Deep learning model optimization method in sleep health monitoring

Abstract

The invention discloses a deep learning model optimization method in sleep health monitoring, and belongs to the technical field of model optimization. Monitoring and counting the exercise data and beverage data of the user on the current day, performing targeted data processing and analysis by using different influence modules in the sleep health monitoring model, and performing reliability verification on basic sleep quality data corresponding to sleep at night by using the exercise data and beverage data of the user on the current day; performing reliability analysis of influence effects on a sleep positive influence module, a sleep negative influence module and a sleep fusion influence module in the sleep health monitoring model by using sleep influence verification data corresponding to a plurality of times of sleep health monitoring of the user; dynamically managing influence processing schemes implemented by different influence modules in the sleep health monitoring model in a self-adaptive manner according to an evaluation result; the method and the device are used for solving the technical problem of poor personalized monitoring analysis and management effects of a deep learning model for different individual users in the existing scheme.

Description

Technical Field

[0001] The present invention relates to the technical field of model optimization, and particularly to a deep learning model optimization method in sleep health monitoring. Background Art

[0002] In the field of sleep health monitoring, deep learning models are increasingly widely used. They can automatically extract features from complex sleep data and perform accurate classification and prediction.

[0003] When implementing existing deep learning models in sleep health monitoring, most of them still stay at the monitoring, statistics, analysis, and prompt of single-user sleep data and sleep environment data, without combining the user's daily exercise data and drink data for supervision and fusion analysis. As a result, the reliability of sleep suggestions and health management prompts for user sleep health monitoring is poor, and there are problems with the personalized monitoring analysis and autonomous optimization effects of deep learning models for sleep health monitoring for different individual users. Summary of the Invention

[0004] The purpose of the present invention is to provide a deep learning model optimization method in sleep health monitoring, which is used to solve the technical problem that the personalized monitoring analysis and autonomous optimization effects of deep learning models for sleep health monitoring in the existing solutions are poor for different individual users.

[0005] The purpose of the present invention can be achieved through the following technical solutions:

[0006] A deep learning model optimization method in sleep health monitoring includes:

[0007] Monitoring and statistics of the user's daily exercise data and drink data, and processing and analyzing them using the sleep positive impact module, sleep negative impact module, or sleep fusion impact module in the sleep health monitoring model to obtain sleep impact processing data corresponding to the user's daily exercise data and drink data;

[0008] Monitoring and analyzing the user's sleep data through the sleep health monitoring model, and outputting basic sleep quality data corresponding to the user's night sleep. Using the sleep impact processing data to verify the reliability of the basic sleep quality data to obtain sleep impact verification data corresponding to the user's single sleep health monitoring;

[0009] Using the sleep impact verification data corresponding to the user's multiple sleep health monitoring to perform a reliability analysis of the impact effects of the sleep positive impact module, sleep negative impact module, and sleep fusion impact module in the sleep health monitoring model, and dynamically manage the impact processing schemes implemented by different impact modules in the sleep health monitoring model according to the evaluation results.

[0010] Preferably, when there is only exercise data, the exercise data of the user on the current day is processed and analyzed. If the exercise time period in the exercise data is not within the user's bedtime influence period, a positive influence processing instruction is generated, and according to the positive influence processing instruction, the exercise intensity value and exercise duration in the exercise data are analyzed through the sleep positive influence processing scheme associated with the sleep positive influence module, and the corresponding sleep positive influence processing data is output.

[0011] Preferably, if the exercise time period in the exercise data is within the user's bedtime influence period, a negative influence processing instruction is generated, and according to the negative influence processing instruction, the exercise intensity value and exercise duration in the exercise data are analyzed through the first sleep negative influence processing scheme associated with the sleep negative influence module, and the corresponding first sleep negative influence processing data is output.

[0012] Preferably, when there is only beverage data, the drinking time point and drinking amount in the beverage data are analyzed through the second sleep negative influence processing scheme associated with the sleep negative influence module, and the corresponding second sleep negative influence processing data is output.

[0013] Preferably, when there are both exercise data and beverage data, the exercise intensity value and exercise duration in the exercise data and the drinking time point and drinking amount in the beverage data are analyzed through the sleep fusion influence processing scheme associated with the sleep fusion influence module, and the corresponding sleep fusion influence processing data is output.

