A drinking health risk assessment system, electronic device and storage medium

Abstract

The present invention provides a health risk assessment system after drinking, an electronic device and a storage medium, which relate to the technical field of data processing, wherein a data acquisition module acquires a PPG signal waveform and physiological data; a feature extraction module extracts five-dimensional time-frequency characteristic values ​​of the PPG signal waveform and physiological characteristic values ​​of the physiological data respectively; a data comparison module compares the five-dimensional time-frequency characteristic values ​​and the physiological characteristic values ​​with historical health data respectively to generate initial risk data; a drunkenness assessment module inputs the five-dimensional time-frequency characteristic values ​​into multiple decision tree sub-models in an integrated classification model to obtain multiple drunkenness degree classification results; and the drunkenness degree of a monitored object is determined according to multiple drunkenness degree classification results; and a risk assessment module generates a health risk level according to the initial risk data and the drunkenness degree. The monitored object wears an intelligent wearable device to achieve real-time monitoring, so that the monitored object can know the health risk level and avoid danger to its health due to drinking.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a drinking health risk assessment system, electronic equipment and storage medium. Background Art

[0002] According to statistics, the overall drinking rate of adults has exceeded 50%, and alcohol consumption is still on the rise every year. Every year, about 3 million people die worldwide due to problems related to drinking. Therefore, it is necessary to prevent the risks of excessive drinking.

[0003] At present, the test methods for the alcohol content of the human body after drinking are generally carried out through breath testing, infrared tester testing, saliva testing, etc. The above common testing methods basically require specific testing devices or testing instruments to achieve the test, are only applicable to specific usage scenarios, and have a certain lag. Summary of the invention

[0004] The problem solved by the present invention is that the existing testing method is only applicable to specific usage scenarios and has a certain hysteresis.

[0005] In order to solve the above problems, the present invention provides a drinking health risk assessment system, an electronic device and a storage medium.

[0006] In a first aspect, the present invention provides a drinking health risk assessment system, comprising: A data acquisition module is used to obtain the current PPG signal waveform and physiological data of the monitored object collected by the smart wearable device; A feature extraction module, used to extract five-dimensional time-frequency feature values ​​of the PPG signal waveform and physiological feature values ​​of the physiological data, wherein the five-dimensional time-frequency feature values ​​include main wave amplitude, dicrotic wave amplitude, beat time ratio, frequency concentration value and concentrated frequency amplitude value; A data comparison module, used to compare the five-dimensional time-frequency characteristic value and the physiological characteristic value with historical health data to generate initial risk data, wherein the historical health data refers to historical data corresponding to the current five-dimensional time-frequency characteristic value and the physiological characteristic value type; a drunkenness assessment module, configured to input the five-dimensional time-frequency feature values ​​into a plurality of decision tree sub-models in an integrated classification model to obtain a plurality of drunkenness degree classification results, wherein the integrated classification model includes a plurality of decision tree sub-models, and different decision tree sub-models input the five-dimensional time-frequency feature values ​​in different types and quantities; and further determine the drunkenness degree of the monitored object according to the plurality of drunkenness degree classification results, wherein the integrated classification model performs voting statistics on the plurality of drunkenness degree classification results and outputs the drunkenness degree of the monitored object; The risk assessment module is used to generate a health risk level according to the initial risk data and the degree of drunkenness.

[0007] Optionally, the decision tree sub-model includes a plurality of split nodes, each of the split nodes has two branches, the two branch ends of each split node are another split node or an output classification result, and each split node corresponds to a split parameter and a five-dimensional time-frequency feature value; at each split node, the corresponding five-dimensional time-frequency feature value is compared with the corresponding split parameter to determine the branch direction of each split node; The number of the five-dimensional time-frequency feature values ​​input into each decision tree sub-model is less than or equal to the number of the splitting nodes, and each of the input five-dimensional time-frequency feature values ​​appears at least once in a plurality of the splitting nodes.

[0008] Optionally, the feature extraction module includes: A time domain analysis unit, configured to perform time domain analysis on the PPG signal waveform to determine the main wave amplitude, the dicrotic wave amplitude and the beat time ratio; A frequency domain analysis unit is used to perform frequency domain analysis on the PPG signal waveform to determine the frequency concentration value and the concentrated frequency amplitude value.

[0009] Optionally, the alcohol-related health risk assessment system further includes: a modeling module, the modeling module including: a training set construction unit, configured to obtain the PPG signal waveforms of a plurality of the monitored subjects, and determine the five-dimensional time-frequency feature values ​​of each of the monitored subjects, wherein a group of the five-dimensional time-frequency feature values ​​of each of the monitored subjects and the actual drunkenness degree of the monitored subjects corresponding to each group of the five-dimensional time-frequency feature values ​​are taken as a sample data, and a plurality of the sample data constitute a training set; and further configured to extract the sample data from the training set with replacement, and respectively construct a plurality of training subsets, wherein the plurality of the training subsets correspond one-to-one to the plurality of the decision tree sub-models, and each of the training subsets is used to train the corresponding decision tree sub-model; A model training unit is used to train the corresponding decision tree sub-model using the sample data in the training subset until the output result of the decision tree sub-model is consistent with the actual drunkenness degree in the sample data, and determine the splitting parameter of the decision tree sub-model.

[0010] Optionally, the model training unit uses the sample data in the training subset to train the corresponding decision tree sub-model, including: Extracting the five-dimensional time-frequency feature values ​​from the sample data in each of the training subsets, wherein the types and quantities of the five-dimensional time-frequency feature values ​​extracted from the same training subset remain consistent, and the types and quantities of the five-dimensional time-frequency feature values ​​extracted from different training subsets are different; The extracted five-dimensional time-frequency feature values ​​are input into the corresponding decision tree sub-model, and the output result of the decision tree sub-model is compared with the corresponding actual drunkenness degree, and the splitting parameter of the decision tree sub-model is adjusted.

[0011] Optionally, before extracting the sample data from the training set with replacement to construct a plurality of training subsets respectively, the training set construction unit is further used to expand the sample data in the training set to obtain virtual sample data so as to balance the number of sample data corresponding to the various drunkenness degrees.

[0012] Optionally, the data comparison module includes: A data processing unit, used to generate a historical time-frequency mean and a historical physiological mean according to the historical health data; A deviation comparison unit, used to compare the five-dimensional time-frequency characteristic value with the historical time-frequency mean, and the physiological characteristic value with the historical physiological mean, to generate corresponding deviation data; The anomaly detection unit is used to perform anomaly detection on the deviation data using an isolation forest algorithm to generate the initial risk data.

[0013] Optionally, the risk assessment module generates a health risk level according to the initial risk data and the degree of drunkenness, including: The initial risk data and the drunkenness degree are normalized to generate a health risk index, and the health risk level is determined based on the health risk index.

[0014] In a second aspect, the present invention provides an electronic device, including a memory and a processor; The memory is used to store computer programs; The processor is configured to implement the following steps when executing the computer program: Obtain the current PPG signal waveform and physiological data of the monitored object collected by the smart wearable device; Respectively extracting five-dimensional time-frequency characteristic values ​​of the PPG signal waveform and physiological characteristic values ​​of the physiological data, wherein the five-dimensional time-frequency characteristic values ​​include main wave amplitude, dicrotic wave amplitude, beat time ratio, frequency concentration value and concentrated frequency amplitude value; Comparing the five-dimensional time-frequency feature value and the physiological feature value with historical health data respectively to generate initial risk data, wherein the historical health data refers to historical data corresponding to the current five-dimensional time-frequency feature value and the physiological feature value type; Inputting the five-dimensional time-frequency feature values ​​into a plurality of decision tree sub-models in an integrated classification model to obtain a plurality of drunkenness degree classification results, wherein the integrated classification model includes a plurality of decision tree sub-models, and different decision tree sub-models input the five-dimensional time-frequency feature values ​​in different types and quantities; further determining the drunkenness degree of the monitored object based on the plurality of drunkenness degree classification results, wherein the integrated classification model performs voting statistics on the plurality of drunkenness degree classification results and outputs the drunkenness degree of the monitored object; A health risk level is generated based on the initial risk data and the degree of intoxication.

