A health monitoring method, device, apparatus and storage medium

Abstract

The embodiment of the application provides a kind of health monitoring method, device, equipment and storage medium, it is related to artificial intelligence field, the method comprises: obtaining the current life index value of the object to be monitored, the current life index value is calibrated by the data calibration model associated with the object to be monitored, obtains life index calibration value, life index calibration value and the current state data of the object to be monitored are input into the baseline model associated with the object to be monitored, the target reference range matched with current state data is determined by baseline model, and based on the association between life index calibration value and target reference range, the health monitoring result of the object to be monitored is obtained. The current life index value of the object to be monitored is calibrated by data calibration model, and the association between life index calibration value and target reference range is judged by baseline model to determine health status, which effectively improves the accuracy of health monitoring result.

Description

Technical Field

[0001] The embodiments of the present invention relate to the field of artificial intelligence, and in particular to a health monitoring method, device, equipment and storage medium. Background Technology

[0002] Bus drivers are in critical operational positions, and their health needs to be closely monitored, because safety accidents caused by health problems can affect the lives of all passengers and have a negative social impact.

[0003] In existing technologies, bus drivers' vital signs data are mainly collected through smart wearable devices. The collected vital signs data is then used to monitor the drivers' health. When an abnormality in the driver's health is detected, an alarm is triggered to prevent accidents from occurring.

[0004] However, the accuracy of vital sign data collected by smart wearable devices is not high, resulting in a low accuracy rate for health monitoring. Summary of the Invention

[0005] This application provides a health monitoring method, apparatus, device, and storage medium to improve the accuracy of vital sign data collected by smart wearable devices, thereby enhancing the accuracy of health monitoring.

[0006] On one hand, embodiments of this application provide a health monitoring method, including: Obtain the current vital signs values ​​of the object to be monitored; The current vital signs value is calibrated by using a data calibration model associated with the object to be monitored, and a vital signs calibration value is obtained. The calibration values ​​of the vital signs and the current status data of the object to be monitored are input into the baseline model associated with the object to be monitored. The baseline model is used to determine the target benchmark range that matches the current state data, and the health monitoring results of the subject to be monitored are obtained based on the correlation between the vital signs calibration values ​​and the target benchmark range.

[0007] In this embodiment, after obtaining the current vital signs (VLS) values ​​of the monitored object, data calibration is performed on these VLS values ​​to obtain more accurate VLS calibration values. Therefore, determining the health monitoring results of the monitored object based on the VLS calibration values ​​can effectively improve the accuracy of health monitoring. Secondly, for the current state data of the monitored object, a corresponding target baseline range is obtained. Then, a health status judgment is made based on the correlation between the VLS calibration values ​​and the target baseline range, effectively improving the accuracy of the health monitoring results.

[0008] Optionally, the baseline model includes the benchmark range of indicators corresponding to the object to be monitored under multiple states; The step of determining the target benchmark range that matches the current state data through the baseline model includes: Using the baseline model, based on the current state data of the object to be monitored, a matching target benchmark range is determined from the corresponding indicator benchmark ranges under the various states.

[0009] In this embodiment of the application, different indicator benchmark ranges are used to determine the health status for different states of the object to be monitored, instead of using the benchmark range for the same state. This avoids errors in health monitoring results due to different states, thereby improving the accuracy of health monitoring.

[0010] Optionally, determining the health monitoring results of the subject based on the correlation between the vital sign calibration values ​​and the target reference range includes: If the vital signs calibration value is within the target reference range, then the health monitoring result is determined to be normal; If the calibrated value of the vital signs is outside the target reference range, the health monitoring result is determined to be abnormal.

[0011] Optionally, it also includes: When the health monitoring result of the object to be monitored is abnormal, an alarm is triggered, and health reminder information is obtained from the knowledge base based on the alarm type; The monitoring alert information is pushed to the object to be monitored.

[0012] In this embodiment, an alarm message will be sent to the monitored object whose health monitoring result is abnormal. Health reminder information will also be pushed to the monitored object according to the type of alarm, which improves the intelligence of the entire health monitoring and thus greatly enhances the user experience.

[0013] Optionally, the data calibration model associated with the object to be monitored is trained in the following manner: Obtain historical vital signs data of the object to be monitored, and select a training sample set from the historical vital signs data; Based on the first loss function and the training sample set, a data calibration model associated with the object to be monitored is trained. The first loss function is associated with the predicted vital sign calibration data, the actual vital sign calibration data, and the fault point limitation parameter. The fault point limitation parameter is used to limit the influence of fault points on the first loss function. The fault point represents a training sample where the residual between the predicted vital sign calibration data and the actual vital sign calibration data is greater than a preset threshold.

[0014] In this embodiment, the reliability of subsequent model learning is improved by screening historical vital sign data of the monitored objects. By optimizing the loss function, parameters that constrain the loss function are added to the traditional linear model, thereby reducing the proportion of fault points in the loss function and improving the accuracy of the data calibration model.

[0015] Optionally, the historical vital signs data includes on-the-job vital signs data collected through smart wearable devices and pre-employment vital signs data collected through medical instruments. The step of selecting a training sample set from the historical vital signs data includes: The on-the-job vital signs data are downsampled to obtain first vital signs data, and the pre-job vital signs data are upsampled to obtain second vital signs data. The validity and data volume of the first vital signs data and the second vital signs data are verified, and a training sample set is obtained based on the verified vital signs data.

