Multidimensional assessment methods and devices for sleep quality, storage media, and computer equipment

Abstract

This application relates to the field of data processing technology, and discloses a multi-dimensional assessment method, device, storage medium, and computer equipment for sleep quality. The method includes: continuously and synchronously collecting the user's physiological and environmental data; extracting time-frequency domain features from the continuously collected physiological and environmental data according to preset time intervals; fusing the extracted results to obtain a fused feature vector for each time interval; identifying the user's sleep stage within the corresponding time interval based on the fused feature vector; dynamically calculating the change in the user's sleep debt within the corresponding time interval based on the sleep stage and corresponding physiological data whenever a new sleep stage is identified, and updating the current sleep debt accordingly; and dynamically determining the user's sleep quality assessment result based on the updated current sleep debt. This application can be applied to sleep quality assessment scenarios in the field of smart healthcare, improving the accuracy and real-time performance of sleep quality assessment.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a method and apparatus for multi-dimensional assessment of sleep quality, a storage medium, and a computer device. Background Technology

[0002] With the increasing awareness of health management and the rapid development of smart healthcare technology, objective and accurate assessment of sleep quality has become an important part of personal health monitoring and disease prevention. Currently, with the help of smart wearable devices and biosensors, it is possible to continuously collect single or a few physiological indicators such as heart rate and body movement, and combine this with mobile applications to provide users with basic sleep duration and structure analysis, which constitutes the initial technological foundation for sleep health management.

[0003] However, as smart healthcare moves towards personalization and precision, existing sleep assessment technologies still have significant limitations. Most solutions rely on data from a single type of sensor, offering only a limited assessment dimension and failing to comprehensively reflect the complex physiological state of sleep. Their analytical models often employ static, uniform threshold standards, which cannot adapt to individual differences, resulting in insufficient correlation between assessment results and users' subjective feelings. More importantly, existing technologies typically only provide post-hoc statistics, lacking quantitative modeling of the dynamic accumulation and repayment process of sleep stress, thus failing to achieve proactive sleep health risk warnings and personalized intervention guidance. Summary of the Invention

[0004] In view of this, this application provides a multi-dimensional assessment method, device, storage medium, and computer equipment for sleep quality. By simultaneously collecting and fusing multi-dimensional physiological and environmental data, it overcomes the limitations of single-sensor data, providing a more comprehensive and integrated reflection of the complex physiological state of sleep and its interaction with the environment, thus providing a more accurate basis for subsequent assessments. Simultaneously, by dynamically identifying sleep stages according to time intervals and calculating changes in sleep debt in real time, a personalized and dynamically evolving quantitative model of sleep debt is established. This abandons static, uniform thresholds, allowing assessment results to adapt to individual differences and greatly improving the correlation between assessment results and users' actual feelings. Furthermore, this method achieves a continuous and quantitative description of the accumulation and repayment process of sleep pressure, enabling assessments to move beyond post-hoc statistics and reflect the surplus and deficit status of sleep physiological load in real time. This provides a reliable and quantitative decision-making basis for proactive sleep health risk warnings and subsequent personalized intervention guidance.

[0005] According to one aspect of this application, a multi-dimensional assessment method for sleep quality is provided, comprising: Continuously and synchronously collect users' physiological and environmental data; The continuously collected physiological and environmental data are subjected to time-frequency domain feature extraction according to a preset time interval. The extraction results are then fused to obtain a fused feature vector representing the user's sleep state in each time interval. Based on the fused feature vector of each time interval, a neural network model is used to identify the user's sleep stage in the corresponding time interval. Whenever a new sleep stage in a new time interval is identified, the change in the user's sleep debt in the corresponding time interval is dynamically calculated based on the sleep stage and the physiological data within the corresponding time interval. The user's current sleep debt is then updated based on the change in sleep debt. If the sleep stage indicates that the user is awake, the user's new sleep pressure is calculated as the change in sleep debt. If the sleep stage indicates that the user is asleep, the user's sleep repayment value is calculated as the change in sleep debt. The user's sleep quality assessment result is dynamically determined based on the updated current sleep debt.

[0006] According to another aspect of this application, a multi-dimensional sleep quality assessment device is provided, comprising: The data acquisition module is used to continuously and synchronously collect the user's physiological and environmental data; The sleep stage identification module is used to extract time-frequency domain features from the continuously collected physiological data and environmental data according to a preset time interval, fuse the extraction results to obtain a fused feature vector representing the user's sleep state in each time interval, and identify the user's sleep stage in the corresponding time interval through a neural network model based on the fused feature vector of each time interval. The sleep debt update module is used to dynamically calculate the change in the user's sleep debt in the corresponding time interval based on the sleep stage and the physiological data in the corresponding time interval whenever a new sleep stage is identified, and update the user's current sleep debt according to the change in sleep debt. If the sleep stage indicates that the user is awake, the user's new sleep pressure is calculated as the change in sleep debt; if the sleep stage indicates that the user is asleep, the user's sleep repayment value is calculated as the change in sleep debt. The sleep quality assessment module is used to dynamically determine the user's sleep quality assessment result based on the updated current sleep debt.

[0007] According to another aspect of this application, a storage medium is provided that stores a computer program thereon, which, when executed by a processor, implements the above-described multi-dimensional assessment method for sleep quality.

[0008] According to another aspect of this application, a computer device is provided, including a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, wherein the processor executes the program to implement the above-described multi-dimensional sleep quality assessment method.