[0015] In a third aspect, the present invention provides a computer-readable storage medium, wherein a computer program is stored on the storage medium. When the computer program is executed by a processor, the following steps are implemented: Obtain the current PPG signal waveform and physiological data of the monitored object collected by the smart wearable device; Respectively extracting five-dimensional time-frequency characteristic values ​​of the PPG signal waveform and physiological characteristic values ​​of the physiological data, wherein the five-dimensional time-frequency characteristic values ​​include main wave amplitude, dicrotic wave amplitude, beat time ratio, frequency concentration value and concentrated frequency amplitude value; Comparing the five-dimensional time-frequency feature value and the physiological feature value with historical health data respectively to generate initial risk data, wherein the historical health data refers to historical data corresponding to the current five-dimensional time-frequency feature value and the physiological feature value type; Inputting the five-dimensional time-frequency feature values ​​into a plurality of decision tree sub-models in an integrated classification model to obtain a plurality of drunkenness degree classification results, wherein the integrated classification model includes a plurality of decision tree sub-models, and different decision tree sub-models input the five-dimensional time-frequency feature values ​​in different types and quantities; further determining the drunkenness degree of the monitored object based on the plurality of drunkenness degree classification results, wherein the integrated classification model performs voting statistics on the plurality of drunkenness degree classification results and outputs the drunkenness degree of the monitored object; A health risk level is generated based on the initial risk data and the degree of intoxication.

[0016] The beneficial effects of the alcohol-related health risk assessment system, electronic device and storage medium of the present invention are: The data acquisition module obtains the current PPG signal waveform and physiological data of the monitored object collected by the smart wearable device, so that the current physical data of the monitored object can be clearly understood, which is convenient for the subsequent health risk assessment after drinking. The feature extraction module extracts the five-dimensional time-frequency characteristic values ​​of the PPG signal waveform and the physiological characteristic values ​​of the physiological data, wherein the five-dimensional time-frequency characteristic values ​​include the main wave amplitude, the dicrotic wave amplitude, the beat time ratio, the frequency concentration value and the concentrated frequency amplitude value, which can facilitate the data comparison module to compare the five-dimensional time-frequency characteristic values ​​and the physiological characteristic values ​​with the historical health data, and make a preliminary assessment of the risk of drinking. The drunkenness assessment module then inputs multiple five-dimensional time-frequency feature values ​​from the five time-frequency feature values ​​into multiple decision tree sub-models in the integrated classification model to obtain multiple drunkenness degree classification results, wherein the integrated classification model includes multiple decision tree sub-models, and different decision tree sub-models input different types and quantities of five-dimensional time-frequency feature values, and each decision tree sub-model performs analysis based on multiple five-dimensional time-frequency feature values, and integrates the characteristics of multiple five-dimensional time-frequency feature values ​​to obtain drunkenness degree classification results, fully considering the mutual influence of multiple five-dimensional time-frequency feature values, and comprehensively analyzing to improve the analysis accuracy, and different decision tree sub-models The tree-based sub-model uses different types and numbers of five-dimensional time-frequency eigenvalues ​​to split and combine the five five-dimensional time-frequency eigenvalues ​​for analysis, which increases the diversity of the analysis and avoids the singleness of the analysis. It can also avoid the problem that an abnormal eigenvalue directly leads to inaccurate analysis results, thereby improving the accuracy of the analysis results; the drunkenness assessment module also determines the drunkenness of the monitored object based on multiple drunkenness degree classification results, wherein the integrated classification model votes and counts multiple drunkenness degree classification results, and outputs the drunkenness of the monitored object, thereby further improving the accuracy of the analysis results; the risk assessment module comprehensively considers the initial risk data and the degree of drunkenness to ensure the accuracy of the results, thereby determining the health risk level of the monitored object, and the obtained health risk level can be displayed in the smart wearable device or sent to the mobile terminal, etc., for the real-time viewing of the monitored object. Since the monitored object wears the smart wearable device in daily life, real-time monitoring of the monitored object is realized, so that the monitored object can view its own health risk level in real time in daily life or during drinking, avoiding the problem that the existing test method is only applicable to specific usage scenarios and has a certain lag, and at the same time serves as a reminder to the monitored object to avoid the danger of the monitored object's body due to drinking. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 A schematic diagram of the structure of a drinking health risk assessment system provided by an embodiment of the present invention; Figure 2 A waveform diagram of a real-time pulse wave provided by an embodiment of the present invention; Figure 3A schematic diagram of the structure of a first decision tree sub-model provided in an embodiment of the present invention; Figure 4 A schematic diagram of the structure of a second decision tree sub-model provided by an embodiment of the present invention; Figure 5 A schematic diagram of the structure of a waveform analysis module provided in an embodiment of the present invention; Figure 6 A verification experiment data table provided for an embodiment of the present invention; Figure 7 A schematic diagram of the structure of an electronic device application environment provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0018] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below in conjunction with the accompanying drawings. Although certain embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be interpreted as being limited to the embodiments described herein. On the contrary, these embodiments are provided to provide a more thorough and complete understanding of the present invention. It should be understood that the drawings and embodiments of the present invention are only for exemplary purposes and are not intended to limit the scope of protection of the present invention.

[0019] It should be understood that the various steps described in the method embodiments of the present invention may be performed in different orders and / or in parallel. In addition, the method embodiments may include additional steps and / or omit the steps shown. The scope of the present invention is not limited in this respect.

[0020] The term "including" and its variations used in this document are open inclusions, that is, "including but not limited to"; the term "based on" means "based at least in part on"; the term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one other embodiment"; the term "some embodiments" means "at least some embodiments"; the term "optionally" means "optional embodiments". The relevant definitions of other terms will be given in the following description. It should be noted that the concepts of "first", "second", etc. mentioned in the present invention are only used to distinguish different devices, modules or units, and are not used to limit the order or interdependence of the functions performed by these devices, modules or units.

[0021] It should be noted that the modifications of "one" and "plurality" mentioned in the present invention are illustrative rather than restrictive, and those skilled in the art should understand that, unless otherwise clearly indicated in the context, it should be understood as "one or more".

[0022] The names of the messages or information exchanged between multiple devices in the embodiments of the present invention are only used for illustrative purposes, and are not used to limit the scope of these messages or information.

[0023] In the related technology, common testing methods include breath testing, blood testing, infrared testing, saliva testing, etc. For example, an alcohol testing instrument is used to test the alcohol concentration of the test subject's breath; or blood is collected and the alcohol content in the blood is tested using a testing device. These testing methods usually require the use of specific equipment for measurement and are mostly used for drunk driving detection or clinical use, rather than for daily testing of physical health conditions, and it is difficult to achieve daily real-time measurement.

[0024] In view of the problems existing in the above-mentioned related technologies, this embodiment provides a drinking health risk assessment system, an electronic device and a storage medium.

[0025] like Figure 1 As shown, a drinking health risk assessment system provided by an embodiment of the present invention includes: The data acquisition module 100 is used to obtain the current PPG signal waveform and physiological data of the monitored object collected by the smart wearable device.

[0026] Specifically, the monitored object wears the smart wearable device on a daily basis. The smart wearable device can be a head-mounted wearable device or a wristwatch, etc. The smart wearable device is used to collect the physical data of the monitored object, and the PPG technology in the smart wearable device is used to pre-process the collected band or physiological data to obtain the overall PPG signal waveform (i.e., pulse wave waveform). Physiological data include the weight, height, body temperature and deep sleep duration of the monitored object. The smart wearable device can also be used to collect the weight, height, body temperature and deep sleep duration of the monitored object. For example, an accelerometer and a heart rate sensor are used to record the sleep cycle of the monitored object, and the deep sleep duration of each night is counted. The average deep sleep duration of the past week is calculated, and its standard deviation is recorded to evaluate the stability of sleep quality. The skin temperature sensor is used to collect the hourly body temperature data of the monitored object, and the sliding window method is used to calculate the average body temperature, maximum and minimum values, and standard deviation of the past 7 days to provide detailed information on body temperature fluctuations. Among them, the deep sleep duration is an important stage in the sleep cycle, which refers to the deep sleep duration of the previous day. The healthy time length of an adult in the deep sleep stage during night sleep is 3-5 hours. Body temperature is a very important factor in evaluating the health of the human body. The normal range is generally between 36.2℃ and 37.2℃.

