzes the number of preset health indicator values ​​of "blood pressure" and "blood sugar" in each test data. It is found that the number of preset health indicator values ​​of "blood pressure" of 120mmHg is 60%, and the number of preset health indicator values ​​of "blood sugar" of 100mg / dl is 40%. Therefore, it can be determined that the number of indicators existing in the key data set for "blood pressure" is 60%, and the number of indicators existing in the key data set for "blood sugar" is 40%. Determination of critical impact degree: The server comprehensively considers the number of types and the number of indicators. It is found that the number of types of "blood pressure" (80%) is much larger than the number of types of "blood sugar" (20%). At the same time, the number of indicators of "blood pressure" (60%) is also larger than the number of indicators of "blood sugar" (40%). In addition, the server also considers the overall parameters of the noisy health indicator collection values ​​in the noisy health indicator collection value set. After calculation and analysis, it is determined that "blood pressure" has a higher critical impact on the target health level, while "blood sugar" has a relatively lower critical impact.

[0147] Through the above steps, the server can determine the critical impact of each preset health indicator in the preset health indicator set on the target health level based on the inference ratio of the test data, taking into account the target health level type and the number of target preset health indicator values, as well as the overall parameters of the noisy health indicator collection value set. This helps to accurately select the preset health indicators that have an important impact on the target health level in the training of the multi-dimensional health status indicator grading model, and improve the accuracy and reliability of the model.

[0148] In an embodiment of the present invention, the critical influence degree of each preset health indicator in the preset health indicator set on the target health level is determined based on the number of existence of each type, the number of existence of each indicator and the overall parameters of the noise-added health indicator collection value in the noise-added health indicator collection value set, which can be implemented through the following examples.

[0149] Determine the information confusion degree corresponding to the preset health status type set based on the number of types of the target health level type existing in the original health indicator collection value set and the overall parameter of the noisy health indicator collection value; the information confusion degree represents an average trend indicating the probability of the health status type existing in the preset health status type set;

[0150] Based on the overall parameters of the noisy health indicator collection value, the number of indicators existing in the original health indicator collection value set for each preset health indicator value under the target preset health indicator, and the inference proportion associated with the target preset health indicator, determine the conditional confusion degree corresponding to the target preset health indicator; the conditional confusion degree indicates the average state of the probability of the target preset health indicator existing under the condition of presetting the preset health state type set;

[0151] Based on the information confusion degree and the condition confusion degree, the critical influence degree of the target preset health indicator on the target health level is determined.

[0152] In an embodiment of the present invention, exemplarily, in the above method, with the server as the execution subject, based on the number of types, the number of indicators and the overall parameters of the noisy health indicator collection value in the noisy health indicator collection value set, the step of determining the critical influence of each preset health indicator in the preset health indicator set on the target health level can be implemented in the following manner: Information confusion determination: The server obtains the number of types of the target health level type existing in the original health indicator collection value set. At the same time, the server obtains the overall parameters of the noisy health indicator collection value. Based on these two parameters, the server can determine the information confusion corresponding to the preset health state type set. The information confusion can characterize the average situation of the probability of the health state type existing in the preset health state type set. Conditional confusion determination: The server further obtains the number of indicators existing in the original health indicator collection value set for each preset health indicator value under the target preset health indicator. At the same time, the server obtains the inference proportion associated with the target preset health indicator. Based on these parameters, the server can determine the conditional confusion corresponding to the target preset health indicator. The conditional confusion can indicate the average situation of the probability of the target preset health indicator existing under the condition of the preset preset health state type set. Determination of critical impact degree: The server comprehensively considers the information chaos degree and condition chaos degree. The critical impact degree of the target preset health indicator on the target health level can be determined by calculating the difference between the two, the correlation coefficient, etc. The critical impact degree reflects the importance of the preset health indicator in the target health level assessment.

[0153] The following is a specific scenario example, taking the server as the execution subject, and describes in detail the execution process of the above steps:

[0154] Assume that the server receives a batch of health indicator data, including an original health indicator collection value set, a noisy health indicator collection value set, a preset health state type set, and a target preset health indicator. Information confusion determination: The server counts the number of types of the target health level type that exist in the original health indicator collection value set. For example, the target health level types are "normal" and "abnormal", and in the original health indicator collection value set, the number of "normal" is 800, and the number of "abnormal" is 200. At the same time, the server obtains the overall parameters of the noisy health indicator collection value, such as the standard deviation of the noisy health indicator collection value. Based on this information, the server can calculate the information confusion corresponding to the preset health state type set. The information confusion can be determined by calculating the variance of the occurrence frequency of the health state type in the preset health state type set. Conditional confusion determination: The server obtains the number of indicators that exist in the original health indicator collection value set for each preset health indicator value under the target preset health indicator. For example, the target preset health indicator is "blood pressure", and its preset health indicator value is 120mmHg. In the original health indicator collection value set, the number of "blood pressure" with 120mmHg is 600. At the same time, the server obtains the inferred proportion associated with the target preset health indicator. The inferred proportion indicates the proportion of health status types corresponding to the target preset health indicator in the original health indicator collection value set. Based on this information, the server can determine the conditional confusion corresponding to the target preset health indicator. The conditional confusion can be determined by calculating the variance of the occurrence frequency of the target preset health indicator under different health status types. Determination of the critical impact degree: The server comprehensively considers the information confusion degree and the conditional confusion degree. The critical impact degree of the target preset health indicator on the target health level can be determined by calculating the difference between the two. The larger the difference, the greater the impact of the target preset health indicator on the target health level.