[0014] Preferably, the total sleep duration, total deep sleep duration, and total light sleep duration of the user in the basic sleep quality data are obtained, and the total sleep duration, total deep sleep duration, and total light sleep duration are respectively used to calculate the corresponding reliable sleep index values through the sleep influence processing data.

[0015] Preferably, the reliable sleep index values are analyzed, and according to the analysis results, the influence identification flag corresponding to the sleep index is set to 0 or 1;

[0016] The influence identification flags obtained by corresponding processing of all sleep indexes are sorted and combined to obtain a single - time influence identification processing sequence.

[0017] Preferably, all elements in the single - time influence identification processing sequence are summed and analyzed;

[0018] If the sum value of all elements is 0, the total number of completely effective processing of the influence processing scheme corresponding to the sleep influence processing data is incremented by 1;

[0019] If the sum value of all elements is greater than 0 and less than 3, the total number of partially effective processing of the influence processing scheme corresponding to the sleep influence processing data is incremented by 1;

[0020] If the sum value of the elements is equal to 3, increment the total number of completely ineffective processing of the impact processing plan corresponding to the sleep impact processing data to which it belongs by one;

[0021] Sort and combine the total number of completely effective processing, the total number of partially effective processing, or the total number of completely ineffective processing for the single impact recognition processing sequence obtained by processing and the data update of the corresponding impact processing plan to obtain the sleep impact verification data corresponding to the user's single sleep health monitoring.

[0022] Preferably, according to the sleep impact verification data corresponding to the user's multiple sleep health monitoring, respectively count the total number of completely effective processing, the total number of partially effective processing, and the total number of completely ineffective processing associated with the sleep positive impact module, the sleep negative impact module, and the sleep fusion impact module in the sleep health monitoring model, and obtain the reliable impact effect values corresponding to different impact modules through calculation;

[0023] Perform data analysis on the reliable impact effect values to determine the sleep impact prediction processing effect of the corresponding impact module.

[0024] Preferably, if the reliable impact effect value is less than or equal to 1, it is determined that the sleep impact prediction processing effect of the corresponding impact module is normal, and the implementation of the existing impact processing plan of the affiliated impact module is maintained in the future;

[0025] Otherwise, it is determined that the sleep impact prediction processing effect of the corresponding impact module is abnormal, and dynamic management is performed on the existing impact processing plan of the affiliated impact module.

[0026] Compared with the existing solutions, the beneficial effects achieved by the present invention are:

[0027] The present invention monitors and counts the user's daily exercise data and beverage data, and uses different impact modules in the sleep health monitoring model to perform targeted data processing and analysis, obtaining the sleep impact processing data corresponding to the user's daily exercise data and beverage data, which can provide diverse sleep impact data support for subsequent sleep suggestions and health management tips for user sleep health monitoring.

[0028] The present invention uses the user's daily exercise data and beverage data to verify the reliability of the basic sleep quality data corresponding to their sleep at night, determines whether the user's daily exercise and beverage consumption have an impact on their sleep quality at night, so as to provide reliable sleep suggestions for the user in the future, improving the reliability and pertinence of sleep health monitoring analysis and suggestions for different users.

[0029] The present invention uses the sleep impact verification data corresponding to several sleep health monitoring of users to perform reliability analysis on the sleep positive impact module, sleep negative impact module, and sleep fusion impact module in the sleep health monitoring model, and adaptively manages the impact processing schemes implemented by different impact modules in the sleep health monitoring model according to the evaluation results, realizing the autonomous supervision and optimization of the implementation effects of different user sleep health monitoring models, and improving the personalized monitoring analysis and autonomous optimization effects of the deep learning model for sleep health monitoring for different individual users. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] The present invention will be further described below with reference to the accompanying drawings.