[0027] The feature extraction module 200 is used to extract the five-dimensional time-frequency feature values ​​of the PPG signal waveform and the physiological feature values ​​of the physiological data, respectively, wherein the five-dimensional time-frequency feature values ​​include the main wave amplitude, the dicrotic wave amplitude, the beat time ratio, the frequency concentration value and the concentrated frequency amplitude value.

[0028] Specifically, the five-dimensional time-frequency eigenvalue is also a real-time pulse wave, such as Figure 2As shown, the main wave amplitude is Figure 2 The height value represented by h1 is the amplitude value of the main wave fluctuation. This characteristic value is used to characterize the cardiac ejection condition and ability. The main wave amplitude is referred to as characteristic 1 below; the dicrotic wave amplitude is Figure 2 The height value represented by h3+h4 is the dicrotic wave amplitude value, which is used to characterize vascular resistance and vascular conditions. The dicrotic wave amplitude value is referred to as feature 2 below; the beat time ratio is Figure 2 The ratio of the mid-dicrotic trough t3 / pulse wave period T, where the beat time ratio represents the fluctuation time of the heart cycle, and the beat time ratio is hereinafter referred to as feature 3. The PPG signal is converted into a frequency signal using continuous Fourier transform, and the formula for continuous Fourier transform is as follows: ; in, is the angular frequency; f(t) is the response function or real-time pulse wave function; is the frequency response function, which is the Fourier transform of f(t); j is the imaginary unit and t is the time.

[0029] After continuous Fourier transform, the relationship between the frequency domain phase and the frequency domain amplitude and the frequency is obtained, and then two characteristic values ​​related to the frequency are obtained: the frequency concentration value represents the frequency value corresponding to the most appearing PPG signal, hereinafter referred to as feature 4; the concentrated frequency amplitude value represents the frequency value corresponding to the largest number of frequency domain signals, hereinafter referred to as feature 5.

[0030] Physiological characteristic values ​​include body mass index (BMI), body temperature, and deep sleep duration. Body mass index (BMI) refers to an indicator that comprehensively measures the degree of human obesity based on the important contributing characteristics of height and weight. It can be determined by height and weight. For example, the body mass index is determined by the BMI formula. The BMI formula is as follows: BMI = W / (H) 2 ; Among them, BMI is body mass index, W is weight, and H is height.

[0031] The BMI is transformed into a Z-score through a standardized formula to adapt to the weight changes of different monitored subjects. The standardized formula includes: ; Among them, Z MBI is the standardized body mass index, is the mean BMI of all monitored subjects, is the standard deviation of all monitored objects.

[0032] The data comparison module 300 is used to compare the five-dimensional time-frequency feature values ​​and the physiological feature values ​​with the historical health data respectively to generate initial risk data, wherein the historical health data refers to the historical data corresponding to the current five-dimensional time-frequency feature values ​​and the physiological feature value types.

[0033] Specifically, historical health data refers to historical data corresponding to the current five-dimensional time-frequency characteristic value and physiological characteristic value type, that is, the historical five-dimensional time-frequency characteristic value and the historical physiological characteristic value. By comparing the five-dimensional time-frequency characteristic value and the physiological characteristic value with the historical health data, the deviation between the current characteristic value and the historical characteristic value can be clarified, thereby determining the outlier. The outlier refers to a sample that deviates significantly from other observed values ​​in the data set. When there are more outliers, it means that the degree of deviation between the monitored object's physical state after drinking and the preset safety standard is higher, and thus the monitored object may face a greater risk of drinking. Therefore, the initial risk data can be determined based on the outliers. The initial risk data may include no initial risk, initial low risk, initial medium risk and initial high risk, wherein no initial risk means that the abnormal scores of the abnormal values ​​are all lower than the preset abnormal threshold, and the probability of the abnormal value is lower than the first preset abnormal probability. The preset abnormal threshold can be set to 0.05, and the first preset abnormal probability can be set to 0.1; the initial low risk means that the number of abnormal values ​​with abnormal scores higher than the preset abnormal threshold is 1, and the probability of the abnormal value is higher than the second preset abnormal probability, and the second preset abnormal probability can be set to 0.3; the initial medium risk means that the number of abnormal values ​​with abnormal scores higher than the preset abnormal threshold is 2, and the probability of the abnormal value is higher than the third preset abnormal probability, and the third preset abnormal probability can be set to 0.6; the initial high risk means that the number of abnormal values ​​with abnormal scores higher than the preset abnormal threshold is 3, and the probability of the abnormal value is higher than the fourth preset abnormal probability, and the fourth preset abnormal probability can be set to 0.9. Among them, the anomaly score refers to the possibility of quantifying each sample to become an anomaly point, and the anomaly numerical probability refers to the probability value further calculated based on the anomaly score. It reflects the degree of deviation between the five-dimensional time-frequency characteristic values ​​and physiological characteristic values ​​and historical data. The anomaly score and the anomaly numerical probability can be determined according to the anomaly detection algorithm, for example, the tree-based unsupervised non-parametric algorithm such as Isolation Forest. Therefore, the initial risk data can be determined according to the division range of the anomaly value and the initial risk data.

[0034] The drunkenness assessment module 400 is used to input the five-dimensional time-frequency feature values ​​into multiple decision tree sub-models in the integrated classification model to obtain multiple drunkenness degree classification results, wherein the integrated classification model includes multiple decision tree sub-models, and different decision tree sub-models input the types and quantities of the five-dimensional time-frequency feature values.

[0035] Specifically, the above five features are input into the integrated classification model. After entering the integrated classification model, different numbers and types of features are respectively input into multiple decision tree sub-models, for example Figure 3 As shown, feature 1, feature 2, feature 3 and feature 4 are input into the first decision tree sub-model, as Figure 4 As shown, feature 1, feature 2, feature 4 and feature 5 are input into the second decision tree sub-model. Similarly, any three or any four of the five features can be input into other decision tree sub-models, so that different decision tree sub-models analyze and process different types and numbers of features. Each decision tree sub-model obtains a drunkenness degree classification result according to the input pulse feature value, thereby obtaining multiple drunkenness degree classification results.

[0036] It should be understood that due to Figure 3 and Figure 4 Each of them corresponds to a model, but this application includes multiple models (actually more than one hundred), and the results of each model are not exactly the same, and the model results can be one, two, three or four, so it is normal that the calculation result does not include "moderate drunkenness", and other models may also not include "severe drunkenness" or "no drunkenness" and so on.

[0037] The drunkenness assessment module 400 is also used to determine the drunkenness degree of the monitored object based on the multiple drunkenness degree classification results, wherein the integrated classification model votes and counts the multiple drunkenness degree classification results and outputs the drunkenness degree of the monitored object.