[0155] Through the above steps, the server can determine the critical impact of each preset health indicator in the preset health indicator set on the target health level based on the number of each type, the number of each indicator, and the overall parameters of the noise-added health indicator collection value. This helps to accurately select the preset health indicators that have an important impact on the target health level in the training of the multi-dimensional health status indicator grading model, and improve the accuracy and reliability of the model.

[0156] In an embodiment of the present invention, the target preset health indicator has a target preset health indicator value; the conditional confusion degree corresponding to the target preset health indicator is determined based on the overall parameters of the noisy health indicator collection value, the number of indicators existing in the original health indicator collection value set and the inference proportion associated with the target preset health indicator for each preset health indicator value under the target preset health indicator, and the package can be implemented through the following examples.

[0157] Determine target test data including the preset health indicator value of the target preset health indicator in the test data set to obtain a target test data set;

[0158] For the target test data including the target health level type and the target preset health indicator value in the target test data set, based on the inferred proportion of each of the test data, determine the number of co-existences of the target test data including the target health level type and the target preset health indicator value in the original health indicator collection value set;

[0159] Determine a first confidence level corresponding to the target preset health indicator value based on a division operation result of the co-existence number and the number of indicators existing in the target preset health indicator value in the key data set; the first confidence level indicates the confidence level of the target health level type appearing in the key data set when the target preset health indicator value appears;

[0160] Determining a second confidence level that the target preset health indicator value appears in the key data set based on a division operation result of the co-existence number and the overall parameter of the noisy health indicator collection value;

[0161] Determining the conditional disorder degree of the target preset health indicator under the target health level type based on the second confidence level of the target health level type appearing in the key data set and the first confidence level corresponding to each preset health indicator value under the target preset health indicator;

[0162] Based on the conditional confusion degree of the target preset health indicator under each health state type in the preset health state type set, the conditional confusion degree corresponding to the target preset health indicator is determined.

[0163] In an embodiment of the present invention, exemplarily, in the above method, with the server as the execution subject, based on the overall parameters of the noised health indicator collection value, the number of indicators existing in the original health indicator collection value set for each preset health indicator value under the target preset health indicator, and the inference ratio associated with the target preset health indicator, the step of determining the conditional chaos degree corresponding to the target preset health indicator can be implemented in the following manner: Target test data determination: The server determines the target test data including the preset health indicator value of the target preset health indicator in the test data set. This can be achieved by searching for test data with a specific preset health indicator value in the test data set. Co-existence number determination: For the target test data including the target health level type and the target preset health indicator value in the target test data set, the server determines the co-existence number of the target test data including the target health level type and the target preset health indicator value in the original health indicator collection value set based on the inference ratio of each test data. The co-existence number represents the number of test data with the same target health level type and target preset health indicator value as the target test data in the original health indicator collection value set. First confidence determination: Based on the result of the division operation between the number of co-existences and the number of indicators existing in the target preset health indicator value in the key data set, the server determines the first confidence corresponding to the target preset health indicator value. The first confidence indicates the confidence that the target health level type appears when the target preset health indicator value appears in the key data set. Second confidence determination: Based on the result of the division operation between the number of co-existences and the overall parameter of the noise-added health indicator collection value, the server determines the second confidence that the target preset health indicator value appears in the key data set. The second confidence indicates the confidence of the overall parameter of the noise-added health indicator collection value when the target preset health indicator value appears in the key data set. Conditional chaos determination: Based on the second confidence that the target health level type appears in the key data set and the first confidence corresponding to each preset health indicator value under the target preset health indicator, the server determines the conditional chaos of the target preset health indicator under the target health level type. The conditional chaos indicates the degree of chaos of the target preset health indicator under the condition that different preset health indicator values ​​appear under the target health level type. Final conditional chaos degree determination: The server determines the conditional chaos degree corresponding to the target preset health indicator based on the conditional chaos degree of the target preset health indicator under each health state type in the preset health state type set. The final conditional chaos degree comprehensively considers the conditional chaos degree of the target preset health indicator under different health state types, and reflects the comprehensive impact of the target preset health indicator on the target health level.