[0009] Using the above technical solution, this application provides a multi-dimensional sleep quality assessment method, device, storage medium, and computer equipment. First, it continuously and synchronously collects the user's physiological and environmental data. Next, the continuously collected physiological and environmental data are segmented according to preset time intervals. Within each time interval, representative time-frequency domain features are extracted from the physiological and environmental data, and then these features are fused into a unified fusion feature vector using a specific algorithm. This fusion feature vector is then input into a trained neural network model, which automatically identifies and outputs the user's specific sleep stage within that time interval. Whenever a new sleep stage is identified in a new time interval, the change in the user's sleep debt within that time interval can be calculated based on that sleep stage and the physiological data within that time interval. By continuously accumulating the change in sleep debt generated in each time interval, dynamic and real-time updates to the user's current sleep debt level are achieved. Finally, based on the dynamically updated current sleep debt, a quantitative sleep quality assessment result is output. This application's embodiments overcome the limitations of single-sensor data by simultaneously collecting and fusing multi-dimensional physiological and environmental data. This allows for a more comprehensive and integrated reflection of the complex physiological state of sleep and its interaction with the environment, providing a more accurate basis for subsequent assessments. Simultaneously, by dynamically identifying sleep stages according to time intervals and calculating changes in sleep debt in real time, a personalized, dynamically evolving sleep debt quantification model is established. This abandons static, uniform thresholds, enabling assessment results to adapt to individual differences and significantly improving the correlation between assessment results and users' actual experiences. Furthermore, this method achieves a continuous and quantitative description of the accumulation and repayment process of sleep pressure, allowing assessments to move beyond post-hoc statistics and reflect the surplus and deficit status of sleep physiological load in real time. This provides a reliable and quantitative decision-making basis for proactive sleep health risk warnings and subsequent personalized intervention guidance.

[0010] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0011] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 A flowchart illustrating a multi-dimensional assessment method for sleep quality provided in an embodiment of this application is shown. Figure 2 This illustration shows a schematic diagram of a recurrent neural network provided in an embodiment of this application; Figure 3 This illustration shows a structural schematic diagram of a multi-dimensional sleep quality assessment system provided in an embodiment of this application; Figure 4 This illustration shows a structural schematic diagram of a multi-dimensional sleep quality assessment device provided in an embodiment of this application; Figure 5 A schematic diagram of the device structure of a computer device provided in an embodiment of this application is shown. Detailed Implementation

[0012] The present application will be described in detail below with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in the embodiments of the present application can be combined with each other.

[0013] This embodiment provides a multi-dimensional assessment method for sleep quality, such as... Figure 1 As shown, the method includes: Step 101: Continuously and synchronously collect the user's physiological data and environmental data.

[0014] Step 102: Extract time-frequency domain features from the continuously collected physiological data and environmental data according to a preset time interval, fuse the extraction results to obtain a fused feature vector representing the user's sleep state in each time interval, and identify the user's sleep stage in the corresponding time interval through a neural network model based on the fused feature vector of each time interval.

[0015] Step 103: Whenever a new sleep stage in a new time interval is identified, the change in the user's sleep debt in the corresponding time interval is dynamically calculated based on the sleep stage and the physiological data within the corresponding time interval. The user's current sleep debt is then updated based on the change in sleep debt. If the sleep stage indicates that the user is awake, the user's new sleep pressure is calculated as the change in sleep debt. If the sleep stage indicates that the user is asleep, the user's sleep repayment value is calculated as the change in sleep debt.

[0016] Step 104: Dynamically determine the user's sleep quality assessment result based on the updated current sleep debt.

[0017] This application provides a multi-dimensional assessment method for sleep quality. First, it can continuously and synchronously collect users' physiological data (such as electrocardiogram, respiration, and body movement) and environmental data (such as noise, temperature, and humidity) through smart wearable devices (such as wristbands) and environmental sensors, laying a solid data foundation for subsequent in-depth analysis.

[0018] Next, the continuously collected physiological and environmental data are segmented according to preset time intervals (e.g., 30 seconds). Here, a fusion feature vector generation operation is performed every time a preset time interval is reached (i.e., every 30 seconds of physiological and environmental data is collected). Specifically, within each time interval, representative time-frequency domain features are extracted from the physiological and environmental data. These features are then fused into a unified fusion feature vector using a specific algorithm. This fusion feature vector can be understood as a digital fingerprint of the user's sleep state within that time interval. This fusion feature vector is then input into a trained neural network model (e.g., a recurrent neural network containing long short-term memory units). The neural network model automatically identifies and outputs the user's sleep stage within that time interval, such as the wakefulness stage or the sleep onset stage.

[0019] Next, we move to the core quantitative assessment stage. Each time a new sleep stage within a new time interval is identified, the change in the user's sleep debt during that time interval can be calculated based on that sleep stage and the physiological data within that time interval. Specifically, if the user is awake during that time interval, the body accumulates sleep pressure, thus increasing the current sleep debt; if the user is asleep during that time interval, the body repays the debt, thus decreasing the current sleep debt. By continuously accumulating the change in sleep debt generated in each time interval, a dynamic and real-time update of the user's current sleep debt level is achieved, thereby accurately quantifying their sleep balance.

[0020] Ultimately, the dynamically updated current sleep debt is used as the core objective indicator for assessing sleep quality. The level of current sleep debt directly and comprehensively reflects the physiological recovery efficiency and sufficiency of sleep. Based on this, quantitative sleep quality assessment results can be output.

[0021] In one specific embodiment, after obtaining the sleep quality assessment results, improvement suggestions can be further generated. Specifically, if the sleep quality assessment results indicate that the current sleep debt level exceeds a high threshold, a high-load state prompt is triggered, and high-priority health suggestions are generated; if the sleep quality assessment results indicate that the current sleep debt level is at a moderate level, regular optimization suggestions are generated; if the sleep quality assessment results indicate that the current sleep debt level is low, it is recommended to maintain the current sleep pattern.