[0038] Specifically, the centralized classification model votes and counts the drunkenness classification results output by multiple decision tree sub-models. The drunkenness classification results may include four classification results: no drunkenness, mild drunkenness, moderate drunkenness, and severe drunkenness. The voting method may be to count the number of each classification result and determine the classification result with the largest number as the drunkenness degree of the monitored object. The drunkenness classification result may also be the probability of one or more of the drunkenness degrees among the probability of no drunkenness, mild drunkenness, moderate drunkenness, and severe drunkenness. The voting method may also be to weightedly add the probabilities of the same drunkenness degree output by multiple decision tree sub-models. For example, the drunkenness classification results output by the first decision tree sub-model are 50% mild drunkenness and 50% moderate drunkenness, and the drunkenness classification results output by the second decision tree sub-model are moderate drunkenness and severe drunkenness. Drunk 80% and severely drunk 30%, a weight value is set corresponding to the output result of each decision tree sub-model. For example, the weight value of the first decision tree sub-model is a, and the weight value of the second decision tree sub-model is b. If there are other decision tree sub-models, and so on, corresponding weight values ​​can also be set for other decision tree sub-models. Then, the probability of each drunkenness level in the drunkenness classification result output by each decision tree sub-model is multiplied by the weight value corresponding to each decision tree sub-model, and then the weighted probabilities of the same drunkenness degree in multiple drunkenness classification results are added together to obtain the final probability corresponding to each drunkenness degree, and the drunkenness degree corresponding to the largest probability is selected as the drunkenness degree of the monitored object.

[0039] It should be understood that due to Figure 3 and Figure 4 Each of them corresponds to a model, but this application includes multiple models (actually more than one hundred), and the results of each model are not exactly the same, and the model results can be one, two, three or four, so it is normal that the calculation result does not include "moderate drunkenness", and other models may also not include "severe drunkenness" or "no drunkenness" and so on.

[0040] The risk assessment module 500 is used to generate a health risk level according to the initial risk data and the degree of drunkenness.

[0041] Specifically, the initial risk data and the degree of drunkenness are combined to conduct a refined assessment of the health risk of drinking for the monitored object. The initial risk data and the degree of drunkenness can be weighted comprehensively to fully consider the different contributions of all indicator characteristics in assessing the health risk of drinking. This comprehensive process aims to integrate information from different dimensions and ultimately form a comprehensive and accurate health risk index for the monitored object after drinking, and determine the health risk level of the monitored object based on the preset health risk level. The preset health risk level can be: No risk: the health risk index value is [0, 0.25]; Low risk: The health risk index value is (0.25, 0.5]; Medium risk: The health risk index value is (0.5, 0.85]; High risk: The health risk index value is (0.85, 1].

[0042] In this embodiment, the current PPG signal waveform and physiological data of the monitored object collected by the smart wearable device are obtained by the data acquisition module 100, so that the current physical data of the monitored object can be clearly understood, which is convenient for the subsequent health risk assessment after drinking. The five-dimensional time-frequency feature values ​​of the PPG signal waveform and the physiological feature values ​​of the physiological data are respectively extracted by the feature extraction module 200, wherein the five-dimensional time-frequency feature values ​​include the main wave amplitude, the dicrotic wave amplitude, the beat time ratio, the frequency concentration value and the concentrated frequency amplitude value, which can facilitate the data comparison module 300 to compare the five-dimensional time-frequency feature values ​​and the physiological feature values ​​with the historical health data respectively, and then the five-dimensional time-frequency feature values ​​and the physiological feature values ​​are respectively compared with the historical health data. A preliminary assessment of drinking risk is performed to obtain initial risk data. The drunkenness assessment module 400 then inputs multiple five-dimensional time-frequency feature values ​​among the five time-frequency feature values ​​into multiple decision tree sub-models within the integrated classification model to obtain multiple drunkenness degree classification results, wherein the integrated classification model includes multiple decision tree sub-models, and different decision tree sub-models input the types and quantities of five-dimensional time-frequency feature values. Each decision tree sub-model performs analysis based on multiple five-dimensional time-frequency feature values, and integrates the characteristics of multiple five-dimensional time-frequency feature values ​​to obtain drunkenness degree classification results, fully considering the mutual influence of multiple five-dimensional time-frequency feature values, and comprehensively analyzing to improve analysis accuracy. , and different decision tree sub-models use different types and numbers of five-dimensional time-frequency eigenvalues ​​to split and combine the five five-dimensional time-frequency eigenvalues ​​for analysis, thereby increasing the diversity of the analysis and avoiding the singleness of the analysis. It can also avoid the problem that a certain eigenvalue abnormality directly leads to inaccurate analysis results, thereby improving the accuracy of the analysis results; the drunkenness assessment module 400 also determines the drunkenness of the monitored object according to multiple drunkenness degree classification results, wherein the integrated classification model votes and counts the multiple drunkenness degree classification results, and outputs the drunkenness of the monitored object, thereby further improving the accuracy of the analysis results; the risk assessment module 500 comprehensively considers the initial risk data and the drunkenness degree to ensure the accuracy of the results, thereby determining the health risk level of the monitored object, and the obtained health risk level can be displayed in the smart wearable device or sent to the mobile terminal, etc., so that the monitored object can view it in real time. Since the monitored object wears the smart wearable device in daily life, real-time monitoring of the monitored object is achieved, so that the monitored object can view its own health risk level in real time in daily life or during drinking, thereby avoiding the problem that the existing test method is only applicable to specific usage scenarios and has a certain lag, and at the same time, it plays a prompting role for the monitored object to avoid the danger of the monitored object's body due to drinking.

[0043] For example, smart wearable devices can be connected to professional software equipped with data service platforms to collect basic information characteristics such as the user's health level and health behavior, and build personalized health records by combining various monitoring data and risk assessment data obtained above. This comprehensive information can provide a solid data foundation for subsequent drinking health risk assessment, disease prevention, exercise guidance, and the formulation of personalized medical plans.

[0044] Exemplarily, the drinking health risk assessment system further includes a waveform processing module for determining the real-time pulse wave of the monitored object based on the PPG signal waveform.

[0045] Specifically, the original PPG signal waveform is filtered and denoised, and the clutter in the waveform and the formed waveform image are processed to form a clear real-time pulse wave, which is convenient for accurately extracting the five-dimensional time-frequency characteristic values ​​of the monitored object. For example, the PPG signal waveform is preprocessed by a Butterworth filter, and then the motion artifact correction algorithm based on the Gaussian function is used to remove the motion artifact waveform to obtain the real-time pulse wave. Among them, the Butterworth filter is a kind of electronic filter, which is also called the maximum flat filter. The characteristic of the Butterworth filter is that the frequency response curve in the passband is maximally flat, without ripples, and gradually decreases to zero in the stopband; the motion artifact correction algorithm based on the Gaussian function compensates the image by using the Gaussian function to reduce the influence of motion artifacts. Under normal circumstances, the motion artifact correction algorithm based on the Gaussian function will smooth the image, thereby blurring the artifacts caused by the movement, making the image look smoother and clearer.

[0046] The alcohol-related health risk assessment system also includes a feature evaluation module, which is used to obtain the physical indicator data of the monitored object, and use the preliminary extreme gradient boosting model to evaluate the feature importance of all the acquired physical indicator data. Through the comprehensive analysis of Gini and Permutation importance, the features that have a significant impact on the risk of drinking are screened out, namely the five-dimensional time-frequency feature values ​​and physiological feature values, to ensure the subsequent prediction application with low noise, low cost and high computation while capturing most of the useful features.

[0047] Optionally, the decision tree sub-model includes a plurality of split nodes, each of the split nodes has two branches, the two branch ends of each split node are another split node or an output classification result, and each split node corresponds to a split parameter and a five-dimensional time-frequency feature value; at each split node, the corresponding five-dimensional time-frequency feature value is compared with the corresponding split parameter to determine the branch direction of each split node; The number of the five-dimensional time-frequency feature values ​​input into each decision tree sub-model is less than or equal to the number of the splitting nodes, and each of the input five-dimensional time-frequency feature values ​​appears at least once in a plurality of the splitting nodes.