[0164] The following is a specific scenario example, with the server as the execution subject, and the execution process of the above steps is described in detail: Assume that the server receives a batch of health indicator data, including an original health indicator collection value set, a noisy health indicator collection value set, and a target preset health indicator and a target health level type. Target test data determination: The server searches for test data with a target preset health indicator value (for example, a blood pressure value of 120 mmHg) in the test data set. After finding the target test data, these data are used as the target test data set. Co-existence number determination: For the target test data including the target health level type and the target preset health indicator value in the target test data set, the server calculates the co-existence number of these data in the original health indicator collection value set. Assume that in the original health indicator collection value set, there are 80 test data with the same target health level type and target preset health indicator value as the target test data. First confidence determination: The server divides the co-existence number by the quotient of the number of indicators existing in the target preset health indicator value in the key data set to determine the first confidence corresponding to the target preset health indicator value. Assuming that the number of indicators that the target preset health indicator value exists in the key data set is 100, the first confidence is 0.8. Determination of the second confidence: The server divides the number of common existences by the quotient of the overall parameter of the noise-added health indicator collection value to determine the second confidence that the target preset health indicator value appears in the key data set. Assuming that the overall parameter of the noise-added health indicator collection value is 20, the second confidence is 4. Determination of conditional chaos: The server determines the conditional chaos of the target preset health indicator under the target health level type according to the second confidence of the target health level type appearing in the key data set and the first confidence corresponding to each preset health indicator value under the target preset health indicator. Assuming that the second confidence of the target health level type appearing in the key data set is 0.6, and the first confidence corresponding to each preset health indicator value under the target preset health indicator is 0.2, 0.3 and 0.5 respectively, the conditional chaos is 0.3. Determination of the final conditional chaos: The server comprehensively considers the conditional chaos of the target preset health indicator under each health state type in the preset health state type set, and determines the conditional chaos corresponding to the target preset health indicator. Assuming that the conditional confusion of the target preset health indicator under different health status types is 0.4, 0.3 and 0.2 respectively, the final conditional confusion is 0.35.

[0165] Through the above steps, the server can determine the conditional confusion corresponding to the target preset health indicator, which reflects the uncertainty of the target preset health indicator under different health status types and the degree of influence on the target health level. This helps to accurately select the preset health indicators that have an important impact on the target health level in the training of the multi-dimensional health status indicator grading model, and improve the accuracy and reliability of the model.

[0166] In an embodiment of the present invention, the critical influence degree of each preset health indicator in the preset health indicator set on the target health level is determined based on the number of existence of each type, the number of existence of each indicator and the overall parameters of the noise-added health indicator collection value in the noise-added health indicator collection value set, which can be implemented through the following examples.

[0167] Determine a deviation statistic between the target preset health indicator and the preset health status type set based on the number of each type, the number of each indicator and the overall parameter of the noise-added health indicator collection value in the noise-added health indicator collection value set; the deviation statistic indicates the correlation between the target preset health indicator and the preset health status type set;

[0168] Based on the deviation statistics between the target preset health indicator and the preset health status type set, the critical influence degree of the target preset health indicator on the target health level is determined.

[0169] In an embodiment of the present invention, exemplarily, in the above method, with the server as the execution subject, based on the number of each type, the number of each indicator, and the overall parameters of the noise-added health indicator collection value in the noise-added health indicator collection value set, the step of determining the critical influence of each preset health indicator in the preset health indicator set on the target health level can be implemented in the following manner:

[0170] Determination of deviation statistics: The server obtains the number of each type and the number of each indicator. The number of types indicates the number of times the target health level type appears in the original health indicator collection value set, and the number of indicators indicates the number of times the target preset health indicator appears in the original health indicator collection value set. At the same time, the server obtains the overall parameters of the noisy health indicator collection value in the noisy health indicator collection value set. Next, the server determines the deviation statistics between the target preset health indicator and the preset health status type set based on these parameters. The deviation statistics can reflect the degree of difference between the target preset health indicator and the preset health status type set. Determination of critical impact degree: The server determines the critical impact degree of the target preset health indicator on the target health level based on the deviation statistics between the target preset health indicator and the preset health status type set. Specifically, it can be determined by calculating the correlation coefficient between the deviation statistics and the critical impact degree. The value range of the correlation coefficient is [-1,1]. When the correlation coefficient is 1, it indicates that there is a strong positive correlation between the deviation statistic and the key impact degree, that is, the difference between the target preset health indicator and the preset health status type set has a very significant impact on the target health level; when the correlation coefficient is -1, it indicates that there is a strong negative correlation between the deviation statistic and the key impact degree, that is, the difference between the target preset health indicator and the preset health status type set has a very insignificant impact on the target health level; when the correlation coefficient is 0, it indicates that there is no linear correlation between the deviation statistic and the key impact degree. Finally, the server determines the key impact degree of the target preset health indicator on the target health level based on the size of the correlation coefficient. Generally speaking, the larger the correlation coefficient, the higher the key impact degree of the target preset health indicator on the target health level.