[0016] In this embodiment, the on-the-job vital signs data are downsampled to prevent excessively large on-the-job vital signs data from causing low training model efficiency; the pre-employment vital signs data are upsampled to prevent excessively small pre-employment vital signs data from causing low training model accuracy. After the vital signs data are collected, validity and data volume verification are performed, providing more accurate training data for model training and thus improving model accuracy.

[0017] Optionally, the validity verification and data volume verification of the first vital sign data and the second vital sign data include: The validity of the first vital signs data and the second vital signs data is verified based on the variance of the first vital signs data and the variance of the second vital signs data. Based on the data volume of the first vital signs data and the data volume of the second vital signs data, a data volume verification is performed on the first vital signs data and the second vital signs data.

[0018] In this embodiment of the application, by performing variance calculation on the vital signs data and verifying the size of the data, it is possible to determine whether the collected first vital signs data and second vital signs data are valid, and whether the size of the data meets the training requirements, thereby improving the quality of the training data and thus improving the accuracy of the model.

[0019] Optionally, training the data calibration model associated with the object to be monitored based on the first loss function and the training sample set includes: Multiple sets of training samples are randomly selected from the training sample set; For multiple sets of training samples, the following steps are performed respectively: a candidate calibration model is trained based on the first loss function and a set of training samples, and the number of fault points in the set of training samples is determined based on the candidate calibration model; The candidate calibration model with the fewest corresponding fault points among the multiple candidate calibration models obtained is taken as the data calibration model.

[0020] In this embodiment, multiple training samples are randomly selected for repeated training, reducing the likelihood of excessive model error due to special circumstances. After model training is complete, the model with the fewest fault points is selected as the candidate model, further increasing the reliability and accuracy of the model.

[0021] Optionally, the baseline model is obtained by training in the following manner: Based on the second loss function and the training sample set, a baseline model associated with the object to be monitored is trained, wherein the second loss function is associated with the predicted benchmark range of the indicators under the multiple states, the actual benchmark range of the indicators under the multiple states, the state values ​​of the multiple states, and the upper and lower limits of the residual tolerance.

[0022] In this embodiment, an associated baseline model is trained for each object to be monitored. Furthermore, when training the baseline model, the state values ​​of multiple states are combined to make the obtained baseline model more targeted and applicable to health monitoring under different states. This avoids the problem of the indicator benchmark range being too rigid due to a single state, which makes the model less flexible.

[0023] On one hand, embodiments of this application provide a health monitoring device, which includes: The acquisition module is used to acquire the current vital signs values ​​of the object to be monitored. The calibration module is used to calibrate the current vital signs value through a data calibration model associated with the object to be monitored, and obtain the vital signs calibration value. The determination module is used to input the vital sign calibration values ​​and the current status data of the object to be monitored into the baseline model associated with the object to be monitored; The matching module determines the target benchmark range that matches the current state data through the baseline model, and obtains the health monitoring results of the subject to be monitored based on the correlation between the vital signs calibration value and the target benchmark range.

[0024] Optionally, the baseline model includes the benchmark range of indicators corresponding to the object to be monitored under multiple states; The matching module is specifically used for: Using the baseline model, based on the current state data of the object to be monitored, a matching target benchmark range is determined from the corresponding indicator benchmark ranges under the various states.

[0025] Optionally, the matching module is specifically used for: If the vital signs calibration value is within the target reference range, then the health monitoring result is determined to be normal; If the calibrated value of the vital signs is outside the target reference range, the health monitoring result is determined to be abnormal.

[0026] Optionally, the matching module is further configured to: When the health monitoring result of the object to be monitored is abnormal, an alarm is triggered, and health reminder information is obtained from the knowledge base based on the alarm type; The monitoring alert information is pushed to the object to be monitored.

[0027] Optionally, the calibration module is further configured to: Obtain historical vital signs data of the object to be monitored, and select a training sample set from the historical vital signs data; Based on the first loss function and the training sample set, a data calibration model associated with the object to be monitored is trained. The first loss function is associated with the predicted vital sign calibration data, the actual vital sign calibration data, and the fault point limitation parameter. The fault point limitation parameter is used to limit the influence of fault points on the first loss function. The fault point represents a training sample where the residual between the predicted vital sign calibration data and the actual vital sign calibration data is greater than a preset threshold.

[0028] Optionally, the calibration module is specifically used for: The on-the-job vital signs data are downsampled to obtain first vital signs data, and the pre-job vital signs data are upsampled to obtain second vital signs data. The validity and data volume of the first vital signs data and the second vital signs data are verified, and a training sample set is obtained based on the verified vital signs data.

[0029] Optionally, the calibration module is specifically used for: The validity of the first vital signs data and the second vital signs data is verified based on the variance of the first vital signs data and the variance of the second vital signs data. Based on the data volume of the first vital signs data and the data volume of the second vital signs data, a data volume verification is performed on the first vital signs data and the second vital signs data.

[0030] Optionally, the calibration module is specifically used for: Multiple sets of training samples are randomly selected from the training sample set; For multiple sets of training samples, the following steps are performed respectively: a candidate calibration model is trained based on the first loss function and a set of training samples, and the number of fault points in the set of training samples is determined based on the candidate calibration model; The candidate calibration model with the fewest corresponding fault points among the multiple candidate calibration models obtained is taken as the data calibration model.