[0048] Specifically, Figure 3 As shown, 12 split nodes are set in the first decision tree sub-model, each split node uses a pulse feature value, and each split node is set with a split parameter, for example, Figure 3 The first split node at the top uses feature 4, and the split parameter is set to 91.5. At the first split node, the value of feature 4 input is compared with 91.5. When the value of feature 4 input is greater than 91.5, the branch direction after the first split node Figure 3 At the location of feature 2 on the left, when the value of feature 4 is less than or equal to 91.5, the branch direction after the first split node Figure 3 The location of feature 3 on the right; and so on, at each split node, the pulse feature value corresponding to each split node is compared with the corresponding split parameter. Sometimes, the drunkenness classification result can be obtained after two split nodes, for example Figure 3 In the example, after passing through the nodes of feature 4 and feature 2, a “heavy” classification result is obtained; sometimes, it is necessary to pass through three or more split nodes for continuous analysis and judgment to obtain a classification result, for example Figure 3 On the right side, after continuous judgment of feature 4, feature 3, and the splitting node where feature 2 is located, the classification result of "mild" or "none" is obtained.

[0049] It should be understood that due to Figure 3 and Figure 4 Each of them corresponds to a model, but this application includes multiple models (actually more than one hundred), and the results of each model are not exactly the same, and the model results can be one, two, three or four, so it is normal that the calculation result does not include "moderate drunkenness", and other models may also not include "severe drunkenness" or "no drunkenness" and so on.

[0050] In addition, from Figure 3 and Figure 4 The structures of the two decision tree sub-models are shown. It can be seen that different decision tree sub-models use different types and numbers of features. The advantage of this is that if one of the features is abnormal, for example, feature 5 is abnormal, but for Figure 3 The first decision tree sub-model in the algorithm has no effect, so the classification result obtained by the first decision tree sub-model is still correct. When voting and counting the classification results of multiple decision tree sub-models, the influence of abnormal classification results can be reduced, and the accuracy and precision of the final determined degree of drunkenness can be higher.

[0051] Optionally, the feature extraction module 200 includes: A time domain analysis unit 210, configured to perform time domain analysis on the PPG signal waveform to determine the main wave amplitude, the dicrotic wave amplitude and the beat time ratio; The frequency domain analysis unit 220 is used to perform frequency domain analysis on the PPG signal waveform to determine the frequency concentration value and the concentrated frequency amplitude value.

[0052] Specifically, time domain analysis refers to the feature extraction of pulse waves in time series. Through time domain analysis, the characteristics of pulse wave amplitude, period, rise time, fall time, main wave amplitude, dicrotic wave amplitude and beat time ratio can be obtained. These parameters can reflect the physiological state of the heart's contraction and relaxation function, blood vessel elasticity and resistance. Frequency domain analysis refers to the feature extraction of pulse waves in the frequency domain. Through frequency domain analysis, the calculation of pulse wave spectrum, extraction of frequency components, frequency distribution of pulse waves, main frequency concentration value and concentrated frequency amplitude value can be obtained. These frequency distributions can reflect the physiological state of the heart and blood vessels. After drinking, under the influence of alcohol, the heart beats and the strength of the pulse will be affected. Therefore, the characteristic value of the pulse wave can be extracted first, and the user's drunkenness degree can be analyzed based on the characteristic value of the pulse wave.

[0053] Optionally, the alcohol-related health risk assessment system further includes a modeling module, which includes: A training set construction unit is used to obtain the PPG signal waveforms of multiple monitored objects and determine the five-dimensional time-frequency feature values ​​of each monitored object, wherein a group of five-dimensional time-frequency feature values ​​of each monitored object and the actual drunkenness degree of the monitored object corresponding to each group of five-dimensional time-frequency feature values ​​are taken as a sample data, and multiple sample data constitute a training set.

[0054] Specifically, the PPG signal waveforms of the plurality of monitored objects are acquired, and the real-time pulse waves of the monitored objects are determined; and the five-dimensional time-frequency characteristic values ​​of each monitored object are determined according to the real-time pulse wave of each monitored object.

[0055] The training set construction unit is also used to extract the sample data from the training set with replacement to construct multiple training subsets respectively, wherein the multiple training subsets correspond one-to-one to the multiple decision tree sub-models, and each training subset is used to train the corresponding decision tree sub-model.

[0056] Specifically, sampling with replacement is a method of random sampling, also called repeated sampling (replacementsampling). Sampling with replacement is to extract a sample each time, put it back into the sample population after extraction, and then extract it again. This process is repeated to reuse the sample, that is, to obtain different training subsets through the same sample. The purpose is to obtain multiple training subsets through the repeated use of samples, so as to obtain more accurate model results by combining multiple training subsets in the future.

[0057] A model training unit is used to train the corresponding decision tree sub-model using the sample data in the training subset until the output result of the decision tree sub-model is consistent with the actual drunkenness degree in the sample data, determine the splitting parameter of the decision tree sub-model, or determine the splitting node position of each of the five-dimensional time-frequency feature values ​​in the decision tree sub-model.

[0058] Specifically, a group of volunteers were selected, and different drinking test plans were formulated for different volunteers based on the relationship between blood alcohol concentration and the physical condition of the volunteers. In the actual test process, the actual drinking environment was simulated, different volunteers were controlled to drink different amounts of alcohol, and the blood alcohol concentration of the volunteers was monitored in real time to calculate the actual blood alcohol concentration of the volunteers. Exemplarily, the actual blood alcohol concentration of the volunteers was calculated using the actual alcohol concentration formula, and the actual degree of drunkenness was determined based on the actual blood alcohol concentration using the drunkenness standard corresponding to driving with alcohol or drunk driving. For example, the drunkenness standard corresponding to driving with alcohol or drunk driving is: when the actual blood alcohol concentration is between 0mg / 100mL and 20mg / 100mL, the actual degree of drunkenness is "none"; when the actual blood alcohol concentration is between 20mg / 100mL and 50mg / 100mL, the actual degree of drunkenness is "mild"; when the actual blood alcohol concentration is between 20mg / 100mL and 50mg / 100mL, the actual degree of drunkenness is "mild"; When the alcohol concentration is between 50mg / 100mL and 80mg / 100mL, the actual drunkenness level is "moderate"; when the actual blood alcohol concentration exceeds 80mg / 100mL, the actual drunkenness level is "severe", where 0mg / 100mL means 0mg of blood alcohol per 100mL of blood, 20mg / 100mL means 20mg of blood alcohol per 100mL of blood, and so on. The numerical range in the drunkenness standard of this embodiment is only an example and can be adjusted according to actual conditions. The actual alcohol concentration formula includes: BAC = x / (m*z)-d*t; Among them, BAC is the actual blood alcohol concentration, x is the monitored blood alcohol concentration, m is the weight of the volunteer, z is the body fluid coefficient, d is the metabolic rate, and t is time.

[0059] The degree of drunkenness is divided into "none", "mild", "moderate" and "severe", and the PPG data of the corresponding volunteers are collected. After extracting the above-mentioned five-dimensional time-frequency features, multiple eigenvalues ​​of the five eigenvalues ​​are randomly combined and then input into different decision tree sub-models for training. The decision tree sub-model performs machine learning to obtain the corresponding decision tree sub-model. Finally, the user's data is evaluated and analyzed in the form of a decision tree for health risk assessment.

[0060] Optionally, the model training unit uses the sample data in the training subset to train the corresponding decision tree sub-model, including: Extract the five-dimensional time-frequency feature values ​​from the sample data in each of the training subsets, wherein the types and quantities of the five-dimensional time-frequency feature values ​​extracted from the same training subset remain consistent, and the types and quantities of the five-dimensional time-frequency feature values ​​extracted from different training subsets are different, for example Figure 3 and Figure 4 The decision tree sub-models shown use different numbers and types of feature values, and some feature values ​​can be reused.

[0061] The extracted five-dimensional time-frequency feature values ​​are input into the corresponding decision tree sub-model, and the output result of the decision tree sub-model is compared with the corresponding actual drunkenness degree, and the splitting parameter of the decision tree sub-model is adjusted, or further, the splitting node position of each pulse feature value in the decision tree sub-model can be adjusted.