[0171] The following is a specific scenario example, with the server as the execution subject, and describes the execution process of the above steps in detail: Assume that the server receives a batch of health indicator data, including the original health indicator collection value set, the noisy health indicator collection value set, the preset health state type set and the target preset health indicator. Deviation statistic value determination: The server obtains the number of times the target health level type appears in the original health indicator collection value set (the number of types), for example, the "normal" type appears 800 times, and the "abnormal" type appears 200 times. At the same time, the server obtains the number of times the target preset health indicator appears in the original health indicator collection value set (the number of indicators), for example, the preset health indicator value of the "blood pressure" indicator is 120mmHg, which appears 600 times. In addition, the server obtains the overall parameters of the noisy health indicator collection value in the noisy health indicator collection value set, such as the standard deviation of the noisy health indicator collection value. Next, the server calculates the deviation statistic between the target preset health indicator and the preset health state type set based on these parameters. Assume that the deviation statistic of the "blood pressure" indicator is 0.3. Determination of critical impact degree: The server calculates the correlation coefficient between the deviation statistic (0.3) between the target preset health indicator and the preset health status type set and the critical impact degree. Assume that the correlation coefficient is 0.6. Finally, the server determines the critical impact degree of the target preset health indicator on the target health level based on the size of the correlation coefficient. Since the correlation coefficient is 0.6, it is a medium-strength correlation, so it is determined that the critical impact degree of the "blood pressure" indicator on the target health level is medium. Through the above steps, the server can determine the deviation statistic between the target preset health indicator and the preset health status type set based on the overall parameters of the number of types, the number of indicators, and the noise-added health indicator collection value, and determine the critical impact degree of the target preset health indicator on the target health level based on the correlation between the deviation statistic and the critical impact degree. This helps to accurately select preset health indicators that have an important impact on the target health level in the training of the multi-dimensional health status indicator grading model, and improve the accuracy and reliability of the model.

[0172] In an embodiment of the present invention, the target preset health indicator has a target preset health indicator value; based on the number of existence of each type, the number of existence of each indicator and the overall parameters of the noisy health indicator collection value in the noisy health indicator collection value set, the deviation statistical value between the target preset health indicator and the preset health status type set is determined, which can be implemented through the following example.

[0173] Determine target test data including the preset health indicator value of the target preset health indicator in the test data set to obtain a target test data set;

[0174] For the target test data including the target health level type and the target preset health indicator value in the target test data set, an integration operation is performed on the number of types of the target health level type existing in the original health indicator collection value set and the number of indicators of the target preset health indicator value existing in the original health indicator collection value set to obtain an integration parameter corresponding to the target test data including the target health level type and the target preset health indicator value;

[0175] Determining an average value of the integrated parameter corresponding to the target test data including the target health level type and the target preset health indicator value based on a division operation result of the integrated parameter and the overall parameter of the noise-added health indicator acquisition value;

[0176] Determine a deviation statistic of the target preset health indicator under the target health level type based on the number of co-existences of the target test data including the target health level type and the target preset health indicator value in the original health indicator collection value set and the average value of the integrated parameter;

[0177] Based on the deviation statistics of the target preset health indicator under each health state type in the preset health state type set, a deviation statistics corresponding to the target preset health indicator is determined.

[0178] In an embodiment of the present invention, exemplarily, in the above method, with the server as the execution subject, based on the number of each type, the number of each indicator and the overall parameters of the noisy health indicator collection value in the noisy health indicator collection value set, the step of determining the deviation statistic value between the target preset health indicator and the preset health state type set can be implemented in the following manner: Target test data determination: The server determines the target test data including the preset health indicator value of the target preset health indicator in the test data set. This can be achieved by searching for test data with a specific preset health indicator value in the test data set. Integration parameter determination: For the target test data including the target health level type and the target preset health indicator value in the target test data set, the server performs an integration operation on the number of types existing in the original health indicator collection value set of the target health level type and the number of indicators existing in the original health indicator collection value set of the target preset health indicator value, and obtains an integration parameter corresponding to the target test data. The integration operation is used to comprehensively consider the occurrence of the target health level type and the target preset health indicator value in the data set. Average value determination: Based on the division operation result of the integration parameter and the overall parameters of the noisy health indicator collection value, the server determines the average value of the integration parameter corresponding to the target test data. The average value reflects the concentration of the integrated parameters in the data set. Determination of the deviation statistic: Based on the number of common existences of the target test data in the original health indicator collection value set and the average value of the integrated parameters, the server determines the deviation statistic of the target preset health indicator under the target health level type. The deviation statistic represents the degree of dispersion of the target preset health indicator under the target health level type. Determination of the final deviation statistic: Based on the deviation statistic of the target preset health indicator under each health state type in the preset health state type set, the server determines the deviation statistic corresponding to the target preset health indicator. The final deviation statistic comprehensively considers the deviation of the target preset health indicator under different health state types.