[0031] Optionally, the determining module is further configured to: Based on the second loss function and the training sample set, a baseline model associated with the object to be monitored is trained, wherein the second loss function is associated with the predicted benchmark range of the indicators under the multiple states, the actual benchmark range of the indicators under the multiple states, the state values ​​of the multiple states, and the upper and lower limits of the residual tolerance.

[0032] On one hand, embodiments of this application provide a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the semantic understanding method described above.

[0033] On one hand, embodiments of this application provide a computer-readable storage medium storing a computer program executable by a computer device, which, when run on the computer device, causes the computer device to perform the steps of the semantic understanding method described above.

[0034] In this embodiment, by calibrating the current vital signs values ​​of the monitored subject, more accurate calibrated vital signs values ​​are obtained, which can effectively improve the accuracy of health monitoring. Secondly, different benchmark ranges of indicators are used to determine the health status for different states of the monitored subject, instead of using the same benchmark range for the same state. This avoids errors in health monitoring results due to different states, thereby improving the accuracy of health monitoring. Attached Figure Description

[0035] Figure 1 A system architecture diagram provided for an embodiment of this application; Figure 2 A flowchart illustrating a health monitoring method provided in this application embodiment. Figure 1 ; Figure 3 A flowchart illustrating a data calibration model training method provided in an embodiment of this application; Figure 4This is a schematic diagram illustrating the comparison between data fitted using the data calibration model and data not fitted using the model, provided in an embodiment of this application. Figure 5 This is a schematic diagram illustrating the blood pressure calibration effect provided in an embodiment of this application; Figure 6 A schematic diagram of a distribution function provided in an embodiment of this application; Figure 7 A schematic diagram comparing a baseline model and a linear model provided in an embodiment of this application; Figure 8 A flowchart illustrating a health monitoring method provided in this application embodiment. Figure 2 ; Figure 9 This is a schematic diagram of the structure of a data transmission device provided in an embodiment of this application; Figure 10 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0036] To make the objectives, technical solutions, and beneficial effects of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0037] refer to Figure 1 This is a system architecture diagram applicable to the embodiments of this application. The system includes at least a smart wearable device 101 and a server 102.

[0038] The smart wearable device 101 comes pre-installed with business applications for health monitoring, including client applications, web applications, and mini-program applications.

[0039] Server 102 is the backend server for business applications. Server 102 can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDN), and big data and artificial intelligence platforms.

[0040] The health monitoring method in this embodiment can be executed by the smart wearable device 101, the server 102, or both. This health monitoring method can be applied to health monitoring of personnel in various scenarios, such as driver health monitoring.

[0041] based on Figure 1 The system architecture diagram in this application illustrates a health monitoring method, such as... Figure 2 As shown, the process of this method is executed by a computer device, which can be... Figure 1 The smart wearable device 101 and / or server 102 shown include the following steps: Step 201: Obtain the current vital signs values ​​of the object to be monitored.

[0042] In this embodiment of the application, the collected vital signs include systolic blood pressure, diastolic blood pressure, body temperature, heart rate, and blood oxygen saturation. The vital sign values ​​are collected through a smart wearable device.

[0043] For example, the subject of monitoring could be a bus driver. While the subject is on duty, data is collected every 10 minutes using a smart wearable device, resulting in 144 data points per person per day.

[0044] Step 202: Calibrate the current vital signs values ​​using the data calibration model associated with the object to be monitored, and obtain the calibrated vital signs values.

[0045] Because data acquired through smart wearable devices can sometimes be inaccurate, this embodiment calibrates the current vital signs (VLS) values ​​of the monitored object after acquisition, thereby obtaining VLS calibration values. In practical applications, different monitored objects use different data calibration models; the acquired VLS values ​​are input into the corresponding data calibration model to obtain the VLS calibration values.

[0046] For example, let driver A be the subject to be monitored. Let a corresponding data calibration model X be established. Let the current systolic blood pressure of driver A collected by the smart wearable device be input into the data calibration model X to obtain the systolic blood pressure calibration value of driver A.

[0047] In addition, when one object to be monitored corresponds to one data calibration model, multiple sub-modules can be set in the data calibration model, each sub-module is used to calibrate a vital indicator value; when one object to be monitored corresponds to multiple data calibration models, each data calibration model is used to calibrate a vital indicator value.

[0048] Step 203: Input the vital signs calibration values ​​and the current status data of the object to be monitored into the baseline model associated with the object to be monitored.

[0049] Specifically, the types of states of the object to be monitored can be set according to the actual situation. For example, when the object to be monitored is a driver, the types of states include driving, entering or leaving a station, waiting for departure command (resting), low-intensity exercise, moderate-intensity exercise, and high-intensity exercise. Each state can be represented by a number, such as "driving" represented by the number 1. The current state of the object to be monitored can be one of the above states. The current state of the object to be monitored can be obtained through the dispatch system or through detection by smart wearable devices.

[0050] Step 204: Determine the target baseline range that matches the current status data through the baseline model, and obtain the health monitoring results of the subject to be monitored based on the correlation between the vital signs calibration values ​​and the target baseline range.

[0051] Specifically, each monitored object corresponds to a baseline model. Multiple sub-modules can be set within a baseline model, each used for monitoring a specific vital sign. Alternatively, multiple baseline models can be set, each used for monitoring a specific vital sign.