[0062] In this embodiment, the number and type of feature values ​​used by each decision tree sub-model are different, so the same training subset can be used by multiple decision tree sub-models. For example, for the same training subset, one decision tree sub-model uses feature 1, feature 3 and feature 5 in each sample data in the training subset, while another decision tree sub-model uses the same training subset but uses feature 1, feature 2, feature 4 and feature 5 in each sample data. Since each decision tree sub-model uses a different type or number of eigenvalues, the reuse rate of the training subset can be improved, the number of sampled data can be reduced, and the training difficulty can be reduced. In addition, not all five eigenvalues ​​are input into each decision tree sub-model. On the one hand, the complexity of the decision tree sub-model is reduced and the analysis speed is improved. On the other hand, a variety of decision tree sub-models can be trained, so that multiple decision tree sub-models with good evaluation and analysis effects can be selected and used, and the classification results of multiple decision tree sub-models can be integrated to improve the accuracy of the classification results. If a single decision tree model is used, the analysis result output by it is unique. Once a certain eigenvalue is abnormal, it will have a greater impact on the analysis result and reduce the accuracy of the analysis result. Multiple decision tree sub-models use different eigenvalues ​​to further reduce the error rate and improve the accuracy of the classification results.

[0063] Optionally, before extracting the sample data from the training set with replacement to construct a plurality of training subsets respectively, the training set construction unit is further used to expand the sample data in the training set to obtain virtual sample data so as to balance the number of sample data corresponding to the various drunkenness degrees.

[0064] Specifically, since there may be a large gap in the number of samples collected from volunteers with different levels of drunkenness, in order to better train the model, the synthetic minority over-sampling technique (SMOTE) is used to balance the sample quantity of the obtained sample data. If the number of samples corresponding to certain levels of drunkenness is small, the oversampling algorithm can be used to obtain virtual samples, thereby increasing the number of samples corresponding to the level of drunkenness, so that the number of samples corresponding to multiple levels of drunkenness is uniform, avoiding the use of unbalanced training sets to train the decision tree sub-model.

[0065] In order to verify the accuracy of the model, we collected drinking experiment data and combined it with the upsampling SMOTE method to select 80 sample data, including 22 cases of no drunkenness, 22 cases of mild drunkenness, 14 cases of moderate drunkenness and 22 cases of severe drunkenness. After the above preprocessing, time-frequency five-dimensional feature extraction and model training, the accuracy of the comparison between the output of the integrated classification model and the actual degree of drunkenness can be obtained. The following randomly selected 24 data examples are shown, as shown in the figure below. Figure 6 As shown in the table.

[0066] After verifying all 66 sample data, the final cross-validation accuracy average was 80%, proving that the above system is meaningful for assessing health risks after drinking.

[0067] In one embodiment, when building a decision tree, each internal node needs to decide how to split the data set into subsets based on the input features. By calculating the Gini coefficient for each possible split, the split that results in the lowest Gini coefficient (i.e., the highest purity) can be selected, thereby ensuring that the resulting decision tree can better classify the data. The Gini coefficient refers to the probability that when two elements are randomly selected from n elements, the two selected elements belong to different groups. The misclassification probability of the target variable with the most derivatives in each section of the decision tree model can be calculated using the Gini formula: ; Among them, Gini Index (p) is the Gini coefficient, n is the total number of categories, and p is i is the proportion of samples belonging to category i.

[0068] In the model of this embodiment, the separation and merging standard value of the decision rule can be set to 0.05, the number of parent nodes can be limited to 200, the number of child nodes can be limited to 100, and the number of branches can be limited to 5.

[0069] Optionally, the data comparison module 300 includes: The data processing unit 310 is used to generate a historical time-frequency mean and a historical physiological mean according to the historical health data; The deviation comparison unit 320 is used to compare the five-dimensional time-frequency characteristic value with the historical time-frequency mean value, and the physiological characteristic value with the historical physiological mean value, to generate corresponding deviation data; The anomaly detection unit 330 is used to perform anomaly detection on the deviation data using an isolation forest algorithm to generate the initial risk data.

[0070] Specifically, the data processing unit 310 generates a historical time-frequency mean and a historical physiological mean according to the historical health data. The historical time-frequency mean and the historical physiological mean can provide feedback on the normal health status of the monitored object. The deviation comparison unit 320 compares and analyzes the real-time collected data with the historical physiological mean to generate corresponding deviation data, that is, the real-time collected five-dimensional time-frequency feature value and the historical time-frequency mean are subtracted to obtain the corresponding deviation data, and the real-time collected physiological feature value and the historical physiological mean are subtracted to obtain the corresponding deviation data for the abnormal detection of the abnormal detection unit 330. The physiological characteristic values ​​include body mass index (BMI), body temperature and deep sleep duration, so the corresponding deviation data include five time-frequency characteristic differences corresponding to the five-dimensional time-frequency characteristic values ​​and three physiological characteristic differences corresponding to the physiological characteristic values. The abnormality detection unit 330 uses the isolation forest algorithm to detect abnormalities in the deviation data. First, the features and the split points are randomly selected. For each sample, a feature and the corresponding deviation data are randomly selected, and then the split point of the feature is randomly selected, that is, the split point of the corresponding deviation data is selected. The sample refers to the current PPG signal waveform and physiological data of the monitored object collected by the smart wearable device, or "for each sample" can also be understood as "for the current PPG signal waveform and physiological data of each monitored object", and the features refer to the main wave amplitude, the dicrotic wave amplitude, the beat time ratio, the frequency concentration value, the concentrated frequency amplitude value, the body mass index (BMI), the body temperature and the deep sleep duration. Then, a tree is constructed, and multiple random trees are constructed by recursive splitting until each observation point is isolated. Among them, the path where the outlier is isolated in the tree is shorter. Finally, the anomaly score is calculated. Through the results of multiple trees, the anomaly formula is used to calculate the anomaly score of each sample. According to the actual needs and the characteristics of the data set, a negative value close to 0 is set as the threshold. When the anomaly score of the sample is lower than the threshold, the sample is considered normal; when the anomaly score is higher than the threshold, the sample is considered abnormal. The anomaly formula includes: Anomaly Score(X)=2-E(X) / c(n); Among them, Anomaly Score(X) is the anomaly score of sample X, E(X) is the expected depth of sample X isolated in the forest, and c(n) is the average depth of the tree.

[0071] Therefore, after obtaining the abnormal score through the abnormal formula, the abnormal numerical probability is further calculated according to the abnormal score, and the initial risk data can be determined based on the division range of the initial risk data, where the division range of the initial risk data includes: No initial risk means that the abnormality scores of the abnormal values ​​are all lower than the preset abnormality threshold, and the probability of the abnormal value is lower than the first preset abnormal probability. The preset abnormal threshold can be set to 0.05, and the first preset abnormal probability can be set to 0.1; initial low risk means that the number of abnormal values ​​with abnormal scores higher than the preset abnormal threshold is 1, and the probability of the abnormal value is higher than the second preset abnormal probability. The second preset abnormal probability can be set to 0.3; initial medium risk means that the number of abnormal values ​​with abnormal scores higher than the preset abnormal threshold is 2, and the probability of the abnormal value is higher than the third preset abnormal probability. The third preset abnormal probability can be set to 0.6; initial high risk means that the number of abnormal values ​​with abnormal scores higher than the preset abnormal threshold is 3, and the probability of the abnormal value is higher than the fourth preset abnormal probability. The fourth preset abnormal probability can be set to 0.9.

[0072] Exemplarily, the key parameter settings of the Isolation Forest algorithm can be n_estimators=100, outliers_fraction = 0.05, max_features=1.0, where n_estimators is the number of trees in the forest, outliers_fraction is the proportion of expected outliers in the specified data set, and max_features is the maximum feature ratio considered at each split.

[0073] Optionally, the risk assessment module generates a health risk level according to the initial risk data and the degree of drunkenness, including: The initial risk data and the drunkenness degree are normalized to generate a health risk index, and the health risk level is determined based on the health risk index.