[0179] The following is a specific scenario example, taking the server as the execution subject, and describes in detail the execution process of the above steps:

[0180] Assume that the server receives a batch of health indicator data, including the original health indicator collection value set, the noisy health indicator collection value set, the preset health state type set and the target preset health indicator. Target test data determination: The server searches for target test data with a preset health indicator value of 120 mmHg in the test data set. After finding the target test data, these data are used as the target test data set. Integration parameter determination: For the target test data including the target health level type and the target preset health indicator value in the target test data set, the server integrates the number of times the target health level type appears in the original health indicator collection value set (type existence number) and the number of times the target preset health indicator value appears in the original health indicator collection value set (indicator existence number). Assume that the target health level type is "normal", the number of types of "normal" in the original health indicator collection value set is 800, the target preset health indicator value is 120 mmHg, and the number of indicators of "120 mmHg" in the original health indicator collection value set is 600. Then the integration parameter is 800*600=480000. Determine the average value: The server divides the integrated parameter by the overall parameter of the noisy health indicator collection value to obtain the average value of the integrated parameter corresponding to the target test data. Assume that the overall parameter of the noisy health indicator collection value is 100, the integrated parameter is 480000, and the average value is 4800. Determine the deviation statistic: The server divides the number of times the target test data appears in the original health indicator collection value set (the number of co-existences) by the average value of the integrated parameter to obtain the deviation statistic of the target preset health indicator under the target health level type. Assume that the number of times the target test data appears in the original health indicator collection value set is 400, and the average value of the integrated parameter is 4800, then the deviation statistic is 400 / 4800=0.0833. Determine the final deviation statistic: The server comprehensively considers the deviation statistic of the target preset health indicator under each health state type in the preset health state type set to obtain the deviation statistic corresponding to the target preset health indicator. Assume that the deviation statistic of the target preset health indicator under the "normal" health status type is 0.1, and the deviation statistic under the "abnormal" health status type is 0.2. Then the final deviation statistic of the target preset health indicator is 0.1*0.8+0.2*0.2=0.12. Through the above steps, the server can determine the deviation statistic between the target preset health indicator and the preset health status type set, reflecting the distribution of the target preset health indicator under different health status types. This helps to accurately select preset health indicators that have an important impact on the target health level in the training of the multi-dimensional health status indicator grading model, and improve the accuracy and reliability of the model.

[0181] In the embodiment of the present invention, the extracting of health status features from the preset health indicator set based on the key impact degree of each preset health indicator in the preset health indicator set can be implemented through the following examples.

[0182] Arrange the positions of each preset health indicator in the preset health indicator set according to their respective key influence levels, and extract a preset number of preset health indicators as health status features according to the position arrangement results; or,

[0183] According to a preset critical impact degree threshold, preset health indicators higher than the critical impact degree threshold are extracted from the preset health indicator set as health status features.

[0184] In the embodiment of the present invention, exemplary, in the above method, with the server as the execution subject, based on the key influence degree of each preset health indicator in the preset health indicator set, the step of extracting the health status feature from the preset health indicator set can be implemented in the following two ways: Position arrangement and extraction: The server arranges the position of each preset health indicator in the preset health indicator set according to their respective corresponding key influence degrees. The key influence degree can be a numerical value obtained by calculation or analysis, reflecting the importance of the preset health indicator to the target health level. Then, according to the position arrangement result, a preset number of preset health indicators are extracted as health status features. The selection of the preset number can be determined according to specific needs or algorithms, for example, extracting the first few preset health indicators with higher key influence degrees. Threshold extraction: The server extracts the preset health indicators above the threshold from the preset health indicator set as health status features according to the preset key influence degree threshold. The preset key influence degree threshold can be a preset fixed value or a dynamic threshold determined according to data distribution and feature importance. Preset health indicators above the threshold are generally considered to have a more significant impact on the target health level, and are therefore selected as health status features.

[0185] The following is a specific scenario example, taking the server as the execution subject, and describes in detail the execution process of the above steps:

[0186] Assume that the server receives a batch of health indicator data and calculates the critical impact level of each preset health indicator.