[0052] For example, let's set the subject to be monitored as driver A, establish a corresponding baseline model M, and input the current systolic blood pressure of driver A collected by the smart wearable device into the baseline model M to obtain the reference range of driver A's systolic blood pressure.

[0053] In addition, when one object to be monitored corresponds to one baseline model, multiple sub-modules can be set in the baseline model, each sub-module is used to monitor the health of one vital sign value; when one object to be monitored corresponds to multiple baseline models, each baseline model is used to monitor the health of one vital sign value.

[0054] In this embodiment, after obtaining the current vital signs (VLS) values ​​of the monitored object, data calibration is performed on these VLS values ​​to obtain more accurate VLS calibration values. Therefore, determining the health monitoring results of the monitored object based on the VLS calibration values ​​can effectively improve the accuracy of health monitoring. Secondly, for the current state data of the monitored object, a corresponding target baseline range is obtained. Then, a health status judgment is made based on the correlation between the VLS calibration values ​​and the target baseline range, effectively improving the accuracy of the health monitoring results.

[0055] In some embodiments, the baseline model includes benchmark ranges for indicators corresponding to the object under monitoring in multiple states. Using the baseline model, a matching target benchmark range is determined from the benchmark ranges for indicators corresponding to the object under monitoring in multiple states.

[0056] After the baseline model is trained, the benchmark range of the monitored object under various states can be obtained for each life indicator.

[0057] For example, regarding systolic blood pressure, the baseline model includes the following five reference ranges for different states: 98-135 during driving, 130-145 when entering or leaving a station, 98-122 when waiting for departure (resting), 135-145 during low-intensity exercise, 138-150 during moderate-intensity exercise, and 145-160 during high-intensity exercise.

[0058] If driver A's current state is "driving", the baseline model shows that driver A's systolic blood pressure baseline range is 98-135 when the state is "driving".

[0059] In this embodiment of the application, health monitoring is performed by using different vital sign data corresponding to different states of the subject under different conditions, instead of using a baseline range under the same state, thus avoiding errors in health monitoring results due to different states.

[0060] In step 204 above, if the vital signs calibration value is within the target reference range, the health monitoring result is determined to be normal; if the vital signs calibration value is outside the target reference range, the health monitoring result is determined to be abnormal.

[0061] For example, if driver A's calibrated systolic blood pressure is 105 during driving, the baseline model determines that driver A's systolic blood pressure range during driving is 98-135. Since driver A's calibrated systolic blood pressure is within the baseline range, the health monitoring result is determined to be normal.

[0062] In some embodiments, when the health monitoring results of the subject being monitored exceed the baseline range, the following steps are taken: When the health monitoring result of the object to be monitored is abnormal, an alarm is triggered, and health reminder information is retrieved from the knowledge base based on the alarm type, and the monitoring reminder information is pushed to the object to be monitored.

[0063] When the calibrated vital signs of the monitored subject exceed the baseline range output by the baseline model, the health monitoring result is considered abnormal, triggering an alarm. Alarm types include: high systolic blood pressure, low systolic blood pressure, high diastolic blood pressure, low diastolic blood pressure, tachycardia, bradycardia, low blood oxygen saturation, high body temperature, and low body temperature. Based on the alarm type, the system retrieves medical explanations, lifestyle guidance, and treatment suggestions from the linked medical knowledge base. After generating the health reminder, it is pushed to the monitored subject for further examination. Furthermore, the health monitoring results are sent to the central monitoring system to assist in the monitoring of the monitored subject. For example, this can be used to adjust vehicle and driver scheduling, protect driver health, and reduce traffic accidents.

[0064] For example, the baseline model determines that driver A's systolic blood pressure should be within the range of 98-135. If driver A's current systolic blood pressure is 85, outside this range, then driver A is considered to have an abnormal health monitoring result, triggering the alarm system with the alarm type "Low Systolic Blood Pressure!". The system will also generate information such as dietary guidance to raise systolic blood pressure. The abnormal health monitoring result "Low Systolic Blood Pressure!" is sent to driver A, along with the dietary guidance. Simultaneously, the abnormal health monitoring result is also sent to the bus company's central monitoring system. Based on the results, the company will take appropriate measures, such as temporarily suspending driver A's driving duties.

[0065] In this embodiment, an alarm message will be sent to the monitored object whose health monitoring result is abnormal. Health reminder information will also be pushed to the monitored object according to the type of alarm, which improves the intelligence of the entire health monitoring and thus greatly enhances the user experience.

[0066] In some embodiments, the data calibration model associated with the object to be monitored is trained in the following manner, specifically as follows: Figure 3 As shown, it includes: Step 301: Obtain historical vital signs data of the object to be monitored, and select a training sample set from the historical vital signs data.

[0067] Specifically, the acquired historical vital signs data undergoes preliminary outlier filtering. Maximum and minimum values ​​are set for five vital signs, and only data falling between these values ​​is retained. Values ​​above the maximum or below the minimum are considered outliers. The specific maximum and minimum values ​​for the five vital signs are shown in the table below: Table 1

[0068] Step 302: Based on the first loss function and the training sample set, train the data calibration model associated with the object to be monitored.