[0031] Figure 1 It is a block diagram of the module for the optimization method of the deep learning model in the sleep health monitoring of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0032] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0033] As Figure 1 shown, the present invention is an optimization method for a deep learning model in sleep health monitoring, including:

[0034] Monitoring and counting the exercise data and beverage data of the user on the current day, and using the sleep positive impact module, sleep negative impact module, or sleep fusion impact module in the sleep health monitoring model to perform processing and analysis to obtain the sleep impact processing data corresponding to the exercise data and beverage data of the user on the current day;

[0035] Among them, the sleep impact processing data is specifically sleep positive impact processing data, first sleep negative impact processing data, second sleep negative impact processing data, or sleep fusion impact processing data; including:

[0036] When only exercise data exists, the exercise data of the user on the current day is processed and analyzed. If the exercise time period in the exercise data is not within the pre-sleep impact period of the user, a positive impact processing instruction is generated, and according to the positive impact processing instruction, the exercise intensity value and exercise duration in the exercise data are analyzed through the sleep positive impact processing scheme associated with the sleep positive impact module, and the corresponding sleep positive impact processing data is output;

[0037] It should be noted that the pre-sleep impact period is specifically one hour before the user goes to bed;

[0038] In addition, the exercise intensity value is determined according to the median value of all heart rates corresponding to the user's exercise time period; in the data test of the existing exercise affecting sleep prediction model, for every additional 30 minutes of moderate-to-high-intensity exercise during the day, the predicted deep sleep duration of the model increases by about 8 - 12 minutes; the exercise time window analysis shows that exercise 3 - 4 hours before bedtime has the greatest promoting effect on deep sleep;

[0039] In addition, the positive impact of exercise on sleep is specifically the total deep sleep duration index; moderate exercise, especially moderate-to-high-intensity activities during the day, significantly increases the deep sleep duration by raising the core body temperature and promoting subsequent body temperature decline.

[0040] The sleep positive impact processing solution is used to match the exercise data with the exercise positive impact experimental data set and output the corresponding sleep positive impact processing data; the exercise positive impact experimental data set includes several experimental exercise intensity value ranges, experimental exercise duration ranges, and the corresponding estimated impact duration of deep sleep;

[0041] The sleep positive impact processing data specifically includes the estimated impact duration of deep sleep;

[0042] If the exercise time period in the exercise data is within the user's pre-sleep impact period, a negative impact processing instruction is generated, and according to the negative impact processing instruction, the exercise intensity value and exercise duration in the exercise data are analyzed through the first sleep negative impact processing solution associated with the sleep negative impact module, and the corresponding first sleep negative impact processing data is output;

[0043] It should be noted that when part of the exercise time period is within the pre-sleep impact period, the data analysis of the overall exercise time period is also carried out using the first sleep negative impact processing solution;

[0044] In addition, the negative impact of exercise on sleep is specifically the total deep sleep duration index;

[0045] Specifically, strenuous exercise within 1 hour before bedtime causes sympathetic nerve excitement, briefly reducing the proportion of deep sleep (about a 5 - 10% decrease), but the overall impact on the total sleep duration index is small;

[0046] The first sleep negative impact processing solution is used to match the exercise data with the exercise negative impact experimental data set and output the corresponding first sleep negative impact processing data;

[0047] The exercise negative impact experimental data set includes several experimental exercise intensity value ranges, experimental exercise duration ranges, and the corresponding estimated impact duration of deep sleep; the experimental data in the exercise negative impact experimental data set is different from the experimental data in the exercise positive impact experimental data set;

[0048] The first sleep negative impact processing data specifically includes the estimated impact duration of deep sleep, with a negative value;

[0049] When only beverage data exists, the drinking time point and drinking amount in the beverage data are analyzed through the second sleep negative impact processing solution associated with the sleep negative impact module, and the corresponding second sleep negative impact processing data is output;

[0050] It should be noted that the beverage is specifically coffee; the beverage data needs to be filled in by the user through the intelligent terminal, for example, filled in and synchronously transmitted to the smart bracelet through the smart phone, or directly filled in through the smart bracelet;

[0051] In addition, the negative impact of beverages on sleep is specifically the total sleep duration index, the total deep sleep duration index, and the total light sleep duration index;

[0052] Caffeine directly inhibits the generation of deep sleep by antagonizing adenosine receptors; Experimental data: For every 100 mg of caffeine intake, the deep sleep stage is reduced by 9.2 ± 2.1 minutes, especially within 6 hours after intake; when the residual caffeine concentration ≥ 20 mg, the deep sleep duration decreases by 18% - 25%;

[0053] Caffeine prolongs the light sleep duration, resulting in sleep fragmentation; Experimental data: Intake of 200 mg of caffeine increases the light sleep proportion from 8% to 12%;