[0187] Position arrangement and extraction:

[0188] The server arranges the preset health indicators in order of their critical impact from high to low. Assume that the numerical range of the critical impact is from 0 to 1 and they have been arranged in this order. Then, the server extracts the first three preset health indicators as health status features based on the position arrangement results. These three preset health indicators are considered to be the most important features affecting the target health level. Threshold extraction: The server sets the preset critical impact threshold to 0.5. Then, the server extracts the preset health indicators with a critical impact higher than 0.5 from the preset health indicator set. In this example, the preset health indicators with a critical impact greater than 0.5 are extracted as health status features. These indicators are considered to play a key role in the assessment of the target health level.

[0189] Through the above two methods, the server can select representative or important preset health indicators as health status features according to the key impact of the preset health indicators. These features can be further used for training of multi-dimensional health status indicator classification models or other related analyses to improve the accuracy of the model and the understanding of health status.

[0190] The embodiment of the present invention provides a computer device 100, which includes a processor and a non-volatile memory storing computer instructions. When the computer instructions are executed by the processor, the computer device 100 executes the user health status monitoring method based on the smart wearable device. Figure 2 As shown, Figure 2 The block diagram of the computer device 100 provided in the embodiment of the present invention. The computer device 100 includes a memory 111, a processor 112 and a communication unit 113. To achieve data transmission or interaction, the memory 111, the processor 112 and the communication unit 113 are directly or indirectly electrically connected to each other. For example, these components can be electrically connected to each other through one or more communication buses or signal lines.

[0191] For illustrative purposes, the foregoing description is made with reference to specific embodiments. However, the above illustrative discussion is not intended to be exhaustive or to limit the present disclosure to the precise form disclosed. Numerous modifications and variations are possible in accordance with the above teachings. These embodiments are selected and described in order to best illustrate the principles of the present disclosure and its practical application, so that those skilled in the art can best utilize the present disclosure and utilize various embodiments with different modifications to suit the intended specific application.

Claims

1. A user health status monitoring method based on smart wearable devices, characterized in that: include: Responding to the current operating mode of the smart wearable device selected by the user; Determining a data collection strategy for the smart wearable device according to the current operating mode; Based on the data collection strategy, target physiological data of the target user is collected through the corresponding sensor of the smart wearable device; Inputting the target physiological data into a pre-trained multi-dimensional health status index grading model to obtain a target health grade corresponding to the target physiological data; Acquire a plurality of target physiological data and corresponding target health levels having a time sequence relationship according to a collection period corresponding to the data collection strategy; A health status monitoring report for the target user is constructed based on the multiple target physiological data having a time-series relationship and the corresponding target health levels.

2. The method according to claim 1, characterized in that The multidimensional health status index grading model is obtained by the following method, including: The sample physiological data instance and multiple health states are respectively identified by using a multi-dimensional health state index classification model to obtain corresponding instance features and multiple health state features; Performing a masking operation on the remaining health status features among the multiple health status features to obtain masked remaining health status features; Perform multi-source data integration on the masked remaining health state feature and the health state feature to obtain a multi-health state integrated feature, and perform multi-source data integration on the health state feature and the instance feature to obtain a first multi-source integrated feature; Perform a combination operation on the multi-health state integrated feature and the first multi-source integrated feature to obtain an integrated feature corresponding to the health state; or The multi-health state integrated feature is multiplied by a preset multi-health state integrated feature influence coefficient to obtain a first intermediate feature; the first intermediate feature and the first multi-source integrated feature are combined to obtain an integrated feature corresponding to the health state, wherein the multi-health state integrated feature influence coefficient is not lower than a first threshold value, and the multi-health state integrated feature influence coefficient is not higher than a second threshold value; Wherein, in the multi-source data integration execution stage, the influence coefficient of the health status feature is higher than the influence coefficient of the remaining health status features; For each of the health status characteristics, integrating or combining the integration characteristics corresponding to the health status characteristics with the health status characteristics multi-source data to obtain the target integration characteristics corresponding to the health status characteristics; The step of grading the sample physiological data instance into a plurality of health states according to the target integration feature to obtain a health level includes: For each of the health states, grading the health states of the sample physiological data instances according to the target integration features corresponding to the health states, to obtain the health grade corresponding to the health states; Through the multidimensional health status indicator grading model, the sample physiological data instance is graded into multiple health states according to the target integration characteristics to obtain a health level, and according to the health level and the instance type of the sample physiological data instance, the network parameters of the multidimensional health status indicator grading model are adjusted until a trained multidimensional health status indicator grading model is obtained.

3. The method according to claim 2, characterized in that The number of the instance features is multiple. After the sample physiological data instance and multiple health states are respectively identified by the multidimensional health state index classification model to obtain corresponding instance features and multiple health state features, the method further includes: For each of the instance features, performing multi-source data integration on the instance feature and the plurality of health status features to obtain a sample integration feature; The step of grading the sample physiological data instance into a plurality of health states according to the target integration feature to obtain a health level includes: The sample physiological data instance is graded into a plurality of health states according to the sample integration feature and the target integration feature to obtain a health grade.