[0022] By applying the technical solution of this embodiment, firstly, the user's physiological and environmental data are continuously and synchronously collected. Then, the continuously collected physiological and environmental data are segmented according to preset time intervals. Within each time interval, representative time-frequency domain features are extracted from the physiological and environmental data, and then these features are fused into a unified fusion feature vector using a specific algorithm. This fusion feature vector is then input into a trained neural network model, which automatically identifies and outputs the specific sleep stage the user is in within that time interval. Whenever a new sleep stage in a new time interval is identified, the change in the user's sleep debt within that time interval can be calculated based on that sleep stage and the physiological data within that time interval. By continuously accumulating the change in sleep debt generated in each time interval, dynamic and real-time updates to the user's current sleep debt level are achieved. Finally, based on the dynamically updated current sleep debt, a quantitative sleep quality assessment result is output. This embodiment, by synchronously collecting and fusing multi-dimensional physiological and environmental data, overcomes the limitations of single-sensor data, and can more comprehensively and holistically reflect the complex physiological state of sleep and its interaction with the environment, providing a more accurate basis for subsequent assessments. Meanwhile, by dynamically identifying sleep stages according to time intervals and calculating changes in sleep debt in real time, a personalized and dynamically evolving quantitative model of sleep debt was established. This model abandons static uniform thresholds, allowing assessment results to adapt to individual differences and greatly improving the correlation between assessment results and users' actual feelings. Furthermore, this method achieves a continuous and quantitative description of the accumulation and repayment process of sleep pressure, enabling assessments to move beyond post-hoc statistics and reflect the surplus and deficit status of sleep physiological load in real time. This provides a reliable and quantitative decision-making basis for proactive sleep health risk warnings and subsequent personalized intervention guidance.

[0023] In this embodiment, optionally, the physiological data includes user age, heart rate variability data, and exercise data; the user's new sleep pressure is calculated based on the following method: personalized parameters characterizing the rate of sleep pressure accumulation during the user's wakefulness are extracted from the physiological data within the corresponding time interval, wherein the personalized parameters include an age factor determined based on the user's age, an autonomic nervous system function index determined based on the heart rate variability data, an activity load calculated based on the exercise data, and a circadian rhythm shift determined based on the historical sleep stage sequence; the personalized parameters are input into a sleep pressure accumulation function to calculate the new sleep pressure, wherein the sleep pressure accumulation function is a weighted nonlinear combination of the age factor, the autonomic nervous system function index, the activity load, and the circadian rhythm shift.

[0024] In this embodiment, when a user is awake, their new sleep pressure can be calculated as follows: Specifically, personalized parameters are extracted from physiological data within the corresponding time interval. This physiological data includes the user's age, heart rate variability (HRV) data, and exercise data. In smart healthcare scenarios, this data is easily and continuously acquired through wearable devices such as smartwatches. Age is a fundamental physiological variable affecting metabolism and recovery ability; HRV data reflects the balance and regulation of the autonomic nervous system and is a core indicator of stress and recovery status; exercise data quantifies the user's daytime physical activity load, directly affecting the degree of fatigue accumulation. Specifically, when extracting personalized parameters, an age factor can be obtained through a non-linear formula based on age; autonomic nervous system function indicators can be derived from heart rate variability data (e.g., by analyzing the geometric features of a Poincaré diagram); activity load can be calculated based on exercise data (such as steps and intensity); and circadian rhythm shift can be obtained by analyzing the user's long-term historical sleep stage sequence and comparing their habitual bedtime with their current actual bedtime.

[0025] Finally, these personalized parameters are input into a preset sleep pressure accumulation function. This function is essentially a weighted nonlinear combination model that captures the synergistic and nonlinear effects of four factors—age, neurological state, physical exertion, and circadian rhythm deviation—on sleep pressure. The new sleep pressure calculated by the sleep pressure accumulation function is a quantified value representing the physiological load accumulated by the user within the current wakefulness period, which requires subsequent sleep to compensate for.

[0026] This application embodiment integrates multi-source physiological data and transforms it into personalized parameters with clear physiological significance. Then, it integrates and calculates the data through a sleep pressure accumulation function. This method can achieve more refined and dynamic quantification of sleep pressure within the framework of smart healthcare, making the updated current sleep pressure more accurate.

[0027] Optionally, in this embodiment of the application, the step of "extracting personalized parameters from physiological data within the corresponding time interval to characterize the rate of sleep pressure accumulation during the user's wakefulness" includes: calculating a nonlinear age factor based on the user's age; extracting the Poincaré plot standard deviation ratio from the heart rate variability data, and calculating the autonomic nervous function index of heart rate variability based on the standard deviation ratio; calculating the user's activity load based on the exercise data; obtaining the user's first historical sleep stage sequence and second historical sleep stage sequence, determining the personal biological clock type based on the first historical sleep stage sequence, determining the user's bedtime for the current sleep cycle based on the second historical sleep stage sequence, calculating the user's sleep time deviation based on the bedtime and the personal biological clock type, and calculating the circadian rhythm offset based on the sleep time deviation, wherein the first historical sleep stage sequence is greater than the second historical sleep stage sequence; Accordingly, the step of "inputting the personalized parameters into the sleep pressure accumulation function to calculate the new sleep pressure" includes: calculating the personalized pressure accumulation rate based on the age factor, the autonomic nervous system function index, the activity load, and the circadian rhythm offset, using a weighted nonlinear sub-function in the sleep pressure accumulation function; and calculating the new sleep pressure based on the personalized pressure accumulation rate and the preset time interval, using a sleep pressure sub-function in the sleep pressure accumulation function.

[0028] In this embodiment, a non-linear age factor is first calculated based on the user's age. The effect of age on sleep stress accumulation is not simply linear; rather, the body's recovery efficiency may decline more rapidly with age. Therefore, processing the age value using a non-linear function (such as a power function) allows for a more reasonable quantification of the fundamental weight of age in stress accumulation.

