Sleep health management method and system based on multi-modal physiological data

Abstract

The application discloses a sleep health management method based on multi-modal physiological data, comprising the following steps: acquiring multi-modal physiological signals of a user, extracting feature parameters of the multi-modal physiological signals to obtain multi-modal physiological feature parameters of the user, wherein the multi-modal physiological signals comprise resting heart rate, heartbeat interval, exercise load, stress level, sleep debt, sleep savings and personal information of the user; preprocessing the multi-modal physiological feature parameters of the user to obtain a user physiological signal feature set; calculating the user physiological signal feature set by using a first neural network model to obtain a sleep staging result of the user; performing first-level feature fusion on the sleep staging result of the user and the user physiological signal feature set and inputting the first-level feature fusion result into a first training set to obtain a sleep score of the user; and performing second-level feature fusion on the sleep score of the user, the sleep staging result of the user and the user physiological signal feature set and inputting the second-level feature fusion result into a random forest model to obtain a sleep guidance suggestion of the user.

Description

Technical Field

[0001] This invention relates to the field of digital health and artificial intelligence, specifically to a sleep health management method and system based on multimodal physiological data. Background Technology

[0002] With the increasing popularity of smart wearable devices, users' demand for personalized health management is growing. Sleep, as a key factor affecting human health, has seen its quality assessment become an important function of wearable devices. Current technologies mainly rely on single indicators for simple judgments, such as sleep duration or heart rate variability, which lack comprehensive modeling of complex physiological states.

[0003] Some technologies use CNN or LSTM models to identify sleep stages, but they do not fully consider the influence of non-sleep-related factors, such as stress and exercise load. Other studies use algorithms such as XGBoost for sleep scoring, but lack the ability to capture dynamic changes over time. Furthermore, existing systems typically process information from different sources in isolation, failing to achieve effective fusion of cross-modal features.

[0004] Therefore, there is a need for a technical solution that can integrate multi-dimensional physiological and behavioral data, combine the advantages of deep learning and traditional machine learning, and achieve high-precision, personalized sleep quality assessment. Summary of the Invention

[0005] Based on the above situation, the main objective of this invention is to provide a method and system that integrates multimodal physiological signals and user behavior data, and generates high-precision sleep staging, dynamic sleep quality scores, and personalized intervention suggestions through deep learning and ensemble learning models.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0007] A sleep health management method based on multimodal physiological data includes the following steps:

[0008] S100, acquire the user's multimodal physiological signals, and extract feature parameters from the multimodal physiological signals to obtain the user's multimodal physiological feature parameters. The multimodal physiological signals include the user's resting heart rate, heart rate interval, exercise load, stress level, sleep debt, sleep savings, and personal information.

[0009] S200, preprocess the user's multimodal physiological characteristic parameters to obtain the user's physiological signal feature set;

[0010] S300, the first neural network model is used to calculate the user's physiological signal feature set to obtain the user's sleep stage results, wherein the user's sleep stage results are in minutes as the smallest unit;

[0011] S400, perform first-level feature fusion on the user's sleep stage results and the user's physiological signal feature set, and input it into the first training set to obtain the user's sleep score;

[0012] S500, the user's sleep score, the user's sleep stage results, and the user's physiological signal feature set are fused into secondary features and input into a random forest model to obtain user sleep guidance suggestions.

[0013] Preferably, the feature parameter extraction of the multimodal physiological signal in step S100 includes:

[0014] The resting heart rate was calculated using spectral subtraction based on the user's PPG signal and ACC data.

[0015] The heartbeat interval is calculated based on the peak-to-peak interval of the user's PPG signal;

[0016] The exercise load is calculated using the following formula:

[0017]

[0018] in, TRIMP For exercise load, The duration of the user's stay in each heart rate zone. Weighting coefficients corresponding to each heart rate zone of the user;

[0019] The stress score is calculated using the following formula:

[0020]

[0021] in, Stress Score For pressure level, RMSSD The root mean square of the difference in heartbeat intervals within the heartbeat interval time window, the SDNN The standard deviation of the inter-heartbeat interval within the inter-heartbeat time window is given by the following. LF The low-frequency energy of the user's PPG signal, HF For the high-frequency energy of the user's PPG signal, pNN50 This represents the proportion of heart rate intervals that are greater than a preset value. These are the fitting parameters;

[0022] The sleep debt is calculated using the following formula:

[0023]

[0024] The sleep savings are calculated using the following formula:

[0025]

[0026] in, Debt for sleep Saving for sleep, For ideal daily sleep duration, For the first i The actual sleep duration per day n To review the number of days.