[0054] Caffeine prolongs the sleep latency and increases the number of nocturnal awakenings, resulting in a shortening of the total sleep duration; Experimental data: Intake of 200 mg of caffeine 3 hours before bedtime: The total sleep duration is reduced by 40 ± 12 minutes; The number of awakenings increases from 1.2 times to 2.8 times;

[0055] In addition, the second sleep negative impact processing solution is used to match the beverage data with the beverage negative impact experimental data set and output the corresponding second sleep negative impact processing data;

[0056] The beverage negative impact experimental data set includes several experimental drinking duration ranges and experimental drinking amount ranges, as well as the corresponding total estimated sleep impact duration, deep sleep estimated impact duration, and light sleep estimated impact duration;

[0057] The second sleep negative impact processing data specifically includes the total estimated sleep impact duration, deep sleep estimated impact duration, and light sleep estimated impact duration; the corresponding values are negative, negative, and positive respectively;

[0058] When both exercise data and beverage data exist, the exercise intensity value and exercise duration in the exercise data, as well as the drinking time point and drinking amount in the beverage data, are analyzed through the sleep fusion impact processing solution associated with the sleep fusion impact module, and the corresponding sleep fusion impact processing data is output;

[0059] Among them, when the user exercises and ingests caffeine on the same day, the physiological effects of the two will produce dynamic antagonism and synergy:

[0060] Interaction between thermoregulation and caffeine metabolism: The increase in body temperature caused by exercise accelerates caffeine metabolism, increasing the rate of decline in residual concentration by about 15%;

[0061] Caffeine safety threshold: When the daily exercise volume ≥ 60 minutes, a caffeine intake ≤ 150 mg can maintain the normal level of deep sleep (±5% fluctuation);

[0062] Exercise compensation limit: When the caffeine intake ≥ 300 mg, even if exercising for 90 minutes, deep sleep is still reduced by ≥ 10 minutes (compensation rate less than 50%);

[0063] In addition, a sleep fusion impact processing scheme is used to match the exercise data and beverage data together with the fusion negative impact experimental data set and output the corresponding sleep fusion impact processing data;

[0064] The fusion negative impact experimental data set includes several ranges of experimental exercise intensity values and experimental exercise duration ranges, several ranges of experimental drinking durations and experimental drinking amounts, as well as the corresponding total estimated sleep impact duration, deep sleep estimated impact duration, and light sleep estimated impact duration;

[0065] The sleep fusion impact processing data specifically includes the total estimated sleep impact duration, deep sleep estimated impact duration, and light sleep estimated impact duration; the corresponding values are negative, negative, and positive respectively;

[0066] The absolute values of the different index values in the sleep fusion impact processing data are all smaller than the absolute values of the corresponding index values in the second sleep negative impact processing data;

[0067] In the embodiments of the present invention, by monitoring and statistically analyzing the user's daily exercise data and beverage data, and using different impact modules in the sleep health monitoring model to perform targeted data processing and analysis, the sleep impact processing data corresponding to the user's daily exercise data and beverage data can be obtained, which can provide diverse sleep impact data support for subsequent sleep suggestions and health management tips for user sleep health monitoring.

[0068] By monitoring and analyzing the user's sleep data through the sleep health monitoring model and outputting the basic sleep quality data corresponding to the user's night sleep, using the sleep impact processing data to verify the reliability of the basic sleep quality data, the sleep impact verification data corresponding to the user's single sleep health monitoring is obtained; including:

[0069] Among them, the sleep health monitoring model is an existing deep learning model, such as CNN (Convolutional Neural Network): processing the temporal features in PPG or EEG signals;

[0070] LSTM (Long Short-Term Memory Network): Analyze the continuity of sleep stages (such as the transition from wakefulness to deep sleep); can monitor sleep data based on the smart bracelet worn by the user;

[0071] Monitoring and analyzing the user's sleep data by the sleep health monitoring model is an existing conventional technical means, and the specific implementation steps will not be elaborated here;

[0072] It can be understood that most of the existing sleep health monitoring and analysis solutions still stay at the monitoring and analysis of the user's sleep data, and conduct a combined analysis by combining the environmental data of the user's sleep to judge the user's sleep quality and give prompts; however, there is no active supervision, processing and analysis of the user's exercise data and beverage data on the same day, which leads to poor reliability of the prompts for the user's poor sleep quality and the corresponding generated sleep suggestions;