[0029] In a specific embodiment, the nonlinear age factor can be calculated using the following formula: ; in, Represents a non-linear age factor. Indicates the user's age. This indicates the preset coefficient.

[0030] Furthermore, it can process heart rate variability (HRV) data, specifically extracting the Poincaré plot standard deviation ratio (SD1 / SD2). The Poincaré plot is a method for analyzing heart rate fluctuation patterns, and its standard deviation ratio (SD1 / SD2) effectively characterizes the balance between the sympathetic and parasympathetic nervous systems in the autonomic nervous system. The autonomic nervous function indicators calculated using this standard deviation ratio directly reflect the user's current level of neural stress and relaxation, serving as a real-time physiological signal of accumulated stress.

[0031] In a specific embodiment, the autonomic nervous system function index can be calculated using the following formula: ; in, This indicates an indicator of autonomic nervous system function.

[0032] In addition, exercise data can be analyzed to calculate activity load. This metric integrates a user's exercise volume and intensity, such as combining daily steps and exercise heart rate data. It quantifies the physiological consumption directly generated by physical activity during wakefulness and is a direct and important component of sleep stress.

[0033] ; in, Indicates activity load, This represents the number of steps taken in the motion data. This represents the intensity of motion in the motion data. Indicates the baseline step number. , This indicates the preset coefficient.

[0034] The circadian rhythm shift of a user can also be determined by analyzing two sets of historical sleep stage sequences. The longer-term historical sleep stage sequences are used to analyze and determine the user's personal circadian rhythm type (e.g., morning or night person). The relatively short-term historical sleep stage sequences (including the current sleep cycle) are used to identify the bedtime of the current sleep cycle. By calculating the sleep time deviation between the user's habitual bedtime and actual bedtime, a quantified circadian rhythm shift is obtained, which accurately measures the degree of misalignment between the current sleep cycle and the user's internal biological clock on the timeline.

[0035] The sleep pressure accumulation function consists of two parts: a weighted nonlinear subfunction and a sleep pressure subfunction. After integrating all the personalized parameters mentioned above, the calculation phase begins. In the first stage, four parameters—age factor, autonomic nervous system function index, activity load, and circadian rhythm shift—are input into the weighted nonlinear subfunction of the sleep pressure accumulation function. This function uses a set of pre-trained or predefined weighting coefficients to weight the personalized parameters and consider their nonlinear interactions, ultimately outputting the personalized pressure accumulation rate, which represents the rate at which sleep pressure increases per unit time under the current physiological and behavioral state.

[0036] In one specific embodiment, an online learning weight update algorithm is employed during the weighted fusion of personalized parameters. This algorithm dynamically optimizes the weight coefficients of various influencing factors (such as age, heart rate variability complexity, activity load, and circadian rhythm shift) in the weighted nonlinear sub-function based on continuously generated physiological data from the user. The algorithm first performs anomaly detection on the input personalized parameters. If a personalized parameter deviates from the user's historical baseline by more than a preset range (e.g., two standard deviations), its reliability weight in this round of learning is significantly reduced to prevent accidental outliers from interfering with the stability of the results. Furthermore, the algorithm includes a baseline correction mechanism. When a continuous change in the user's lifestyle is detected, the user's historical baseline is automatically updated to ensure long-term adaptation to the user's dynamic changes. This closed-loop learning process allows the weighted nonlinear sub-function to continuously self-adjust and evolve over time, ultimately achieving a highly personalized calculation of sleep pressure accumulation rate.

[0037] In the second stage, the personalized pressure accumulation rate calculated above and the duration of the preset time interval are input into the sleep pressure sub-function (e.g., personalized pressure accumulation rate × duration of the preset time interval) to calculate the new sleep pressure accumulated in the current specific time interval.

[0038] This application not only considers traditional factors such as age and activity level, but also innovatively introduces a heart rate variability depth index based on Poincaré diagrams and a precise circadian rhythm offset based on dual-sequence analysis. This allows sleep pressure calculation to dynamically match the user's neural state and biological clock rhythm. Compared with the crude mode of ordinary fitness trackers that only rely on activity duration to estimate calorie consumption, this application provides a more accurate basis for fatigue quantification.

[0039] Optionally, in this embodiment, the sleep stage is further used to indicate the sleep type of the user; the user's sleep repayment value is calculated based on the following method: determining the sleep type indicated by the sleep stage, and calculating the corresponding debt repayment efficiency parameter according to the sleep type; and calculating the sleep repayment value based on the debt repayment efficiency parameter and the preset time interval.

[0040] In this embodiment, firstly, the identified sleep stages (such as the wakefulness stage, the sleep onset stage, etc.) are used to further determine the specific sleep type. Here, sleep type can include light sleep type, deep sleep type, and REM sleep type, etc. In sleep medicine, different sleep types have significantly different effects on restoring physical and mental energy (i.e., sleep debt repayment efficiency). Therefore, determining the sleep type can accurately determine the corresponding debt repayment efficiency parameter.

[0041] Next, a debt repayment efficiency parameter can be calculated based on the determined sleep type. This parameter is not a fixed value, but rather dynamically correlated with the determined sleep type, quantifying the efficiency of that sleep type in repaying sleep debt per unit time. For example, in smart healthcare practices, the debt repayment efficiency per unit time for deep sleep is usually set to be higher than that for light sleep.

[0042] Finally, the debt repayment efficiency parameter obtained above (representing the repayment capacity per unit time) is combined with the current preset time interval length (representing the duration) and a sleep repayment value generated within the time interval is obtained through mathematical calculation. This value represents the total amount of sleep debt that the user actually offsets physiologically during this specific duration and quality of sleep.