[0027] Preferably, the calculation of the heartbeat interval based on the peak-to-peak interval of the user's PPG signal includes:

[0028] Heartbeat interval values ​​that are less than the first threshold and greater than the second threshold are deleted, and the interval values ​​preceding the deleted heartbeat intervals are used as the basis for further deletion. M The average of the valid heartbeat interval values ​​is supplemented as the deleted heartbeat interval value;

[0029] The moving window Z-score method was used to remove heart rate interval fluctuations, and the values ​​before and after the removed heart rate interval fluctuations were compared. N The weighted values ​​of each value are supplemented as the deleted heartbeat interval fluctuation values.

[0030] Preferably, the moving window Z-score method is used to delete the heartbeat interval fluctuation values, and the values ​​before and after the deleted heartbeat interval fluctuation values ​​are used as the basis for further deletion. N The weighted values ​​supplemented to the deleted heartbeat interval fluctuation values ​​include:

[0031] Before and after each heartbeat interval N A window is formed by several points, and the mean value of the intrabeat interval within the window is calculated. μ and standard deviation σ ;

[0032] like >k σ ,like k If the value is >3, then the heartbeat interval is deleted, and the following formula is used to calculate the intervals before and after the heartbeat interval point. N The weighted values ​​of each value are supplemented to the deleted heartbeat interval values:

[0033]

[0034] Among them, the For the first i Heart rate interval value.

[0035] Preferably, the user's heart rate is divided into intervals based on the user's maximum heart rate, wherein the user's maximum heart rate = 220 - user's age;

[0036] The weighting factor for the 50–60% range of the user's maximum heart rate is 1;

[0037] The weighting factor for the 61–70% range of the user's maximum heart rate is 2;

[0038] The weighting factor for the 71–80% range of the user's maximum heart rate is 3;

[0039] The weighting factor for the 81–90% range of the user's maximum heart rate is 4;

[0040] The weighting factor for the 90–100% range of the user's maximum heart rate is 5.

[0041] Preferably, the preprocessing of the user's multimodal physiological feature parameters in step S200 includes using a hierarchical normalization strategy to normalize the user's multimodal physiological feature parameters, including the following steps:

[0042] S201, the original data of the ordered discrete feature parameters in the user's multimodal physiological feature parameters is retained, and the continuous feature parameters in the user's multimodal physiological feature parameters are normalized using the following formula:

[0043] =

[0044] in, These represent the minimum and maximum values ​​of features in the global training set.

[0045] S202, Calculate the historical mean vector for each user:

[0046]

[0047] The user's historical average resting heart rate. This represents the historical average heart rate interval for users. The user's historical average exercise load. This represents the historical average of user stress levels. The historical average of users' sleep debt. The historical average of user sleep savings;

[0048] The user's multimodal physiological characteristic parameters are updated using the following formula:

[0049]

[0050] in, The normalized mean for all users. For experience weights;

[0051] S203, the user's resting heart rate, heart rate interval, and stress level in the user's multimodal physiological characteristic parameters are assigned one feature value per minute. The user's exercise load, sleep debt, sleep savings, and personal information in the user's multimodal physiological characteristic parameters are expanded to assign one feature value per minute, thus obtaining the user's physiological signal feature set:

[0052] X=

[0053] in, For resting heart rate, For the inter-cardiac period, For exercise load, For pressure level, Debt for sleep For sleep savings, For gender, For age, For height, For weight, For signs of alcoholism, For medication signage and This is a symbol for drinking coffee.

[0054] Preferably, the first neural network model in step S300 includes a convolutional neural network and a long short-term memory neural network model.

[0055] The convolutional neural network includes two cascaded one-dimensional convolutional layers, each with 32 convolutional kernels and a convolutional window length of 7 time units. The user's physiological signal feature set moves by one time unit each time, and max pooling is performed sequentially after each convolutional operation.

[0056] The long short-term memory neural network model adopts a two-layer bidirectional long short-term memory network with 64 hidden units in each layer. The long short-term memory neural network model receives the temporal feature sequence output by the convolutional neural network model and outputs a hidden state vector that integrates historical and future contextual information at each time step.

[0057] The hidden state vector is passed through a fully connected layer and... Softmax The function is mapped to the sleep staging results at the corresponding time step.

[0058] Preferably, the first training set in step S400 is XGBoost, LightGBM, CatBoost, or a random forest model.

[0059] The present invention also discloses a wearable electronic device, which uses the method described in any one of the present invention to manage the user's sleep health.