4. The method according to claim 3, characterized in that Each of the health states has a corresponding target integration feature, and the sample physiological data instance is graded into a plurality of health states according to the sample integration feature and the target integration feature to obtain a health grade, including: Perform multi-source data integration on the target integrated feature corresponding to the health state, the target integrated features corresponding to the remaining health states in the multiple health states, and the sample integrated feature to obtain the integrated feature to be used; The health status of the sample physiological data instance is graded according to the integrated feature to be used to obtain a health grade corresponding to the health status.

5. The method according to claim 2, characterized in that: The health status characteristics include a first health status characteristic and a second health status characteristic; For each of the health status features, performing multi-source data integration on the health status feature, the remaining health status features in the plurality of health status features, and the instance feature to obtain an integrated feature, including: For each of the health status features, a multiplication result of a second health status feature of the remaining health status features in the multiple health status features and a preset first multi-source integration feature influence coefficient is obtained, and a product feature is obtained according to the first health status feature of the health status feature and the multiplication result, wherein the first multi-source integration feature influence coefficient is not higher than a first threshold value; For each of the health status features, performing multi-source data integration on the first health status feature of the health status feature and the health status feature to obtain a health status integration sub-feature, and performing multi-source data integration on the first health status feature of the health status feature and the instance feature to obtain an instance integration sub-feature; Perform a combination operation on the product feature, the health status integrated sub-feature and the instance integrated sub-feature to obtain an integrated feature corresponding to the health status; or, The product feature is multiplied by a preset product feature influence coefficient to obtain a second intermediate feature; the second intermediate feature, the health status integration sub-feature and the instance integration sub-feature are combined to obtain an integrated feature corresponding to the health status, wherein the product feature influence coefficient is not lower than a first threshold value, and the product feature influence coefficient is not higher than a second threshold value.

6. The method according to claim 2, characterized in that The number of the health status features is not less than three, and for each of the health status features, the health status feature is integrated with the remaining health status features in the multiple health status features and the instance feature to obtain an integrated feature, including: For each of the health status characteristics, selecting a preset number of remaining health status characteristics from a plurality of remaining health status characteristics in the plurality of health status characteristics as first remaining health status characteristics, and selecting the remaining remaining health status characteristics in the plurality of remaining health status characteristics as second remaining health status characteristics; performing a masking operation on the first remaining health state feature to obtain a masked first remaining health state feature; Perform multi-source data integration on the health state feature, the masked first remaining health state feature, and the second remaining health state feature to obtain a health state mask feature, and perform multi-source data integration on the health state feature and the instance feature to obtain a second multi-source integrated feature; Perform a combination operation on the health status masking feature and the second multi-source integrated feature to obtain an integrated feature corresponding to the health status; or, The health status masking feature is multiplied by the influence coefficient of the preset health status integration sub-feature to obtain a third intermediate feature; the third intermediate feature is combined with the second multi-source integration feature to obtain an integration feature corresponding to the health status; The health status integration sub-feature influence coefficient is not lower than a first threshold, and the health status integration sub-feature influence coefficient is not higher than a second threshold.

7. The method according to claim 2, characterized in that The method further comprises: Acquire corresponding noisy health index collection values ​​sent by multiple test smart wearable devices to obtain a set of noisy health index collection values; each of the noisy health index collection values ​​is data after the original health index collection value of the corresponding test smart wearable device is noisy according to the noisy strategy; the noisy strategy includes a first noisy strategy and a second noisy strategy; the first noisy strategy is that the element value in the original health index collection value is maintained in its original state after noisy based on a first preset probability; the second noisy strategy is that the element value in the original health index collection value is changed to a preset value after noisy based on a second preset probability; the preset value is a value indicating that the key data collected by the corresponding test smart wearable device exists in the targeted test data; Acquire a test data set, the test data set includes multiple test data, each test data includes any preset health indicator value of any preset health indicator in a preset health indicator set, and any health status type in a preset health status type set; the noisy health indicator collection value and the original health indicator collection value are in the form of a preset matrix, and the element value in the preset matrix indicates whether the key data collected by the corresponding test smart wearable device exists in the test data; the preset health status type set includes a target health level type; the preset health indicator set includes a target preset health indicator; For each test data in the test data set, extract the associated element value corresponding to the test data from each noisy health indicator collection value in the noisy health indicator collection value set; the position arrangement of each associated element value in the corresponding noisy health indicator collection value corresponds to the position arrangement of the test data in the test data set; For each of the multiple associated element values, determining whether the test data indicated by the targeted associated element value is the targeted test data, and obtaining a recognition result corresponding to the targeted associated element value; Obtaining a first transition parameter based on the corresponding identification results of each of the associated element values, the first noise adding strategy, and the data scale of the noisy health indicator collection values ​​in the noisy health indicator collection value set; Obtaining a second transition parameter based on a subtraction result of the first noise adding strategy and the second noise adding strategy; Based on the result of the division operation of the first transition parameter and the second transition parameter, the inferred proportion of the targeted test data in the key data set is obtained, and the inferred proportion of the targeted test data in the original health indicator collection value set is determined; the original health indicator collection value set includes the original health indicator collection values ​​of each of the multiple tested smart wearable devices; Based on the inferred proportion of each of the test data, determining the number of types of the target health level type existing in the key data set; Based on the inferred proportion of each of the test data, determine the target preset health indicator A k The number of indicators existing in the key data set for each preset health indicator value below; Determine the critical influence of each preset health indicator in the preset health indicator set on the target health level based on the number of each type, the number of each indicator, and the overall parameters of the noise-added health indicator collection value in the noise-added health indicator collection value set; Arrange the positions of each preset health indicator in the preset health indicator set according to their respective key influence levels, and extract a preset number of preset health indicators as health status features according to the position arrangement results; or, According to a preset critical impact degree threshold, preset health indicators higher than the critical impact degree threshold are extracted from the preset health indicator set as health status features; the health status features are used to train a multidimensional health status indicator grading model for the target health level.