[0043] In smart healthcare scenarios, such as integrated smart mattresses or sleep aid systems, the embodiments of this application can not only report how long a user slept deeply, but also accurately calculate how much sleep debt was cleared during this stage. This transforms the restorative value of sleep from a vague subjective feeling of "whether I slept well or not" into objective, quantifiable indicators based on sleep type and duration. This quantification capability provides core data support for achieving truly personalized sleep management. For example, it provides data to assess the actual contribution of different interventions (such as environmental regulation and breathing guidance) to improving deep sleep efficiency, thus promoting the deepening of health management from experience and intuition to data-driven decision-making.

[0044] In this embodiment, optionally, the physiological data includes heart rate variability data, exercise data, and respiratory pattern data; the sleep type includes light sleep, deep sleep, and REM sleep; the step of "calculating the corresponding debt repayment efficiency parameter according to the sleep type" includes: for light sleep, calculating a sleep fragmentation index based on the sleep stage sequence identified from the beginning of the current sleep cycle to the current time interval, and calculating the corresponding debt repayment efficiency parameter based on the sleep fragmentation index; for deep sleep, extracting the low-frequency power ratio as a deep sleep indicator from the frequency domain analysis of the heart rate variability data in the corresponding time interval, and calculating the corresponding debt repayment efficiency parameter based on the deep sleep indicator; for REM sleep, calculating a rapid eye movement density index based on the exercise data and respiratory pattern data, and calculating the corresponding debt repayment efficiency parameter based on the rapid eye movement density index.

[0045] In this embodiment, each sleep type is not a fixed debt repayment efficiency parameter, but is dynamically calculated, thereby achieving personalization and accuracy of the debt repayment efficiency parameter.

[0046] Specifically, for light sleep, a key sleep fragmentation index is calculated by analyzing the sequence of sleep stages identified from the start of the current sleep cycle to the current time interval. This index reflects the continuity of sleep; frequent micro-awakenings or stage transitions reduce the quality of sleep recovery. Based on this sleep fragmentation index, the debt repayment efficiency parameter for light sleep is dynamically adjusted; the higher the degree of fragmentation, the lower the debt repayment efficiency parameter, thus quantifying the relatively low recovery utility of fragmented light sleep.

[0047] In a specific embodiment, the debt repayment efficiency parameter for the light sleep type can be calculated based on the following formula: ; A parameter representing the debt repayment efficiency of light sleep type. , Indicates the preset coefficient. This represents the sleep fragmentation index. In one specific embodiment, , .

[0048] In a specific embodiment, the sleep fragmentation index can be calculated as follows: from the start of the current sleep cycle to the current time interval, consecutive sleep stages of the same type are merged. After merging, the duration of each merged light sleep stage is determined. If the duration is less than a preset duration threshold, a count is performed. Finally, the quotient between the total count and the total number of merged sleep stages is calculated, and the result is used as the sleep fragmentation index.

[0049] For deep sleep types, frequency domain analysis can be performed based on heart rate variability data for the corresponding time intervals. In this analysis, the low-frequency power ratio is a key indicator, which is closely related to sleep depth and the activity of the autonomic nervous system. This ratio is used as an indicator of deep sleep; the higher the ratio, the deeper the sleep and the stronger the recovery ability. The debt repayment efficiency parameter is then calculated based on this.

[0050] In a specific embodiment, the debt repayment efficiency parameter corresponding to the deep sleep type can be calculated based on the following formula: ; This parameter represents the debt repayment efficiency corresponding to the type of deep sleep. , , Indicates the preset coefficient. This represents an indicator of deep sleep. In one specific embodiment, , , .

[0051] For REM sleep, the REM density index can be calculated based on motion and breathing pattern data. During REM sleep, muscle tone almost disappears, but the eyes move rapidly, and breathing becomes irregular and shallow. By analyzing the degree of inhibition of body movement signals and the variability of respiratory waveforms, the REM density index is calculated comprehensively. Based on this, the debt repayment efficiency parameter of this sleep stage is dynamically determined to measure its contribution to the recovery of brain function (such as memory integration).

[0052] In a specific embodiment, the debt repayment efficiency parameter corresponding to the REM sleep type can be calculated based on the following formula: ; This parameter represents the debt repayment efficiency corresponding to the REM sleep type. , , Indicates the preset coefficient. This represents a rapid eye movement density index. In one specific embodiment, , , .

[0053] In one specific embodiment, the rapid eye movement density index (REM density index) can be calculated by integrating key indicators from motion data and respiratory pattern data. First, a somatokinesis index is extracted from the motion data. During REM sleep, skeletal muscle tone almost completely disappears, resulting in a significant reduction in body movement. By analyzing triaxial accelerometer data from the motion data, the proportion of movement amplitude below a certain threshold is calculated, and combined with the detection of minute, sudden electromyographic activities, muscle twitching features associated with REM are identified, and these muscle twitching features are quantified as a somatokinesis index. Simultaneously, a respiratory variability index is extracted from the respiratory pattern data. Respiration during REM sleep is typically irregular, characterized by increased fluctuations in respiratory rate and significant changes in respiratory depth. By analyzing the time and frequency domain characteristics of the respiratory signal, indicators such as the coefficient of variation of the respiratory interval, the irregularity index of respiratory amplitude, and the chaos degree of the respiratory waveform are calculated to quantify the degree of deviation of the respiratory pattern from regularity, thus obtaining the respiratory variability index. Subsequently, the above-mentioned somatokinesis index and respiratory variability index are weighted and integrated into a comprehensive REM density index.

[0054] This application's embodiments design differentiated and dynamic debt repayment efficiency calculation models for different sleep types. Compared to the traditional approach of assigning fixed debt repayment efficiency values ​​to various sleep types, this approach can reflect the actual repayment amount of each sleep segment in real time based on multi-dimensional physiological signals, thus laying a precise and reliable quantitative foundation for generating truly personalized sleep improvement suggestions.