[0060] The present invention also discloses a sleep health management system based on multimodal physiological data. The system includes an electronic terminal and a cloud server. The electronic terminal and the cloud server are capable of wireless communication. The electronic terminal executes steps S100-S300 of any of the methods described in the present invention, and the cloud server executes steps S400-S500 of any of the methods described in the present invention.

[0061] This invention presents a sleep health management method based on multimodal physiological data. For the first time, it integrates indirect sleep data such as exercise load, stress levels, and sleep history into the evaluation system, enhancing model robustness. A dynamic scoring mechanism introduces concepts such as "stress level," "personal information," "sleep debt," and "sleep savings" to reflect long-term sleep health trends. Personalized guidance generates actionable suggestions based on individual differences, enhancing user compliance. Lightweight deployment allows neural networks, training sets, and random forest models to run efficiently on portable devices (such as smartwatches), making it suitable for real-time applications. This results in more accurate sleep staging, scientific sleep scoring, and the generation of personalized sleep improvement suggestions.

[0062] Other beneficial effects of the present invention will be explained in detail through the introduction of specific technical features and technical solutions in specific embodiments. Those skilled in the art should be able to understand the beneficial technical effects brought about by these technical features and technical solutions through the introduction of these technical features and technical solutions. Attached Figure Description

[0063] The preferred embodiment of the sleep health management method based on multimodal physiological data according to the present invention will be described below with reference to the accompanying drawings. In the drawings:

[0064] Figure 1 This is a flowchart of a sleep health management method based on multimodal physiological data according to a preferred embodiment of the present invention;

[0065] Figure 2 This is a schematic diagram of the workflow of a sleep health management method based on multimodal physiological data according to a preferred embodiment of the present invention. Detailed Implementation

[0066] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of this application. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0067] In the description of this invention, "several" means one or more, "multiple" means two or more, "greater than," "less than," and "exceeding" are understood to exclude the stated number, while "above," "below," and "within" are understood to include the stated number. The use of "first" and "second" in the description is merely for distinguishing technical features and should not be construed as indicating or implying relative importance, or implicitly indicating the number of indicated technical features, or implicitly indicating the order of the indicated technical features.

[0068] In the description of this invention, unless otherwise explicitly defined, terms such as "setup" should be interpreted broadly, and those skilled in the art can reasonably determine the specific meaning of the above terms in this invention in conjunction with the specific content of the technical solution.

[0069] In the description of this invention, the terms "one embodiment," "some embodiments," "illustrative embodiment," "example," "specific example," or "some examples," etc., refer to specific features or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0070] Figure 1 The flowchart of a sleep health management method based on multimodal physiological data according to a preferred embodiment of the present invention includes the following steps: S100, acquiring the user's multimodal physiological signals and extracting feature parameters from the multimodal physiological signals to obtain user multimodal physiological feature parameters, wherein the multimodal physiological signals include the user's resting heart rate, heart rate interval (BBI), exercise load, stress level, sleep debt, sleep savings, and personal information; S200, preprocessing the user's multimodal physiological feature parameters to obtain a user physiological signal feature set; S300, using a first neural network model to calculate the user's physiological signal feature set to obtain a user sleep staging result, wherein the user sleep staging result is in minutes as the smallest unit; S400, performing first-level feature fusion of the user sleep staging result and the user physiological signal feature set, and inputting it into a first training set to obtain a user sleep score; S500, performing second-level feature fusion of the user sleep score, the user sleep staging result, and the user physiological signal feature set, and inputting it into a random forest model to obtain user sleep guidance suggestions.

[0071] In specific implementations, green light, infrared light, red light PPG signals, ambient light, and behavioral data can be used as the user's multimodal physiological signals. Among them, behavioral data includes, but is not limited to, three-axis ACC data from an accelerometer, namely, three-axis x, y, and z-axis data.

[0072] This invention presents a sleep health management method based on multimodal physiological data. For the first time, it integrates indirect sleep data such as exercise load, stress levels, and sleep history into the evaluation system, enhancing model robustness. A dynamic scoring mechanism introduces concepts such as "stress level," "personal information," "sleep debt," and "sleep savings" to reflect long-term sleep health trends. Personalized guidance generates actionable suggestions based on individual differences, enhancing user compliance. Lightweight deployment allows neural networks, training sets, and random forest models to run efficiently on portable devices (such as smartwatches), making it suitable for real-time applications. This results in more accurate sleep staging, scientific sleep scoring, and the generation of personalized sleep improvement suggestions.