[0074] Specifically, the initial risk data is determined based on the abnormal score and the abnormal numerical probability further calculated based on the abnormal score. The drunkenness degree is determined based on the maximum probability of the probabilities corresponding to each drunkenness degree. Therefore, the health risk index can be determined based on the abnormal numerical probability of the initial risk data and the maximum probability of the drunkenness degree, and then the health risk level can be determined based on the division range of the health risk index and the health risk level. First, the abnormal numerical probability of the initial risk data and the maximum drunkenness probability of the drunkenness degree are normalized to standardize the abnormal numerical probability and the maximum drunkenness probability. For the abnormal numerical probability, a linear transformation is performed according to its maximum and minimum values ​​so that all values ​​are mapped to the [0, 1] interval, where 0 represents no deviation (i.e., fully meets the preset standard) and 1 represents the maximum deviation. For the maximum drunkenness probability, although it may be close to or in the [0, 1] interval, in order to ensure consistency, it still needs to be subjected to necessary standardization to ensure that it fully meets the requirements of subsequent calculations. Then, the normalized abnormal value probability and the maximum drunkenness probability are comprehensively weighted by a weighted formula to fully consider the different contributions of all indicator characteristics in assessing the health risk of drinking, obtain the health risk index, and determine the health risk level of the test object according to the preset health risk level. The weighted formula includes: R=0.416P(ASi)+0.584Pi; Among them, R is the health risk index, P(ASi) is the probability of abnormal values ​​corresponding to the initial risk data, and Pi is the maximum drunkenness probability of the drunkenness degree. 0.416 is the fusion coefficient of P(ASi), and 0.584 is the fusion coefficient of Pi. For example, a dual verification mechanism based on feature selection and permutation importance based on Gini impurity is used to quantitatively fuse the significant contribution of PPG signal waveform and physiological data to obtain the corresponding fusion coefficient. The steps include: The importance scores of PPG signal waveform and physiological data are calculated and quantified by random forest classifier to capture the information gain efficiency of PPG signal waveform and physiological data in the decision tree splitting process and the sensitivity contribution to the decline in prediction accuracy; The importance scores of the PPG signal waveform and physiological data are mapped to a unified dimension to obtain the importance scores of the PPG signal waveform and physiological data in the same dimension, and then the fusion weight formula group is allocated based on the contribution ratio to determine the fusion coefficient. The importance scores of the PPG signal waveform and physiological data can be obtained according to actual tests. For example, the importance scores of the PPG signal waveform are 2.9597, 2.604, 2.3087, 2.1611 and 1.906, and the total importance score of the PPG signal waveform is 11.9395. The importance scores of the physiological data include 2.5973, 2.0738, 1.9396 and 1.8993, and the total importance score of the physiological data is 8.51.

[0075] The contribution ratio distribution fusion weight formula group includes: ; in, α is the fusion coefficient of Pi, β is the fusion coefficient of P(ASi), is the overall importance score of the PPG signal waveform, Score the overall importance of the physiological data.

[0076] In an optional embodiment of the present invention, there is also provided an electronic device, including a memory and a processor; The memory is used to store computer programs; The processor is configured to implement the following steps when executing the computer program: Obtain the current PPG signal waveform and physiological data of the monitored object collected by the smart wearable device; Respectively extracting five-dimensional time-frequency characteristic values ​​of the PPG signal waveform and physiological characteristic values ​​of the physiological data, wherein the five-dimensional time-frequency characteristic values ​​include main wave amplitude, dicrotic wave amplitude, beat time ratio, frequency concentration value and concentrated frequency amplitude value; Comparing the five-dimensional time-frequency feature value and the physiological feature value with historical health data respectively to generate initial risk data, wherein the historical health data refers to historical data corresponding to the current five-dimensional time-frequency feature value and the physiological feature value type; Inputting the five-dimensional time-frequency feature values ​​into a plurality of decision tree sub-models in an integrated classification model to obtain a plurality of drunkenness degree classification results, wherein the integrated classification model includes a plurality of decision tree sub-models, and different decision tree sub-models input the five-dimensional time-frequency feature values ​​in different types and quantities; further determining the drunkenness degree of the monitored object based on the plurality of drunkenness degree classification results, wherein the integrated classification model performs voting statistics on the plurality of drunkenness degree classification results and outputs the drunkenness degree of the monitored object; A health risk level is generated based on the initial risk data and the degree of intoxication.

[0077] In this embodiment, the electronic device may be a processing device, a computer, a server, etc. Figure 7 As shown, the electronic device can be wirelessly connected with the smart wearable device to achieve data transmission and interaction. The smart wearable device can be a head-mounted wearable device or a wristwatch, etc., which is used to collect the physiological data of the monitored object and transmit the collected physiological data to the electronic device through a wireless network. The electronic device can also feedback the execution result of the computer program to the smart wearable device to remind the wearer to pay attention to his own health risk status, so that the user can drink in moderation and avoid health and safety problems caused by drinking. In addition, the electronic device can also be connected to a mobile terminal, which can be a mobile phone, tablet computer or other device of the monitored object or the manager of the monitored object. The smart wearable device and the mobile terminal can also be connected through the network. The mobile terminal is connected through a wireless network to receive the processing results of the electronic device, that is, to receive the five-dimensional parameters obtained by the electronic device after processing the physiological data and the assessed health risk level. After receiving the feedback data, the corresponding prompt operation is performed so that the monitored object or its manager can grasp the physical condition in real time and make effective adjustments to ensure personal safety.

[0078] In an optional embodiment of the present invention, a computer-readable storage medium is further provided, wherein a computer program is stored on the storage medium. When the computer program is executed by a processor, the following steps are implemented: Obtain the current PPG signal waveform and physiological data of the monitored object collected by the smart wearable device; Respectively extracting five-dimensional time-frequency characteristic values ​​of the PPG signal waveform and physiological characteristic values ​​of the physiological data, wherein the five-dimensional time-frequency characteristic values ​​include main wave amplitude, dicrotic wave amplitude, beat time ratio, frequency concentration value and concentrated frequency amplitude value; Comparing the five-dimensional time-frequency feature value and the physiological feature value with historical health data respectively to generate initial risk data, wherein the historical health data refers to historical data corresponding to the current five-dimensional time-frequency feature value and the physiological feature value type; Inputting the five-dimensional time-frequency feature values ​​into a plurality of decision tree sub-models in an integrated classification model to obtain a plurality of drunkenness degree classification results, wherein the integrated classification model includes a plurality of decision tree sub-models, and different decision tree sub-models input the five-dimensional time-frequency feature values ​​in different types and quantities; further determining the drunkenness degree of the monitored object based on the plurality of drunkenness degree classification results, wherein the integrated classification model performs voting statistics on the plurality of drunkenness degree classification results and outputs the drunkenness degree of the monitored object; A health risk level is generated based on the initial risk data and the degree of intoxication.

[0079] An electronic device that can be used as a server or client of the present invention will now be described, which is an example of a hardware device that can be applied to various aspects of the present invention. Electronic devices are intended to represent various forms of digital electronic computer devices, such as laptop computers, desktop computers, workbenches, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. Electronic devices can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples, and are not intended to limit the implementation of the present invention described and / or required herein.

[0080] The electronic device includes a computing unit, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) or a computer program loaded from a storage unit into a random access memory (RAM). In the RAM, various programs and data required for the operation of the device can also be stored. The computing unit, ROM, and RAM are connected to each other through a bus. An input / output (I / O) interface is also connected to the bus.

[0081] A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment method can be implemented by instructing the relevant hardware through a computer program, and the program can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, the storage medium can be a disk, an optical disk, a read-only memory (ROM) or a random access memory (RAM), etc. In the present application, the unit described as a separate component may or may not be physically separated, and the component displayed as a unit may or may not be a physical unit, that is, it may be located in one place, or it may be distributed on multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the embodiment of the present invention. In addition, each functional unit in each embodiment of the present invention can be integrated in a processing unit, or each unit can exist physically separately, or two or more units can be integrated in one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of a software functional unit.

[0082] Although the present invention is disclosed as above, the protection scope of the present invention is not limited thereto. Those skilled in the art may make various changes and modifications without departing from the spirit and scope of the present invention, and these changes and modifications will fall within the protection scope of the present invention.