[0069] Specifically, the first loss function is associated with the predicted vital sign calibration data, the actual vital sign calibration data, and the fault point limitation parameter. The fault point limitation parameter is used to limit the influence of fault points on the first loss function. The fault point refers to a training sample where the residual between the predicted vital sign calibration data and the actual vital sign calibration data is greater than a preset threshold.

[0070] The data calibration model is represented by the following formula (1): …(1) Where w and b represent model parameters, This represents the input vital signs value. This indicates the predicted calibration values ​​for vital signs.

[0071] In traditional linear models, the loss function design is greatly affected by fault points. In order to reduce the proportion of fault points in the loss function, the loss function of the model needs to be optimized. Therefore, in this embodiment, the first loss function is determined based on the predicted vital signs calibration data, the actual vital signs calibration data, and the fault point limitation parameters, as shown in the following formula (2): …(2) Here, var represents the fault point limitation parameter, and w, b, and var are all learned functions. Tool The body is shown in the following formula (3): …(3) Additionally, a regularization term can be added to the loss function to prevent overfitting.

[0072] To verify the performance of the data calibration model in this application, the fitting effect of the data calibration model was compared with the fitting effect of the traditional linear model, such as... Figure 4 As shown, the solid line represents the fitting effect of the data calibration model in this application, and the dashed line represents the fitting effect of the traditional linear model. By comparison, it can be seen that the regression effect of the data calibration model in this application is better than that of the traditional linear model.

[0073] In addition, this application conducted experiments to test the predictive performance of the data calibration model, such as... Figure 5 As shown, compared to the pre-calibration vital signs values ​​obtained directly from smart wearable devices, the accuracy of the vital signs values ​​calibrated using the data calibration model is significantly improved and is closer to the true values ​​of the vital signs.

[0074] In this embodiment, the reliability of subsequent model learning is improved by screening historical vital sign data of the monitored objects. By optimizing the loss function, parameters that constrain the loss function are added to the traditional linear model, thereby reducing the proportion of fault points in the loss function and improving the accuracy of the data calibration model.

[0075] In some embodiments, historical vital sign data includes on-the-job vital sign data collected through smart wearable devices and pre-employment vital sign data collected through medical instruments; the on-the-job vital sign data is downsampled to obtain first vital sign data, and the pre-employment vital sign data is upsampled to obtain second vital sign data; the first and second vital sign data are validated for validity and data volume, and a training sample set is obtained based on the validated vital sign data.

[0076] Specifically, the historical vital signs data used are all derived from on-the-job and pre-job vital signs data, excluding vital signs data of the subjects on rest days. Therefore, based on the bus company's driver dispatch system, the drivers' work hours are queried, and vital signs data collected by smart wearable devices and medical instruments during work hours are selected.

[0077] While on duty, the monitored individuals acquire one vital signs data point every 10 minutes using smart wearable devices, totaling 144 data points per person per day. Before starting work, vital signs data are collected using medical instruments, with one data point collected per person per day. Due to the large amount of collected on-duty vital signs data, downsampling was performed to reduce the number of data points collected. Conversely, due to the limited amount of collected pre-work vital signs data, upsampling was performed to increase the number of data points collected.

[0078] In this embodiment, the on-the-job vital signs data are downsampled to prevent excessively large on-the-job vital signs data from causing low efficiency in training the model; the pre-employment vital signs data are upsampled to prevent excessively small pre-employment vital signs data from causing low accuracy in training the model. After the vital signs data are collected, validity and data volume are verified, providing a more accurate amount of data for subsequent model training, thereby improving the model's accuracy.

[0079] In some embodiments, the process of validating the first vital sign data and the second vital sign data and verifying the data volume includes: The validity of the first and second life indicator data is validated based on their variances. The data volume of both the first and second life indicator data is also validated based on their respective data sizes.

[0080] Specifically, for the individuals to be monitored, the second vital signs data collected before employment include: j1, j2, ..., j n The first vital signs data collected by smart wearable devices include: k1, x2, ..., x n Calculate the mean u1 and variance t1 of the second vital signs data collected before employment, and the mean u2 and variance t2 of the first vital signs data collected by the smart wearable device. Construct the statistic t1 / t2, learn the distribution function of t1 / t2, and then determine whether the current value is a low-probability event based on the distribution function. Whether it is a low-probability event can be determined according to business requirements; generally, events with a probability less than 0.01 are considered low-probability events. Figure 6 The solid line shown represents the distribution of data volume 30, and the dashed line represents the distribution of data volume 200. Figure 6 The probability of the distribution function represented by the solid and dashed lines is greater than 0.01, so they are not low-probability events, which also shows that the vital signs data corresponding to the solid and dashed lines have passed the validity check.

[0081] Because model accuracy decreases with too little data and efficiency decreases with too much data, it's necessary to validate the vital signs data based on its volume. Specifically, if the number of primary vital signs (PMR) data points is less than 30 or the number of secondary vital signs (TMR) data points is less than 3, model learning is not performed. If the number of PMR or TMR data points is greater than 200, the data is downsampled to 200 data points. For example, if fewer than 3 data points are collected by medical instruments, they are not used as a reference; if more than 200 data points are collected by wearable devices, only 200 data points are retained, which can be randomly selected.

[0082] In this embodiment of the application, by performing variance calculation on the vital signs data and verifying the size of the data, it is possible to determine whether the collected first vital signs data and second vital signs data are valid, and whether the size of the data meets the training requirements, thereby improving the quality of the training data and thus improving the accuracy of the model.