[0055] Optionally, in this embodiment, step 102, "identifying the user's sleep stage within a corresponding time interval based on the fusion feature vector of each time interval using a neural network model," includes: arranging the fusion feature vectors in chronological order of their generation, and selecting a preset number of consecutive target fusion feature vectors from the sorting results using a preset sliding window to form time-series data, wherein the generation time of each target fusion feature vector is earlier than the generation time of the remaining fusion feature vectors; inputting the time-series data into a recurrent neural network containing at least one layer of long short-term memory (LSTM), receiving the fusion feature vector of each time interval in the time-series data through the input layer of the recurrent neural network, and learning and memorizing the dependency relationship between the fusion feature vectors of preceding and following time intervals through the at least one layer of LTM in the recurrent neural network. The system generates a hidden state vector for each time interval. Through the attention mechanism layer of the recurrent neural network, it calculates the importance weight for each time interval and weights the hidden state vectors of each time interval with their corresponding importance weights to obtain a weighted feature vector. Through the fully connected layer of the recurrent neural network, based on the weighted feature vector, it outputs the probability distribution of the latest time interval corresponding to multiple preset sleep stages. Based on the probability distribution corresponding to the latest time interval, it determines the user's initial sleep stage in the latest time interval from among the multiple preset sleep stages. It calculates the confidence level corresponding to the initial sleep stage. If the confidence level is less than a preset confidence threshold, it determines the sleep stage of the latest time interval based on adjacent sleep stages; otherwise, it directly uses the initial sleep stage as the sleep stage of the latest time interval.

[0056] In this embodiment, the fusion feature vectors generated in each time interval are first arranged in chronological order. To capture the temporal evolution of sleep states, a preset sliding window method is used to select multiple consecutive (e.g., 10) generated fusion feature vectors from the sorting results to form a set of time-series data. Here, the generation time of all target fusion feature vectors within the preset time window is earlier than the remaining fusion feature vectors outside the preset time window. This ensures that the recurrent neural network makes current judgments based only on known, past, and latest information, meeting the requirements of real-time processing or causal analysis.

[0057] Next, the time-series data is fed into a specially designed recurrent neural network (RNN). This RNN may include an input layer, at least one Long Short-Term Memory (LSTM) layer, an attention mechanism layer, and a fully connected layer. Specifically, the input layer first receives the fused feature vector for each time interval in the time-series data. Then, the LSTM begins operation. LSTM has a unique memory gating mechanism, enabling it to effectively learn and remember long-term dependencies between the fused feature vectors of consecutive time intervals, such as recognizing common patterns of a smooth transition from "light sleep" to "deep sleep." After LSTM processing, each time interval can be transformed into a hidden state vector containing contextual information. To further improve the performance of the RNN, an attention mechanism layer is introduced, which automatically evaluates and calculates the importance weights of different time intervals within a preset time window for determining the latest time interval. Then, the hidden state vectors of all time intervals are weighted and summed according to their importance weights to obtain a comprehensive weighted feature vector. Finally, this weighted feature vector is transformed through a fully connected layer, outputting a probability distribution for the latest time interval that covers multiple preset sleep stages (such as wakefulness, falling asleep, etc.). This distribution reflects the confidence level of a recurrent neural network in identifying a user as being in each sleep stage.

[0058] Then, a preliminary decision is made based on the above probability distribution. Specifically, the sleep stage with the highest probability value is selected as the user's initial sleep stage in the latest time interval. To assess the reliability of this judgment, its confidence level can be calculated, or the highest probability value itself can be used directly. If this confidence level is lower than a preset confidence threshold (e.g., 0.9), it indicates that the recurrent neural network's judgment is not very certain and may be in a period of ambiguity during stage transition. In this case, smoothing logic will be activated, referencing information from previous sleep stages adjacent to this time interval to correct the final result, for example, adopting the more stable and certain one among the adjacent stages. If the confidence level is high enough, the initial sleep stage is directly adopted as the final sleep stage for this latest time interval.

[0059] In a specific embodiment, the structure of the recurrent neural network is as follows: Figure 2 As shown. Specifically, it includes: an input layer: responsible for receiving and standardizing the fusion feature vector for each time interval. The input to this layer is a 64-dimensional fusion feature vector generated according to the aforementioned method, which consists of four categories of 16-dimensional time-frequency domain features: heart rate, respiration, body movement, and environment.

[0060] Feature Processing Layer: First, each fused feature vector is normalized to eliminate differences in the dimensions of different physiological signals. Then, the fused feature vectors from multiple consecutive time intervals (e.g., 10) are organized into time-series data in chronological order to capture the dynamic evolution of sleep states. In practical deployments of smart healthcare, this process ensures that raw data streaming in real-time from wearable devices can be transformed into a standardized, processable time-series input in real time.

[0061] LSTM Layer: Used for deep mining of temporal dependencies in sleep states. This embodiment employs a three-layer stacked Long Short-Term Memory (LSTM) unit, with each layer containing 128 neurons. LSTM is a special type of recurrent neural network whose internal gating mechanism effectively learns and memorizes the dependencies between states in long sequences, such as the typical pattern of a smooth transition from "light sleep" to "deep sleep." The output sequence of the previous layer serves as the input to the next layer. This deep structure allows the model to progressively extract more abstract and discriminative temporal representations from the fused feature vectors, generating a series of hidden state vectors containing contextual information.

[0062] Attention Mechanism Layer: This layer dynamically focuses on key time intervals to improve recognition accuracy. It applies temporal attention to the hidden state vectors output by the LSTM. Using a trainable parameter matrix, it calculates the importance weight of the hidden state vectors for each time interval. Then, it sums all the hidden state vectors according to their corresponding importance weights to obtain the final weighted feature vector.