[0073] In a preferred embodiment, the feature parameter extraction of the multimodal physiological signal in step S100 may include:

[0074] The resting heart rate is calculated using spectral subtraction based on the user's PPG signal and ACC data.

[0075] The heartbeat interval is calculated based on the peak-to-peak interval of the user's PPG signal;

[0076] The exercise load is calculated using the following formula. TRIMP :

[0077]

[0078] in, The duration of the user's stay in each heart rate zone. Weighting coefficients corresponding to each heart rate zone of the user;

[0079] The stress score is calculated using the following formula:

[0080]

[0081] in, RMSSD This is the root mean square of the difference in heart rate intervals within the heart rate interval time window. SDNN The standard deviation of the inter-heartbeat interval within the time window. LF For the low-frequency energy of the user's PPG signal, HF For the high-frequency energy of the user's PPG signal, pNN50 This represents the proportion of heart rate intervals greater than a preset value, which can be set to 50ms. These are the fitting parameters;

[0082] Sleep debt is calculated using the following formula:

[0083]

[0084] Sleep savings are calculated using the following formula:

[0085]

[0086] in, Debt for sleep Saving for sleep, For ideal daily sleep duration, For the first i The actual sleep duration per day n To review the number of days.

[0087] Resting heart rate, also known as sleep heart rate, refers to the heart rate during the period from when a user falls asleep to when they wake up. The output range is 30-210 Bpm, with an output frequency of 1 Hz, denoted as HR. PPG (Photoplethysmography) is a photoplethysmography technique primarily used to monitor changes in blood flow, commonly found in heart rate and blood oxygen monitoring. PPG works by illuminating the skin with light and using photoelectric sensors to detect changes in transmitted or reflected light caused by blood flow. As blood volume fluctuates with the heartbeat, the amount of light absorbed changes synchronously, thus capturing the pulse wave signal for analysis of cardiovascular parameters.

[0088] The beat-to-beat interval (BBI) is the period between the peaks of the PPG signal that represents one heartbeat cycle.

[0089] Exercise load is an indicator that measures the physiological burden an individual bears during exercise or physical activity. The calculation period is from the time an individual wakes up to the time they fall asleep, excluding short naps, such as afternoon naps of less than 3 hours. In other implementations, exercise load can be replaced by metabolic equivalents (METs).

[0090] Sleep debt refers to the long-term accumulated amount of sleep deprivation, that is, the actual sleep time obtained is consistently less than the amount of sleep required by the individual's physiology. Sleep savings is a concept opposite to sleep debt, referring to the portion of sleep time obtained by an individual that exceeds their physiological needs. It is generally considered to have a buffering or restorative effect, which can partially offset the negative impact of subsequent sleep deprivation. It is only counted as "savings" when the actual sleep exceeds the need.

[0091] When extracting the heartbeat interval based on photoplethysmography (PPG) signals, the original BBI sequence often contains outliers, such as intervals that are too short or too long, due to motion artifacts, poor signal quality, baseline drift, or peak detection errors.

[0092] In a preferred embodiment, calculating the heartbeat interval based on the peak-to-peak interval of the user's PPG signal may include: deleting heartbeat interval values ​​that are less than a first threshold and greater than a second threshold, and using the deleted heartbeat interval as the preceding value. M The average of the valid heartbeat interval values ​​is used to supplement the deleted heartbeat interval values; the moving window Z-score method is used to delete heartbeat interval fluctuation values, and the values ​​before and after the deleted heartbeat interval fluctuation values ​​are used as the basis for comparison. N The weighted sum of these values ​​is used to supplement the deleted heartbeat interval fluctuation value. In a specific implementation, the first threshold for the heartbeat interval can be 300ms, and the second threshold for the heartbeat interval can be 1500ms. M It can be 5. N It can also be set to 5. By setting these two thresholds, out-of-range BBI values ​​are removed and the average of the top 5 valid values ​​of that value is used to supplement the results.

[0093] In a preferred embodiment, the moving window Z-score method is used to delete the heart rate interval fluctuation values, and the values ​​before and after the deleted heart rate interval fluctuation values ​​are used as the basis for further deletion. N The weighted sum of these values ​​to supplement the deleted heartbeat interval fluctuation values ​​may include:

[0094] Before and after each heartbeat interval N A window is formed by several points, and the mean value of the intrabeat interval within the window is calculated. μ and standard deviation σ ;

[0095] like >k σ, if k>3, then delete the heartbeat interval, and use the following formula to calculate the interval before and after the heartbeat interval point. N The weighted values ​​of each value are supplemented to the deleted heartbeat interval values:

[0096]

[0097] Among them, the For the first i Heart rate interval value.