[0083] In some embodiments, the process of training a data calibration model associated with the object to be monitored based on a first loss function and a training sample set includes: Multiple training samples are randomly selected from the training sample set.

[0084] Multiple training samples are randomly selected from all training samples, with each group containing the same amount of data. For example, if there are 100 training samples in the set, 50 samples are randomly selected from these 100 samples as one group. Multiple training samples are randomly selected using replacement.

[0085] For multiple sets of training samples, the following steps are performed respectively: a candidate calibration model is trained based on the first loss function and a set of training samples, and the number of fault points in a set of training samples is determined based on the candidate calibration model.

[0086] Each set of training samples is used to train a candidate calibration model based on the first loss function. The number of fault points in the set of training samples under the candidate calibration model is calculated. A fault point represents a training sample in the set of training samples where the residual between the life indicator calibration data predicted by the candidate calibration model and the actual life indicator calibration data is greater than a preset threshold.

[0087] The candidate calibration model with the fewest corresponding fault points among the multiple candidate calibration models obtained is used as the data calibration model.

[0088] For example, five training sample sets are randomly selected from the training sample set, with each set containing 50 training samples. The first set of training samples and the first loss function are used to train candidate calibration model 1, and the number of fault points in the first set of training samples is calculated as a1. Similarly, four other models are trained, denoted as candidate calibration model 2, candidate calibration model 3, candidate calibration model 4, and candidate calibration model 5, with corresponding numbers of fault points a2, a3, a4, and a5, respectively.

[0089] By comparing the values ​​of a1, a2, a3, a4, and a5, it was found that a5 is the smallest, meaning that the fifth group of training samples has the fewest fault points. Therefore, the fifth group is selected as the candidate data for model calibration.

[0090] In this embodiment, by randomly selecting multiple sets of training samples, the potential for excessive model error due to special circumstances is reduced. After model training is completed, the model with the fewest fault points is selected as the candidate model, further increasing the reliability and accuracy of the model.

[0091] In some embodiments, the baseline model is trained in the following manner: a baseline model associated with the object to be monitored is trained based on a second loss function and a training sample set, wherein the second loss function is associated with the predicted benchmark range of indicators under multiple states, the true benchmark range of indicators under multiple states, the state values ​​under multiple states, and the upper and lower limits of the residual tolerance.

[0092] Constructing the dataset: Using calibrated smart wearable device data m1, m2, ..., m n Construct M = {m1, m2, ..., m} n-1}, y={ m2,m3,…,m n}, where n>=5.

[0093] When the amount of data is less than 5, the baseline range of life indicators for each state is as shown in Table 2 below.

[0094] Table 2

[0095] As shown in Table 2, the baseline range for systolic blood pressure is 100-140, the baseline range for diastolic blood pressure is 60-100, the baseline range for body temperature is 35.5-37.5, the baseline range for heart rate is 50-100, and the baseline range for blood oxygen is 87-107.

[0096] Model training: The baseline model is represented by the following formula (4): …(4) Where y(w,b,k,x) is the target value to be predicted, w, b, and k are model parameters, w is the correlation coefficient, b is the intercept, and k is the weight of different states. x and I are inputs, where x is the monitored vital signs data and I is the value of different states.

[0097] In this embodiment of the application, the second loss function is determined based on the predicted benchmark range of indicators under multiple states, the actual benchmark range of indicators under multiple states, the state values ​​of multiple states, and the upper and lower limits of the residual tolerance, as shown in the following formula (5): …(5) Among them, the function Specifically, as shown in the following formula (6): …(6) Here, q represents the residual tolerance. In actual training, q needs to learn two values: an upper limit (default 0.95) and a lower limit (default 0.05) for the residual tolerance. The value of q needs to be learned from data collected from all drivers. The parameters of the other items need to be learned from data collected from one driver, ultimately resulting in a personalized baseline model.

[0098] After determining the upper limit of the residual tolerance, a set of w, b, k, and the corresponding upper limit baseline model can be learned using the upper limit value of the residual tolerance and the second loss function. Based on the upper limit baseline model, the benchmark upper limit of the indicators of the monitored object under various states can be obtained.

[0099] Similarly, after determining the lower limit of the residual tolerance, another set of w, b, and k, as well as the corresponding lower limit baseline model, can be learned using the lower limit of the residual tolerance and the second loss function. Based on the lower limit baseline model, the benchmark lower limits of the indicators of the monitored object under various states can be obtained.

[0100] Based on the upper and lower limits of the indicator benchmarks for the monitored object under various states, the indicator benchmark range for the monitored object under various states is obtained. Specifically, as follows: Figure 7 As shown, the dashed line represents the curve learned by the linear model, while the solid line represents the baseline range of the indicators output by the model in this application. It can be observed that the range of vital signs values ​​changes continuously under different states, consistent with our expected results, and is more accurate than methods using fixed thresholds.

[0101] In this embodiment, an associated baseline model is trained for each object to be monitored. Furthermore, when training the baseline model, the state values ​​of multiple states are combined to make the obtained baseline model more targeted and applicable to health monitoring under different states. This avoids the problem of the indicator benchmark range being too rigid due to a single state, which makes the model less flexible.