[0063] Output Layer and Classification Results: The weighted feature vector is fed into a fully connected layer for linear transformation, mapping it to the same dimension as the number of sleep stage categories. Subsequently, the Softmax activation function converts the output into a probability distribution, where each value represents the confidence probability that the latest time interval belongs to "awake," "light sleep," "deep sleep," or "REM sleep." Finally, the category with the highest probability is selected as the initial sleep stage for that time interval. It's important to note that the sleep stage can be the same as or different from the sleep type.

[0064] This application combines an advanced deep learning architecture with rigorous post-processing logic. In the field of smart healthcare, compared to the misjudgments caused by traditional devices relying on fixed rules or simple models, this application, through LSTM and attention mechanisms, enables recurrent neural networks to understand the dynamic transition patterns of sleep stages like an expert, and to pay more attention to key time intervals. Simultaneously, the introduction of confidence-based post-processing smoothing effectively filters out occasional, physiologically inconsistent, jump-like misjudgments, making the final output sleep stage more reliable.

[0065] Optionally, in this embodiment of the application, the method further includes: obtaining the current system state, wherein the current system state includes the current sleep debt, the environmental data, the third historical sleep stage sequence, and the user-configured personal preferences; inputting the current system state into a reinforcement learning policy network based on an Actor-Critic architecture; based on the current system state, outputting the probability distribution of various sleep improvement actions through the Actor network in the reinforcement learning policy network; and selecting target improvement actions from the probability distribution through an exploration strategy in the reinforcement learning policy network. Accordingly, the method further includes: after confirming that the target improvement action has been performed, continuously monitoring the user's physiological data and environmental data to obtain subsequent sleep quality assessment results; calculating a reward value based on the subsequent sleep quality assessment results, wherein the reward value is a weighted combination of the degree of sleep quality improvement, the amount of sleep debt reduction, and the degree of sleep disturbance caused by performing the target improvement action; evaluating the state value through the Critic network in the reinforcement learning policy network based on the reward value and the current system state, and updating the network parameters of the Actor network and the Critic network according to the state value, so as to update the reinforcement learning policy network.

[0066] In this embodiment, the current system state is first obtained. This includes current sleep debt, environmental data (such as bedroom temperature and noise level), the third historical sleep stage sequence used to understand sleep patterns, and personal preferences actively set by the user (such as not wanting to be advised to wake up too early). These four dimensions of information together constitute the context for decision-making, ensuring that subsequent recommendations are generated based on objective physiological indicators while respecting personal lifestyle habits and subjective wishes.

[0067] Next, the current system state is input into a reinforcement learning policy network based on an "Actor-Critic architecture." The Actor network analyzes the current system state and outputs a probability distribution of a series of possible sleep improvement actions (such as "going to bed 30 minutes earlier" or "adjusting the air conditioner to 22°C"), representing the probability that each sleep improvement action is deemed appropriate. To balance utilizing known effective strategies with exploring new possibilities, an exploration strategy (e.g., an ε-greedy strategy) is employed. Based on the probability distribution and incorporating randomness, a target improvement action is ultimately selected and recommended to the user. Here, the target improvement action is also known as the sleep improvement suggestion.

[0068] After the target improvement actions are identified, the feedback collection phase can begin. Once the target improvement actions are confirmed to have been implemented (e.g., the user followed the early bedtime recommendation, or the smart home automatically adjusted the environment), the user's subsequent physiological and environmental data are continuously monitored. Based on this new physiological and environmental data, a new sleep quality assessment result is calculated again, serving as an objective basis for measuring the effectiveness of the intervention.

[0069] Based on the comparison before and after the intervention, a reward value can be calculated, which is a key feedback signal in reinforcement learning. Here, the reward function can be designed as a multi-dimensional weighted combination: positive rewards are set for "the degree of improvement in sleep quality" and "the amount of reduction in sleep debt," reflecting the core health benefits of the intervention; at the same time, negative penalties are set for "the degree of disruption to sleep caused by the action" (such as discomfort caused by a significant change in sleep schedule) to avoid making the recommendations too aggressive.

[0070] Finally, the calculated reward value is used for learning and self-optimization. The Critic network in the reinforcement learning policy network is responsible for evaluating the long-term cumulative reward, i.e., the state value, in the current system state. By comparing the actual reward value with the Critic's prediction, the prediction error is calculated, and the parameters of both the Actor and Critic networks are updated accordingly. This process makes the Actor network more inclined to choose high-reward target improvement actions in similar states in the future, while the Critic network's evaluation becomes increasingly accurate, thus gradually improving the intelligence level of the entire reinforcement learning policy network through iterative iteration.

[0071] The reinforcement learning process in this embodiment learns and adapts to each user's unique physiological responses and life constraints through continuous interaction. This personalized ability, which evolves through closed-loop learning with real-world feedback, can significantly improve the acceptability of targeted improvement actions and the user's experience.

[0072] Furthermore, as a refinement and extension of the specific implementation methods of the above embodiments, and to fully illustrate the specific implementation process of this embodiment, a multi-dimensional sleep quality assessment system is provided, such as... Figure 3 As shown, the system includes: The first layer is a multi-dimensional data acquisition layer responsible for simultaneously acquiring the user's physiological and environmental data. This layer deploys physiological and environmental sensing units. The physiological sensing units specifically include a heart rate variability sensor for monitoring the state of the autonomic nervous system, a respiratory monitoring sensor for capturing breathing patterns and depth, and a body motion sensor with a triaxial accelerometer for detecting body movement and posture. The environmental sensing units include environmental sensors for simultaneously acquiring noise, light, temperature, and humidity data within the sleep microenvironment. In smart healthcare applications, these sensors can be integrated into wearable devices to achieve non-intrusive, long-term, comprehensive data acquisition for the user, providing a high-quality multimodal data source for subsequent analysis.