[0073] The basic sleep quality data includes the total sleep duration, total deep sleep duration and total light sleep duration of the user; the content of the basic sleep quality data can also be increased, deleted or modified according to the application requirements of the actual application scenario;

[0074] Obtain the total sleep duration, total deep sleep duration and total light sleep duration of the user in the basic sleep quality data, all in minutes, and respectively compare the total sleep duration, total deep sleep duration and total light sleep duration with the sleep impact processing data through the formula Calculate the corresponding sleep index reliability value Zk; in the formula, k is 1, 2, 3, which are the total sleep duration index, total deep sleep duration index, and total light sleep duration index respectively; Tk is T1, T2, T3, which are the total sleep duration, total deep sleep duration and total light sleep duration of the user in the basic sleep quality data respectively; t0k is t01, t02, t03, which are the standard total sleep duration, standard total deep sleep duration and standard total light sleep duration of the user respectively, and are obtained according to the median of different indicators of all the basic sleep quality data of the user without exercise and beverage consumption in the past; tk is t1, t2, t3, which are the total sleep duration impact value, total deep sleep duration impact value, and total light sleep duration impact value in the sleep impact processing set respectively, and are determined according to the sleep impact processing data obtained by corresponding processing of the user on the same day;

[0075] It should be noted that the sleep index reliability value is used to process and calculate the impact prediction data and sleep monitoring data corresponding to different indicators, so as to digitally represent the corresponding impact identification status;

[0076] If Zk ∈ [w1k, w2k], it is prompted that the impact identification corresponding to the sleep index is reliable, and the impact identification flag corresponding to the sleep index is set to 0; w1k and w2k are respectively the minimum error and the maximum error corresponding to different sleep indexes, both of which can be determined according to the test error data of several previous sleep experiments in the laboratory;

[0077] Otherwise, it is prompted that the impact identification corresponding to the sleep index is unreliable, and the impact identification flag corresponding to the sleep index is set to 1;

[0078] Sort and combine the impact identification flags obtained by processing all sleep indexes to obtain a single impact identification processing sequence;

[0079] In the embodiment of the present invention, by processing, digitally representing and combining the impact identification states corresponding to different sleep indexes, it is possible to not only monitor and analyze different sleep indexes, but also provide reliable local index data support for the sleep impact processing corresponding to the user's single sleep health monitoring;

[0080] Sum and analyze all elements in the single impact identification processing sequence;

[0081] If the sum value of all elements is 0, the total number of completely effective processing of the impact processing scheme corresponding to the sleep impact processing data is incremented by one;

[0082] If the sum value of all elements is greater than 0 and less than 3, the total number of partially effective processing of the impact processing scheme corresponding to the sleep impact processing data is incremented by one;

[0083] If the sum value of the elements is equal to 3, the total number of completely ineffective processing of the impact processing scheme corresponding to the sleep impact processing data is incremented by one;

[0084] Sort and combine the single impact identification processing sequence obtained by processing and the total number of completely effective processing, partially effective processing or completely ineffective processing of the corresponding impact processing scheme data update to obtain the sleep impact verification data corresponding to the user's single sleep health monitoring;

[0085] In the embodiment of the present invention, by using the user's daily exercise data and drink data to verify the reliability of the basic sleep quality data corresponding to his night sleep, it is determined whether the user's daily exercise and drink consumption have an impact on his night sleep quality, so as to provide reliable sleep suggestions for the user subsequently, improving the reliability and pertinence of different users' sleep health monitoring analysis and suggestions.

[0086] Perform reliability analysis on the sleep positive impact module, sleep negative impact module, and sleep fusion impact module in the sleep health monitoring model using the sleep impact verification data corresponding to several times of the user's sleep health monitoring, and dynamically manage the impact processing schemes implemented by different impact modules in the sleep health monitoring model according to the evaluation results; including:

[0087] According to the sleep impact verification data corresponding to several times of the user's sleep health monitoring, due to the differences in the bodies of different users, the reliability of their sleep health monitoring analysis is not good. Therefore, the specific values for several times should not be too small, at least 15 times. Respectively count the total number of completely effective processes, the total number of partially effective processes, and the total number of completely ineffective processes associated with the sleep positive impact module, sleep negative impact module, and sleep fusion impact module in the sleep health monitoring model, and through the formula Calculate the reliable value Kd of the impact effect corresponding to different impact modules; in the formula, d is 1, 2, 3, which are the sleep positive impact module, sleep negative impact module, and sleep fusion impact module respectively; n1d, n2d, n3d are the total number of completely effective processes, the total number of partially effective processes, and the total number of completely ineffective processes corresponding to different impact modules respectively; α is the error impact factor and is a real number greater than one, and the specific value is not limited; Ad is the reliable threshold of the impact effect corresponding to different impact modules, which is determined according to the test error data of several previous sleep experiments in the laboratory;

[0088] It should be noted that the reliable value of the impact effect is used to process and calculate all sleep impact data of different impact modules to digitally represent the corresponding sleep impact prediction processing effect;

[0089] It can be understood that the larger the values of the total number of completely ineffective processes and the total number of partially effective processes obtained by monitoring the corresponding impact module, the worse the sleep impact prediction processing effect on the user. The corresponding reason may be due to the user's physical constitution. Even if the user exercises and consumes coffee on the same day, it will not have a great impact on their sleep at night. Then, targeted management of the subsequent exercise of this impact module is required;

[0090] Perform data analysis on the reliable value of the impact effect to determine the sleep impact prediction processing effect of the corresponding impact module;

[0091] If the reliable value of the impact effect is less than or equal to 1, it is determined that the sleep impact prediction processing effect of the corresponding impact module is normal, and the subsequent implementation of the existing impact processing scheme of the affiliated impact module is maintained;

[0092] Otherwise, it is determined that the sleep impact prediction processing effect of the corresponding impact module is abnormal, and the existing impact processing scheme of the affiliated impact module is dynamically managed;

[0093] Among them, the existing impact processing solutions of the impact module are dynamically managed. Specifically, it can be to stop the subsequent implementation of the existing impact processing solution of the affiliated impact module, or to supplement and optimize the training of the experimental data set of the existing impact processing solution of the affiliated impact module with all the user's historical movement data, drink data, and corresponding basic sleep quality data, and use the optimized impact processing solution to perform subsequent sleep impact prediction processing on the user;

[0094] In addition, supplementing and optimizing the training of the experimental data set of the existing impact processing solution is a conventional technical means. For example, it can be optimized and trained through a Q-learning model. The specific implementation steps are not elaborated here.

[0095] In the embodiments of the present invention, the reliability analysis of the impact effects of the sleep positive impact module, sleep negative impact module, and sleep fusion impact module in the sleep health monitoring model is performed through the sleep impact verification data corresponding to several sleep health monitoring of the user, and the impact processing solutions implemented by different impact modules in the sleep health monitoring model are dynamically managed adaptively according to the evaluation results, realizing the autonomous supervision and optimization of the implementation effects of different user sleep health monitoring models, and improving the personalized monitoring analysis and autonomous optimization effects of the deep learning model for sleep health monitoring for different individual users.

[0096] In several embodiments provided by the present invention, it should be understood that the disclosed system can be implemented in other ways. For example, the above-described invention embodiments are merely illustrative. For example, the division of modules is only a logical function division, and there can be other division methods in actual implementation.

[0097] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical modules. They can be located in one place or distributed to multiple network modules. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0098] In addition, in each embodiment of the present invention, the various functional modules can be integrated in a processing module, or each module can exist physically alone, or two or more modules can be integrated in one module. The above-mentioned integrated modules can be implemented in the form of hardware or in the form of a combination of hardware and software functional modules.

[0099] For those skilled in the art, it is obvious that the present invention is not limited to the details of the above-mentioned exemplary embodiments, and can be implemented in other specific forms without departing from the basic characteristics of the present invention.