[0102] To better explain the embodiments of this application, a flowchart of a health monitoring method provided by this application is described below with reference to a specific implementation scenario, taking the collection of driver A's systolic blood pressure as an example. Figure 8 As shown: The system acquires driver A's systolic blood pressure data via a smart wearable device. This data is then input into a calibration model to obtain a calibrated systolic blood pressure value. Combining this data with motion information from the dispatch system and other sensors, driver A's status data is determined. The status data and calibrated systolic blood pressure value are then input into a baseline model. A baseline systolic blood pressure reference range matching driver A's status data is determined from the baseline model, and the calibrated value is compared to this range. If the calibrated value is outside the reference range, an alarm is generated, and relevant medical information regarding raising / lowering systolic blood pressure is retrieved from a medical knowledge base and sent to driver A (the recipient of care). Simultaneously, the alarm is sent to the dispatch system so that it can adjust driver A's work schedule.

[0103] Based on the same technical concept, this application provides a health monitoring device 900, such as... Figure 9 As shown, the device includes: Module 901 is used to acquire the current vital signs values ​​of the object to be monitored; Calibration module 902 is used to calibrate the current vital signs value through a data calibration model associated with the object to be monitored, and obtain a vital signs calibration value; The determination module 903 is used to input the vital sign calibration value and the current status data of the object to be monitored into the baseline model associated with the object to be monitored; The matching module 904 determines the target benchmark range that matches the current state data through the baseline model, and obtains the health monitoring results of the subject to be monitored based on the correlation between the vital signs calibration value and the target benchmark range.

[0104] Optionally, the matching module 904 is specifically used for: Using the baseline model, based on the current state data of the object to be monitored, a matching target benchmark range is determined from the corresponding indicator benchmark ranges under the various states.

[0105] Optionally, the matching module 904 is specifically used for: If the vital signs calibration value is within the target reference range, then the health monitoring result is determined to be normal; If the calibrated value of the vital signs is outside the target reference range, the health monitoring result is determined to be abnormal.

[0106] Optionally, the matching module 904 is further configured to: When the health monitoring result of the object to be monitored is abnormal, an alarm is triggered, and health reminder information is obtained from the knowledge base based on the alarm type; The monitoring alert information is pushed to the object to be monitored.

[0107] Optionally, the calibration module 902 is further configured to: Obtain historical vital signs data of the object to be monitored, and select a training sample set from the historical vital signs data; Based on the first loss function and the training sample set, a data calibration model associated with the object to be monitored is trained. The first loss function is associated with the predicted vital sign calibration data, the actual vital sign calibration data, and the fault point limitation parameter. The fault point limitation parameter is used to limit the influence of fault points on the first loss function. The fault point represents a training sample where the residual between the predicted vital sign calibration data and the actual vital sign calibration data is greater than a preset threshold.

[0108] Optionally, the acquisition module 901 is specifically used for: The on-the-job vital signs data are downsampled to obtain first vital signs data, and the pre-job vital signs data are upsampled to obtain second vital signs data. The validity and data volume of the first vital signs data and the second vital signs data are verified, and a training sample set is obtained based on the verified vital signs data.

[0109] Optionally, the calibration module 902 is specifically used for: The validity of the first vital signs data and the second vital signs data is verified based on the variance of the first vital signs data and the variance of the second vital signs data. Based on the data volume of the first vital signs data and the data volume of the second vital signs data, a data volume verification is performed on the first vital signs data and the second vital signs data.

[0110] Optionally, the calibration module 902 is specifically used for: Multiple sets of training samples are randomly selected from the training sample set; For multiple sets of training samples, the following steps are performed respectively: a candidate calibration model is trained based on the first loss function and a set of training samples, and the number of fault points in the set of training samples is determined based on the candidate calibration model; The candidate calibration model with the fewest corresponding fault points among the multiple candidate calibration models obtained is taken as the data calibration model.

[0111] Optionally, the determining module 903 is specifically used for: Based on the second loss function and the training sample set, a baseline model associated with the object to be monitored is trained, wherein the second loss function is associated with the predicted benchmark range of the indicators under the multiple states, the actual benchmark range of the indicators under the multiple states, the state values ​​of the multiple states, and the upper and lower limits of the residual tolerance.

[0112] Based on the same technical concept, embodiments of this application provide a computer device, which may be a terminal or a server, such as... Figure 10 The embodiment includes at least one processor 1001 and a memory 1002 connected to the at least one processor. The specific connection medium between the processor 1001 and the memory 1002 is not limited in this embodiment. Figure 10 Taking the connection between processor 1001 and memory 1002 via a bus as an example, the bus can be divided into address bus, data bus, control bus, etc.

[0113] In this embodiment of the application, the memory 1002 stores instructions that can be executed by at least one processor 1001. By executing the instructions stored in the memory 1002, at least one processor 1001 can perform the steps included in the above-described health monitoring method.

[0114] The processor 1001 is the control center of the computer device, capable of connecting to various parts of the computer device via various interfaces and lines. It performs health monitoring by running or executing instructions stored in the memory 1002 and accessing data stored in the memory 1002. Optionally, the processor 1001 may include one or more processing units. The processor 1001 may integrate an application processor and a modem processor. The application processor primarily handles the operating system, user interface, and applications, while the modem processor primarily handles wireless communication. It is understood that the modem processor may not be integrated into the processor 1001. In some embodiments, the processor 1001 and the memory 1002 may be implemented on the same chip; in other embodiments, they may be implemented on separate chips.