[0098] In a specific implementation method N The value can be 5, meaning that for each BBI point, the mean BBI within the window formed by the five points before and after it is calculated. μ and standard deviation σ The output frequency of the BBI during the intercardiac period can be 1 Hz.

[0099] In a preferred embodiment, the user's heart rate can be divided into zones based on the user's maximum heart rate (HRmax), where HRmax = 220 - user age, and a weighting coefficient is applied to the 50-60% range of the user's maximum heart rate. A weight of 1 represents the weighting factor for the 61-70% range of the user's maximum heart rate. The weighting factor is 2, representing the 71-80% range of the user's maximum heart rate. A weighting factor of 3 for the 81–90% range of the user's maximum heart rate. The weighting factor is 4, representing the 90-100% range of the user's maximum heart rate. It is 5.

[0100] In a preferred embodiment, when calculating the pressure level, the BBI time window length is 5 minutes, the sliding step size is 1 minute, and the overlapping data is 4 minutes to achieve the calculation, with an output frequency of 1 Hz. The RMSSD value range can be 27 ± 13 ms, the SDNN value range can be 141 ± 40 ms, LF is the low-frequency part of the PPG signal, using a bandpass of 0.15~0.4 Hz. Alternatively, the BBI interval sequence within the BBI time window can be extended, and then bandpass signals of [0.04, 0.15] Hz and [0.15, 0.4] Hz can be extracted. The low-frequency energy LF and high-frequency energy HF are obtained by calculating the sum of squares of the amplitudes of each passband signal, and the ratio of LF to HF, LF / HF, is calculated, with a value range of 1.4~2.1. The fitting parameters can be set to the following values: 0.1774, -0.1172, 0.8154, -0.0919, 98.2624, or other values. This invention does not limit these values.

[0101] In other implementations, the pressure level may be replaced by a single indicator such as RMSSD or LF / HF, or other forms of fusion.

[0102] In specific implementations, an individual's ideal daily sleep requirement is typically 7–9 hours for adults, which can be personalized through the personal information section. (Review of days) n It can usually be set to 3 days, 7 days or 14 days.

[0103] In a preferred embodiment, personal information may include gender (e.g., male is marked as 0, female as 1), age, height, weight, history of snoring (0 for no, 1 for yes), history of alcohol abuse (0 for no, 1 for yes), history of medication use (0 for no, 1 for yes), and history of drinking coffee before bed (0 for no, 1 for yes). The features used in this scheme, namely gender, age, height, weight, history of snoring, history of alcohol abuse, history of medication use, and history of drinking coffee before bed, can be represented as:

[0104] F1=

[0105] The extraction of user multimodal physiological feature parameters is now complete. To eliminate the dimensional differences between different physiological and behavioral indicators and ensure the stability of the subsequent deep learning model, in a preferred embodiment, the preprocessing of the user multimodal physiological feature parameters in step S200 can employ a hierarchical normalization strategy to normalize the user multimodal physiological feature parameters, including:

[0106] Step S201: For the ordered discrete feature parameters in the user's multimodal physiological feature parameters, retain the original data; for the continuous feature parameters in the user's multimodal physiological feature parameters, normalize them using the following formula:

[0107] =

[0108] in, These are the minimum and maximum values ​​of the features in the global training set; the global training set refers to the sample data of all test users.

[0109] To address baseline differences among users, such as user A having an average heart rate of 55 bpm and user B having an average heart rate of 75 bpm, user-centered offset compensation is introduced. Step S202 is used to calculate the historical mean vector for each user.

[0110]

[0111] The user's historical average resting heart rate. This represents the historical average heart rate interval for users. The user's historical average exercise load. This represents the historical average of user stress levels. The historical average of users' sleep debt. The historical average of user sleep savings;

[0112] The user's multimodal physiological characteristic parameters are updated using the following formula:

[0113]

[0114] in, The normalized mean for all users. For experience weights;

[0115] S203, the user's resting heart rate, heart rate interval, and stress level in the user's multimodal physiological characteristic parameters are assigned one feature value per minute. The user's exercise load, sleep debt, sleep savings, and personal information in the user's multimodal physiological characteristic parameters are expanded to assign one feature value per minute, thus obtaining the user's physiological signal feature set:

[0116] X=

[0117] in, For resting heart rate, For the inter-cardiac period, For exercise load, For pressure level, Debt for sleep For sleep savings, For gender, For age, For height, For weight, For signs of alcoholism, For medication signage and This is a symbol for drinking coffee.