8. The method according to claim 7, characterized in that The target preset health indicator has a target preset health indicator value, and determining the critical influence degree of each preset health indicator in the preset health indicator set on the target health level based on the number of each type, the number of each indicator, and the overall parameter of the noise-added health indicator collection value in the noise-added health indicator collection value set includes: Determine the information confusion degree corresponding to the preset health status type set based on the number of types of the target health level type existing in the original health indicator collection value set and the overall parameter of the noisy health indicator collection value; the information confusion degree represents an average trend indicating the probability of the health status type existing in the preset health status type set; Determine target test data including the preset health indicator value of the target preset health indicator in the test data set to obtain a target test data set; For the target test data including the target health level type and the target preset health indicator value in the target test data set, based on the inferred proportion of each of the test data, determine the number of co-existences of the target test data including the target health level type and the target preset health indicator value in the original health indicator collection value set; Determine a first confidence level corresponding to the target preset health indicator value based on a division operation result of the co-existence number and the number of indicators existing in the target preset health indicator value in the key data set; the first confidence level indicates the confidence level of the target health level type appearing in the key data set when the target preset health indicator value appears; Determining a second confidence level that the target preset health indicator value appears in the key data set based on a division operation result of the co-existence number and the overall parameter of the noisy health indicator collection value; Determining the conditional disorder degree of the target preset health indicator under the target health level type based on the second confidence level of the target health level type appearing in the key data set and the first confidence level corresponding to each preset health indicator value under the target preset health indicator; Based on the conditional confusion degree of the target preset health indicator under each health state type in the preset health state type set, determining the conditional confusion degree corresponding to the target preset health indicator; the conditional confusion degree indicates the average situation of the probability of the target preset health indicator existing under the condition of the preset health state type set; Based on the information confusion degree and the condition confusion degree, the critical influence degree of the target preset health indicator on the target health level is determined.

9. The method according to claim 7, characterized in that: The target preset health indicator has a target preset health indicator value, and determining the critical influence degree of each preset health indicator in the preset health indicator set on the target health level based on the number of each type, the number of each indicator, and the overall parameter of the noise-added health indicator collection value in the noise-added health indicator collection value set includes: Determine target test data including the preset health indicator value of the target preset health indicator in the test data set to obtain a target test data set; For the target test data including the target health level type and the target preset health indicator value in the target test data set, an integration operation is performed on the number of types of the target health level type existing in the original health indicator collection value set and the number of indicators of the target preset health indicator value existing in the original health indicator collection value set to obtain an integration parameter corresponding to the target test data including the target health level type and the target preset health indicator value; Determining an average value of the integrated parameter corresponding to the target test data including the target health level type and the target preset health indicator value based on a division operation result of the integrated parameter and the overall parameter of the noise-added health indicator acquisition value; Determine a deviation statistic of the target preset health indicator under the target health level type based on the number of co-existences of the target test data including the target health level type and the target preset health indicator value in the original health indicator collection value set and the average value of the integrated parameter; Based on the deviation statistics of the target preset health indicator under each health state type in the preset health state type set, determining the deviation statistics corresponding to the target preset health indicator; the deviation statistics indicate the correlation between the target preset health indicator and the preset health state type set; Based on the deviation statistics between the target preset health indicator and the preset health status type set, the critical influence degree of the target preset health indicator on the target health level is determined.

10. A server system, characterized in that: The method comprises a server, wherein the server is used to execute the method described in any one of claims 1 to 9.

Images