Claims

1. A drinking health risk assessment system, characterized in that: include: A data acquisition module is used to obtain the current PPG signal waveform and physiological data of the monitored object collected by the smart wearable device; A feature extraction module, used to extract five-dimensional time-frequency feature values ​​of the PPG signal waveform and physiological feature values ​​of the physiological data, wherein the five-dimensional time-frequency feature values ​​include main wave amplitude, dicrotic wave amplitude, beat time ratio, frequency concentration value and concentrated frequency amplitude value; A data comparison module, used to compare the five-dimensional time-frequency characteristic value and the physiological characteristic value with historical health data to generate initial risk data, wherein the historical health data refers to historical data corresponding to the current five-dimensional time-frequency characteristic value and the physiological characteristic value type; a drunkenness assessment module, configured to input the five-dimensional time-frequency feature values ​​into a plurality of decision tree sub-models in an integrated classification model to obtain a plurality of drunkenness degree classification results, wherein the integrated classification model includes a plurality of decision tree sub-models, and different decision tree sub-models input the five-dimensional time-frequency feature values ​​in different types and quantities; and further determine the drunkenness degree of the monitored object according to the plurality of drunkenness degree classification results, wherein the integrated classification model performs voting statistics on the plurality of drunkenness degree classification results and outputs the drunkenness degree of the monitored object; The risk assessment module is used to generate a health risk level according to the initial risk data and the degree of drunkenness.

2. The alcohol-related health risk assessment system according to claim 1, characterized in that: The decision tree sub-model includes a plurality of split nodes, each of which has two branches, and the two branch ends of each split node are another split node or an output classification result, and each split node corresponds to a split parameter and a five-dimensional time-frequency feature value; at each split node, the corresponding five-dimensional time-frequency feature value is compared with the corresponding split parameter to determine the branch direction of each split node; The number of the five-dimensional time-frequency feature values ​​input into each decision tree sub-model is less than or equal to the number of the splitting nodes, and each of the input five-dimensional time-frequency feature values ​​appears at least once in a plurality of the splitting nodes.

3. The alcohol-related health risk assessment system according to claim 1, characterized in that: The feature extraction module comprises: A time domain analysis unit, configured to perform time domain analysis on the PPG signal waveform to determine the main wave amplitude, the dicrotic wave amplitude and the beat time ratio; A frequency domain analysis unit is used to perform frequency domain analysis on the PPG signal waveform to determine the frequency concentration value and the concentrated frequency amplitude value.

4. The alcohol-related health risk assessment system according to claim 2, characterized in that: Also includes: A modeling module, the modeling module comprising: a training set construction unit, configured to obtain the PPG signal waveforms of a plurality of the monitored subjects, and determine the five-dimensional time-frequency feature values ​​of each of the monitored subjects, wherein a group of the five-dimensional time-frequency feature values ​​of each of the monitored subjects and the actual drunkenness degree of the monitored subjects corresponding to each group of the five-dimensional time-frequency feature values ​​are taken as a sample data, and a plurality of the sample data constitute a training set; and further configured to extract the sample data from the training set with replacement, and respectively construct a plurality of training subsets, wherein the plurality of the training subsets correspond one-to-one to the plurality of the decision tree sub-models, and each of the training subsets is used to train the corresponding decision tree sub-model; A model training unit is used to train the corresponding decision tree sub-model using the sample data in the training subset until the output result of the decision tree sub-model is consistent with the actual drunkenness degree in the sample data, and determine the splitting parameter of the decision tree sub-model.

5. The drinking health risk assessment system according to claim 4, characterized in that: The model training unit uses the sample data in the training subset to train the corresponding decision tree sub-model, including: Extracting the five-dimensional time-frequency feature values ​​from the sample data in each of the training subsets, wherein the types and quantities of the five-dimensional time-frequency feature values ​​extracted from the same training subset remain consistent, and the types and quantities of the five-dimensional time-frequency feature values ​​extracted from different training subsets are different; The extracted five-dimensional time-frequency feature values ​​are input into the corresponding decision tree sub-model, and the output result of the decision tree sub-model is compared with the corresponding actual drunkenness degree, and the splitting parameter of the decision tree sub-model is adjusted.

6. The alcohol-related health risk assessment system according to claim 4, characterized in that: Before extracting the sample data from the training set with replacement to construct a plurality of training subsets respectively, the training set construction unit is further used to expand the sample data in the training set to obtain virtual sample data so as to balance the number of sample data corresponding to the various drunkenness degrees.

7. The drinking health risk assessment system according to claim 1, characterized in that: The data comparison module includes: A data processing unit, used to generate a historical time-frequency mean and a historical physiological mean according to the historical health data; A deviation comparison unit, used to compare the five-dimensional time-frequency characteristic value with the historical time-frequency mean, and the physiological characteristic value with the historical physiological mean, to generate corresponding deviation data; The anomaly detection unit is used to perform anomaly detection on the deviation data using an isolation forest algorithm to generate the initial risk data.

8. The alcohol-related health risk assessment system according to claim 1, characterized in that: The risk assessment module generates a health risk level according to the initial risk data and the degree of drunkenness, including: The initial risk data and the drunkenness degree are normalized to generate a health risk index, and the health risk level is determined based on the health risk index.

9. An electronic device, characterized in that: including memory and processor; The memory is used to store computer programs; The processor is configured to implement the following steps when executing the computer program: Obtain the current PPG signal waveform and physiological data of the monitored object collected by the smart wearable device; Respectively extracting five-dimensional time-frequency characteristic values ​​of the PPG signal waveform and physiological characteristic values ​​of the physiological data, wherein the five-dimensional time-frequency characteristic values ​​include main wave amplitude, dicrotic wave amplitude, beat time ratio, frequency concentration value and concentrated frequency amplitude value; Comparing the five-dimensional time-frequency feature value and the physiological feature value with historical health data respectively to generate initial risk data, wherein the historical health data refers to historical data corresponding to the current five-dimensional time-frequency feature value and the physiological feature value type; Inputting the five-dimensional time-frequency feature values ​​into a plurality of decision tree sub-models in an integrated classification model to obtain a plurality of drunkenness degree classification results, wherein the integrated classification model includes a plurality of decision tree sub-models, and different decision tree sub-models input the five-dimensional time-frequency feature values ​​in different types and quantities; further determining the drunkenness degree of the monitored object based on the plurality of drunkenness degree classification results, wherein the integrated classification model performs voting statistics on the plurality of drunkenness degree classification results and outputs the drunkenness degree of the monitored object; A health risk level is generated based on the initial risk data and the degree of intoxication.

10. A computer-readable storage medium, characterized in that: The storage medium stores a computer program. When the computer program is executed by the processor, the following steps are implemented: Obtain the current PPG signal waveform and physiological data of the monitored object collected by the smart wearable device; Respectively extracting five-dimensional time-frequency characteristic values ​​of the PPG signal waveform and physiological characteristic values ​​of the physiological data, wherein the five-dimensional time-frequency characteristic values ​​include main wave amplitude, dicrotic wave amplitude, beat time ratio, frequency concentration value and concentrated frequency amplitude value; Comparing the five-dimensional time-frequency feature value and the physiological feature value with historical health data respectively to generate initial risk data, wherein the historical health data refers to historical data corresponding to the current five-dimensional time-frequency feature value and the physiological feature value type; Inputting the five-dimensional time-frequency feature values ​​into a plurality of decision tree sub-models in an integrated classification model to obtain a plurality of drunkenness degree classification results, wherein the integrated classification model includes a plurality of decision tree sub-models, and different decision tree sub-models input the five-dimensional time-frequency feature values ​​in different types and quantities; further determining the drunkenness degree of the monitored object based on the plurality of drunkenness degree classification results, wherein the integrated classification model performs voting statistics on the plurality of drunkenness degree classification results and outputs the drunkenness degree of the monitored object; A health risk level is generated based on the initial risk data and the degree of intoxication.

Images