[0118] In specific implementations, ordered discrete indicators such as gender, alcohol consumption, smoking, medication use, and coffee consumption before bed, all have values ​​ranging from 0 to 1, thus preserving the original data. Continuous indicators such as resting heart rate (HR), heart rate interval (BBI), exercise load (TRIMP), stress score, sleep debt, sleep savings, height, and weight can be normalized using Min-Max normalization based on global statistics to avoid scaling distortion caused by extreme values ​​from a single user.

[0119] Specifically, to ensure consistency between the normalized parameters and inference, during the training phase... , Saved as source data for the model. During inference: New user data is normalized using the same global parameters to ensure a consistent model input distribution and avoid prediction drift caused by online parameter updates. The user physiological signal feature set can also be filtered based on the feature's influence weight coefficients to reduce model complexity.

[0120] In a preferred embodiment, the first neural network model in step S300 includes a Convolutional Neural Network (CNN) and a Long Short-Term Memory (LSTM) neural network. The CNN comprises two cascaded one-dimensional convolutional layers, each with 32 kernels and a convolutional window length of 7 time units. The user's physiological signal feature set moves one time unit at a time. Max pooling is performed sequentially after each convolutional operation to enhance the model's robustness to individual physiological differences and sensor noise. The LSTM uses a two-layer bidirectional long short-term memory network (BiLSTM), with 64 hidden units per layer. The LSTM receives the temporal feature sequence output by the CNN and outputs a hidden state vector that fuses historical and future contextual information at each time step. The hidden state vector is then passed through a fully connected layer and... Softmax The function maps to the sleep staging results corresponding to the time step. In a specific implementation, one time unit can refer to the window length of the BBI signal.

[0121] The user's physiological signal feature set is used as the input of the CNN+LSTM model. The output of the CNN+LSTM training set is the calibrated sleep staging results: 5-category sleep staging results: Wake, N1, N2, N3, REM. Each staging result is in minutes as the smallest unit, realizing minute-level sleep staging.

[0122] In other implementations, a Transformer neural network or a Temporal Convolutional Network (TCN) can also be used.

[0123] In a preferred embodiment, the first training set in step S400 can be an XGBoost, LightGBM, CatBoost, or a random forest model. The sleep staging result output in step S300 is fused with the user's physiological signal feature set using the first-level feature fusion (S304) as input to the XGBoost training set. The output is a calibrated sleep score, a standardized sleep score from 0 to 100. First-level feature fusion refers to adding the sleep staging result to the user's physiological signal feature set, thus adding a sleep staging dimension to the user's physiological signal feature set.

[0124] In step S500, the sleep stage, sleep score, and user physiological signal feature set are fused into secondary features and used as input to the random forest model. At the same time, the sleep score is expanded with the original value and aligned with other features. The output of the trained model is the calibrated sleep guidance suggestion.

[0125] In specific implementations, the sleep guidance recommendations include, but are not limited to:

[0126] If you have a short-term sleep debt of 1-2 days, it is recommended that if you have accumulated 2.5 hours of sleep deprivation in the past two days, you can add 30-60 minutes of sleep tonight, or arrange a 20-minute nap, and avoid exceeding 30 minutes to prevent affecting your nighttime sleep;

[0127] If you have enough sleep reserves, you can go to bed a little later tonight, but do not exceed one hour more than usual;

[0128] If you have been experiencing a prolonged period of high exercise load and high stress, it is recommended that you schedule active recovery activities for tomorrow, such as walking or stretching, and ensure that you get at least 9 hours of sleep.

[0129] If alcohol abuse is combined with snoring, it is suggested that smoking and snoring are risk factors for sleep apnea.

[0130] If you frequently wake up feeling suffocated or have a headache upon waking, it is recommended to consult a doctor for screening.

[0131] like Figure 2 As shown, after extracting and preprocessing the user's physiological signal feature set x, the user's physiological signal feature set x is first input into the CNN+LSTM neural network model to calculate the user's sleep stage results. Then, the user's sleep stage results and the user's physiological signal feature set x are fused together at the first level and input into the Xgboost training set to calculate the user's sleep score. Finally, the user's sleep stage results, the user's sleep score and the user's physiological signal feature set x are fused together at the second level and input into the random forest model to calculate the user's sleep guidance suggestions.

[0132] This invention also discloses a wearable electronic device that uses the method described in any one of the present invention to manage a user's sleep health. Examples include smartwatches, bracelets, rings, and other wearable devices.

[0133] This invention also discloses a sleep health management system based on multimodal physiological data. The system includes an electronic terminal and a cloud server, wherein the electronic terminal and the cloud server are capable of wireless communication.