[0115] The processor 1001 can be a general-purpose processor, such as a central processing unit (CPU), digital signal processor, application-specific integrated circuit (ASIC), field-programmable gate array (FPGA), or other programmable logic device, discrete gate or transistor logic device, or discrete hardware component, capable of implementing or executing the methods, steps, and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this application can be directly manifested as being executed by a hardware processor, or executed by a combination of hardware and software modules within the processor.

[0116] Memory 1002, as a non-volatile computer-readable storage medium, can be used to store non-volatile software programs, non-volatile computer-executable programs, and modules. Memory 1002 may include at least one type of storage medium, such as flash memory, hard disk, multimedia card, card-type memory, random access memory (RAM), static random access memory (SRAM), programmable read-only memory (PROM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), magnetic memory, magnetic disk, optical disk, etc. Memory 1002 can be any other medium capable of carrying or storing desired program code in the form of instructions or data structures that can be accessed by a computer, but is not limited thereto. In the embodiments of this application, memory 1002 can also be a circuit or any other device capable of implementing storage functions for storing program instructions and / or data.

[0117] Based on the same inventive concept, embodiments of this application provide a computer-readable storage medium storing a computer program executable by a computer device, which, when run on the computer device, causes the computer device to perform the steps of the above-described health monitoring method.

[0118] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0119] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to this application. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0120] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0121] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0122] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A health monitoring method, characterized in that, include: Obtain the current vital signs values ​​of the object to be monitored; The current vital signs value is calibrated by using a data calibration model associated with the object to be monitored, and a vital signs calibration value is obtained. The calibration values ​​of the vital signs and the current status data of the object to be monitored are input into the baseline model associated with the object to be monitored. The baseline model is used to determine the target benchmark range that matches the current state data, and the health monitoring results of the subject to be monitored are obtained based on the correlation between the vital signs calibration values ​​and the target benchmark range. The data calibration model associated with the monitored object was trained in the following manner: Obtain historical vital signs data of the object to be monitored, and select a training sample set from the historical vital signs data; Based on the first loss function and the training sample set, a data calibration model associated with the object to be monitored is trained. The first loss function is associated with the predicted vital signs calibration data, the actual vital signs calibration data, and the fault point limitation parameter. The fault point limitation parameter is used to limit the influence of fault points on the first loss function. The fault point represents a training sample in which the residual between the predicted vital signs calibration data and the actual vital signs calibration data is greater than a preset threshold. The specific relationship of the first loss function determined by the fault point limiting parameters is as follows: Where var represents the fault point limitation parameter, x i This represents the input vital signs value, y. i The function represents the predicted calibration value of vital signs, w represents the model parameters, and the function is... The specifics are as follows: ; The step of training a data calibration model associated with the object to be monitored based on the first loss function and the training sample set includes: Multiple sets of training samples are randomly selected from the training sample set; For multiple sets of training samples, the following steps are performed respectively: a candidate calibration model is trained based on the first loss function and a set of training samples, and the number of fault points in the set of training samples is determined based on the candidate calibration model; The candidate calibration model with the fewest corresponding fault points among the multiple candidate calibration models obtained is taken as the data calibration model.

2. The method as described in claim 1, characterized in that, The baseline model includes the benchmark range of indicators corresponding to the monitored object under various states; The step of determining the target benchmark range that matches the current state data through the baseline model includes: Using the baseline model, based on the current state data of the object to be monitored, a matching target benchmark range is determined from the corresponding indicator benchmark ranges under the various states.

3. The method as described in claim 1, characterized in that, The determination of the health monitoring results of the subject under monitoring based on the correlation between the calibrated vital signs values ​​and the target baseline range includes: If the vital signs calibration value is within the target reference range, then the health monitoring result is determined to be normal; If the calibrated value of the vital signs is outside the target reference range, the health monitoring result is determined to be abnormal.

4. The method as described in claim 1, characterized in that, Also includes: When the health monitoring result of the object to be monitored is abnormal, an alarm is triggered, and health reminder information is obtained from the knowledge base based on the alarm type; The health reminder information is pushed to the subject to be monitored.

5. The method as described in claim 1, characterized in that, The historical vital signs data includes on-the-job vital signs data collected through smart wearable devices, and pre-employment vital signs data collected through medical instruments. The step of selecting a training sample set from the historical vital signs data includes: The on-the-job vital signs data are downsampled to obtain first vital signs data, and the pre-job vital signs data are upsampled to obtain second vital signs data. The validity and data volume of the first vital signs data and the second vital signs data are verified, and a training sample set is obtained based on the verified vital signs data.

6. The method as described in claim 5, characterized in that, The validity and data volume of the first vital sign data and the second vital sign data are verified. The validity of the first vital signs data and the second vital signs data is verified based on the variance of the first vital signs data and the variance of the second vital signs data. Based on the data volume of the first vital signs data and the data volume of the second vital signs data, a data volume verification is performed on the first vital signs data and the second vital signs data.

7. The method as described in claim 1, characterized in that, The baseline model was obtained by training in the following manner: Based on the second loss function and the training sample set, a baseline model associated with the object to be monitored is trained, wherein the second loss function is associated with the predicted benchmark range of indicators under multiple states, the actual benchmark range of indicators under multiple states, the state values ​​of the multiple states, and the upper and lower limits of the residual tolerance.

8. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the method according to any one of claims 1 to 7.

Images