[0100] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A deep learning model optimization method for sleep health monitoring, characterized in that: include: Monitor and count the user's daily exercise data and beverage data, and use the sleep positive impact module, sleep negative impact module or sleep fusion impact module in the sleep health monitoring model to process and analyze them, and obtain sleep impact processing data corresponding to the user's daily exercise data and beverage data; The sleep health monitoring model is used to monitor and analyze the user's sleep data, and the basic sleep quality data corresponding to the user's sleep at night is output. The sleep impact processing data is used to verify the reliability of the basic sleep quality data, and the sleep impact verification data corresponding to the user's single sleep health monitoring is obtained; The sleep impact verification data corresponding to several sleep health monitorings of the user is used to analyze the reliability of the impact effects of the sleep positive impact module, sleep negative impact module and sleep fusion impact module in the sleep health monitoring model, and the impact treatment schemes implemented by different impact modules in the sleep health monitoring model are adaptively managed dynamically based on the evaluation results; When only exercise data exists, the user's exercise data for the day is processed and analyzed. If the exercise time period in the exercise data is not within the user's pre-sleep influence time period, a positive influence processing instruction is generated, and according to the positive influence processing instruction, the exercise intensity value and exercise duration in the exercise data are analyzed through the sleep positive influence processing scheme associated with the sleep positive influence module, and the corresponding sleep positive influence processing data is output; If the exercise time period in the exercise data is within the user's pre-sleep impact period, a negative impact processing instruction is generated, and according to the negative impact processing instruction, the exercise intensity value and exercise duration in the exercise data are analyzed by the first sleep negative impact processing scheme associated with the sleep negative impact module, and the corresponding first sleep negative impact processing data is output; When both exercise data and beverage data exist, the exercise intensity value and exercise duration in the exercise data and the drinking time and drinking amount in the beverage data are analyzed through the sleep fusion impact processing solution associated with the sleep fusion impact module to output the corresponding sleep fusion impact processing data.

2. The deep learning model optimization method in sleep health monitoring according to claim 1, characterized in that: When only beverage data exists, the drinking time point and drinking amount in the beverage data are analyzed through the second negative sleep impact processing scheme associated with the negative sleep impact module, and the corresponding second negative sleep impact processing data is output.

3. The deep learning model optimization method in sleep health monitoring according to claim 2, characterized in that: The total sleep time, deep sleep time and light sleep time of the user in the basic sleep quality data are obtained, and the total sleep time, deep sleep time and light sleep time are respectively combined with the sleep impact processing data to obtain the corresponding sleep index reliability value through calculation.

4. The deep learning model optimization method in sleep health monitoring according to claim 3, characterized in that: Performing data analysis on the reliability values ​​of the sleep indicators, and setting the impact identification flag corresponding to the sleep indicators to 0 or 1 according to the analysis results; All sleep indicators are sorted and combined with the impact identification marks obtained by the corresponding processing to obtain a single impact identification processing sequence.

5. The deep learning model optimization method in sleep health monitoring according to claim 4, characterized in that: All elements in a single impact identification processing sequence are summed and analyzed; If the sum of all elements is 0, the total number of completely effective processing of the impact processing scheme corresponding to the sleep impact processing data is increased by one; If the sum of all elements is greater than 0 and less than 3, the total number of effective partial processing of the impact processing scheme corresponding to the sleep impact processing data is increased by one; If the sum of the elements is equal to 3, the total number of completely ineffective treatments of the impact treatment scheme corresponding to the sleep impact treatment data is increased by one; The obtained single impact identification processing sequence and the total number of completely effective processing, the total number of partially effective processing or the total number of completely invalid processing updated by the corresponding impact processing scheme data are sorted and combined to obtain the sleep impact verification data corresponding to the user's single sleep health monitoring.

6. The deep learning model optimization method in sleep health monitoring according to claim 5, characterized in that: According to the sleep impact verification data corresponding to several sleep health monitorings of the user, the total number of completely effective treatments, the total number of partially effective treatments, and the total number of completely invalid treatments associated with the sleep positive impact module, the sleep negative impact module, and the sleep fusion impact module in the sleep health monitoring model are counted respectively, and the impact effect reliability values ​​corresponding to different impact modules are obtained by calculation; Data analysis is performed on the reliability value of the impact effect to determine the sleep impact prediction processing effect of the corresponding impact module.

7. The deep learning model optimization method in sleep health monitoring according to claim 6, characterized in that: If the impact effect reliability value is less than or equal to 1, it is determined that the sleep impact prediction processing effect of the corresponding impact module is normal, and the subsequent implementation of the existing impact processing plan of the corresponding impact module is maintained; Otherwise, it is determined that the sleep impact prediction processing effect of the corresponding impact module is abnormal, and the existing impact processing scheme of the corresponding impact module is dynamically managed.

Images