[0134] The electronic terminal executes steps S100-S300 of any one of the methods described in this invention, and the cloud server executes steps S400-S500 of any one of the methods described in this invention. That is, sleep staging is deployed on the device side, while the sleep scoring and sleep guidance models are deployed on the cloud side. This combines sleep scores, staging details, and long-term user behavior patterns to generate actionable, causally traceable improvement suggestions and personalized sleep health management plans.

[0135] Specifically, based on the embodiments disclosed above in this application, reference is made to... Figure 2The processes described in the flowcharts of the embodiments can be implemented as computer programs. For example, embodiments of this application also provide a computer program product including a computer program carried on a computer-readable medium, the computer program comprising methods for executing... Figure 1 The flowchart of the embodiment describes the program code for the method. In such an embodiment, the computer program can be downloaded and installed from a network via a communication interface, or installed from memory. When the computer program is executed by a processor, it performs the functions defined in the method of the above embodiment.

[0136] It should be noted that the aforementioned computer-readable media may include, but is not limited to: volatile memory, such as random access memory (RAM); non-volatile memory, such as read-only memory (ROM), flash memory, hard disk drive (HDD), or solid-state drive (SSD); and combinations of the above types of memory.

[0137] In this application, a computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0138] The aforementioned computer-readable medium may be included in the aforementioned electronic device; or it may exist independently and not assembled into the electronic device.

[0139] In an optional embodiment, this application also provides a computer storage medium that can be used for computer software instructions, including a computer program, which, when executed by a processor, performs the presentation whiteboard display method as described in the above embodiments. This storage medium includes, but is not limited to, flash memory, hard disk, and solid-state drive.

[0140] It should be noted that the use of step numbers (letter or number) to refer to certain specific method steps in this invention is merely for the purpose of convenience and brevity in description, and is by no means intended to restrict the order of these method steps. Those skilled in the art will understand that the order of the relevant method steps should be determined by the technology itself and should not be unduly restricted by the existence of step numbers.

[0141] Those skilled in the art will understand that, without conflict, the above-mentioned preferred solutions can be freely combined and superimposed.

[0142] It should be understood that the above embodiments are merely exemplary and not restrictive. Various obvious or equivalent modifications or substitutions that can be made by those skilled in the art regarding the above details without departing from the basic principles of the present invention will be included within the scope of the claims of the present invention.

Claims

1. A sleep health management method based on multimodal physiological data, characterized in that, Including the following steps: S100, acquire the user's multimodal physiological signals, and extract feature parameters from the multimodal physiological signals to obtain the user's multimodal physiological feature parameters. The multimodal physiological signals include the user's resting heart rate, heart rate interval, exercise load, stress level, sleep debt, sleep savings, and personal information. The feature parameter extraction from the multimodal physiological signals includes: The resting heart rate was calculated using spectral subtraction based on the user's PPG signal and ACC data. The heartbeat interval is calculated based on the peak-to-peak interval of the user's PPG signal; The exercise load is calculated using the following formula: in, TRIMP For exercise load, The duration of the user's stay in each heart rate zone. Weighting coefficients corresponding to each heart rate zone of the user; The pressure level is calculated using the following formula: in, Stress Score For pressure level, RMSSD The root mean square of the difference in heartbeat intervals within the heartbeat interval time window, the SDNN The standard deviation of the inter-heartbeat interval within the inter-heartbeat time window is given by the following. LF The low-frequency energy of the user's PPG signal, HF For the high-frequency energy of the user's PPG signal, pNN50 This represents the proportion of heart rate intervals that are greater than a preset value. These are the fitting parameters; The sleep debt is calculated using the following formula: The sleep savings are calculated using the following formula: in, Debt for sleep Save for sleep For ideal daily sleep duration, For the first i The actual sleep duration per day n To review the number of days; S200, preprocess the user's multimodal physiological feature parameters to obtain a user physiological signal feature set. The preprocessing of the user's multimodal physiological feature parameters includes using a hierarchical normalization strategy to normalize the user's multimodal physiological feature parameters, comprising the following steps: S201, the original data of the ordered discrete feature parameters in the user's multimodal physiological feature parameters is retained, and the continuous feature parameters in the user's multimodal physiological feature parameters are normalized using the following formula: = in, These represent the minimum and maximum values ​​of features in the global training set. S202, Calculate the historical mean vector for each user: The user's historical average resting heart rate. This represents the historical average heart rate interval for users. The user's historical average exercise load. This represents the historical average of user stress levels. The historical average of users' sleep debt. The historical average of user sleep savings; The user's multimodal physiological characteristic parameters are updated using the following formula: in, The normalized mean for all users. For experience weights; S203, the user's resting heart rate, heart rate interval, and stress level in the user's multimodal physiological characteristic parameters are assigned one feature value per minute. The user's exercise load, sleep debt, sleep savings, and personal information in the user's multimodal physiological characteristic parameters are expanded to assign one feature value per minute, thus obtaining the user's physiological signal feature set: X= in, For resting heart rate, For the inter-cardiac period, For exercise load, For pressure level, Debt for sleep For sleep savings, For gender, For age, For height, For weight, For signs of alcoholism, For medication signage and This is a symbol for drinking coffee; S300, the first neural network model is used to calculate the user's physiological signal feature set to obtain the user's sleep stage results, wherein the user's sleep stage results are in minutes as the smallest unit; S400, perform first-level feature fusion on the user's sleep stage results and the user's physiological signal feature set, and input it into the first training set to obtain the user's sleep score; S500, the user's sleep score, the user's sleep stage results, and the user's physiological signal feature set are fused into secondary features and input into a random forest model to obtain user sleep guidance suggestions.

2. The sleep health management method based on multimodal physiological data according to claim 1, characterized in that, The heartbeat interval is calculated based on the peak-to-peak interval of the user's PPG signal, including: Heartbeat intervals that are less than the first threshold and greater than the second threshold are deleted, and the intervals preceding the deleted heartbeat intervals are used as the basis for the deletion. M The average of the effective heartbeat intervals is supplemented with the deleted heartbeat intervals; The moving window Z-score method was used to remove heart rate interval fluctuations, and the values ​​before and after the removed heart rate interval fluctuations were compared. N The weighted values ​​of each individual value are supplemented by the deleted heartbeat interval fluctuation values.

3. The sleep health management method based on multimodal physiological data according to claim 2, characterized in that, The moving window Z-score method was used to remove heart rate interval fluctuations, and the values ​​before and after the removed heart rate interval fluctuations were compared. N The weighted values ​​supplemented by the deleted heartbeat interval fluctuation values ​​include: Before and after each heartbeat interval N A window is formed by several points, and the mean value of the intrabeat interval within the window is calculated. μ and standard deviation σ ; like >k*σ ,like k If the value is >3, then the heartbeat interval is deleted, and the following formula is used to calculate the intervals before and after the heartbeat interval point. N The weighted values ​​of each value are used to supplement the deleted heartbeat interval values: Among them, the For the first i A period between heartbeats.

4. The sleep health management method based on multimodal physiological data according to claim 1, characterized in that, The user's heart rate is divided into zones based on the user's maximum heart rate, where the maximum heart rate = 220 - the user's age; The weighting factor for the 50–60% range of the user's maximum heart rate is 1; The weighting factor for the 61–70% range of the user's maximum heart rate is 2; The weighting factor for the 71–80% range of the user's maximum heart rate is 3; The weighting factor for the 81–90% range of the user's maximum heart rate is 4; The weighting factor for the 90–100% range of the user's maximum heart rate is 5.

5. The sleep health management method based on multimodal physiological data according to claim 1, characterized in that, The first neural network model in step S300 includes a convolutional neural network and a long short-term memory neural network model. The convolutional neural network includes two cascaded one-dimensional convolutional layers, each with 32 convolutional kernels and a convolutional window length of 7 time units. The user's physiological signal feature set moves by one time unit each time, and max pooling is performed sequentially after each convolutional operation. The long short-term memory neural network model adopts a two-layer bidirectional long short-term memory network with 64 hidden units in each layer. The long short-term memory neural network model receives the temporal feature sequence output by the convolutional neural network model and outputs a hidden state vector that integrates historical and future contextual information at each time step. The hidden state vector is passed through a fully connected layer and... Softmax The function is mapped to the sleep staging results at the corresponding time step.

6. The sleep health management method based on multimodal physiological data according to claim 1, characterized in that, The first training set in step S400 is XGBoost, LightGBM, CatBoost, or a random forest model.

7. A wearable electronic device, characterized in that, The electronic device uses the method described in any one of claims 1-6 to manage the user's sleep health.

8. A sleep health management system based on multimodal physiological data, the system comprising an electronic terminal and a cloud server, wherein the electronic terminal and the cloud server are capable of wireless communication, characterized in that... The electronic terminal performs steps S100-S300 in the method as described in any one of claims 1-6, and the cloud server performs steps S400-S500 in the method as described in any one of claims 1-6.

Images