[0073] The second layer, the data processing layer, is responsible for standardizing and preprocessing the raw sensor data and fusing its features. This layer first uses a data preprocessing module to clean, denoise, and time-synchronize heterogeneous data from multiple sources to eliminate signal artifacts and ensure spatiotemporal consistency. Subsequently, a feature extraction module extracts time-frequency domain features from various data types according to a preset time interval (e.g., 30 seconds), such as calculating the time-domain standard deviation and frequency-domain power spectral density from heart rate signals. Finally, the data fusion module employs a feature-level fusion algorithm to integrate time-frequency domain feature vectors from different sensors into a unified, high-dimensional fused feature vector. This vector comprehensively represents the user's sleep state information within the corresponding time interval.

[0074] The third layer, the intelligent analysis layer, serves as the core of the system, enabling accurate identification of sleep stages and dynamic quantitative modeling of sleep debt. This layer first uses a neural network recognition module to input time-series data into a recurrent neural network model containing long short-term memory units and an attention mechanism layer. Leveraging its powerful ability to learn temporal dependencies, it outputs the probability distribution of the latest time interval corresponding to different sleep stages (wakefulness, light sleep, deep sleep, and REM sleep), which is then post-processed and smoothed to obtain the final judgment result. Simultaneously, the sleep debt modeling module introduces a "stress-repayment" quantitative model: when the user is awake, the system calculates the accumulated sleep stress (i.e., increased debt) for that period based on their age, real-time heart rate variability index, activity load, and deviation from their personal circadian rhythm using a non-linear weighting function. When the user is asleep, the system dynamically calculates the sleep repayment value (i.e., decreased debt) for that period based on the identified specific sleep type and its corresponding physiological signal quality (e.g., low-frequency power of heart rate variability during deep sleep). The state assessment module accumulates these changes in real time, dynamically updating and outputting a quantified current sleep debt as a core objective indicator for evaluating sleep recovery efficiency.

[0075] The fourth layer is the optimization decision layer, which generates personalized intervention strategies based on sleep quality assessment results and continuously optimizes them. The core of this layer is a reinforcement learning engine based on an Actor-Critic architecture. The Actor network in the engine receives a system state containing current sleep debt, environmental data, historical sleep stage sequences, and user preferences, and outputs a probability distribution of a series of potential sleep improvement actions (such as adjusting environmental parameters and suggesting bedtime). After exploring strategies, selecting specific actions, and executing them, the system continuously monitors subsequent changes in sleep debt and calculates feedback values ​​based on a reward function that integrates the degree of sleep quality improvement, the amount of debt reduction, and the degree of disruption to the user's sleep schedule. The Critic network evaluates the state value and uses this feedback to update the parameters of both the Actor and Critic networks simultaneously using gradient descent. This allows the policy network to continuously learn and adapt to the user's personalized response patterns over time, generating increasingly accurate and effective improvement suggestions.

[0076] The fifth layer is the user interaction layer, responsible for information exchange and visualization between the system and the user. This layer is typically implemented as a front-end mobile application or web interface. Its core functions include: clearly presenting complex sleep debt curves, the proportion of each sleep stage, and environmental data trends in chart form through a data visualization module; allowing users to set their own preferences and constraints (such as the earliest wake-up time) through a personalized configuration module to ensure the feasibility of the recommendations; and converting the actions output by the optimization decision layer into natural language suggestions or directly controlling smart home devices through a suggestion generation module, thereby forming a complete personalized service loop of "evaluation-suggestion-feedback-optimization".

[0077] This application's embodiments, through the synergistic effect of the aforementioned five-layer architecture, achieve a complete technical closed loop for sleep quality assessment, from multi-dimensional perception and intelligent analysis to personalized decision-making. Specifically, by using multi-source data fusion and recurrent neural networks, the accuracy and robustness of sleep stage identification are significantly improved; through an innovative dynamic sleep debt quantification model, real-time and personalized assessment of the sleep recovery process is achieved, surpassing the limitations of traditional static threshold methods; furthermore, by leveraging a reinforcement learning framework, the system possesses the adaptive ability to continuously optimize intervention strategies based on long-term interactive feedback. This solution provides an implementable, evolvable, and forward-looking systematic solution for personalized sleep health management in the field of smart healthcare.

[0078] Furthermore, as Figure 1 In terms of specific implementation, this application provides a multi-dimensional sleep quality assessment device, such as... Figure 4 As shown, the device includes: The data acquisition module is used to continuously and synchronously collect the user's physiological and environmental data; The sleep stage identification module is used to extract time-frequency domain features from the continuously collected physiological data and environmental data according to a preset time interval, fuse the extraction results to obtain a fused feature vector representing the user's sleep state in each time interval, and identify the user's sleep stage in the corresponding time interval through a neural network model based on the fused feature vector of each time interval. The sleep debt update module is used to dynamically calculate the change in the user's sleep debt in the corresponding time interval based on the sleep stage and the physiological data in the corresponding time interval whenever a new sleep stage is identified, and update the user's current sleep debt according to the change in sleep debt. If the sleep stage indicates that the user is awake, the user's new sleep pressure is calculated as the change in sleep debt; if the sleep stage indicates that the user is asleep, the user's sleep repayment value is calculated as the change in sleep debt. The sleep quality assessment module is used to dynamically determine the user's sleep quality assessment result based on the updated current sleep debt.

[0079] Optionally, the physiological data includes user age, heart rate variability data, and exercise data; the sleep debt update module is used for: Personalized parameters are extracted from the physiological data within the corresponding time interval to characterize the rate of sleep pressure accumulation during the user's wakefulness. The personalized parameters include an age factor determined based on the user's age, an autonomic nervous system function index determined based on the heart rate variability data, an activity load calculated based on the exercise data, and a circadian rhythm offset determined based on the historical sleep stage sequence. The personalized parameters are input into the sleep pressure accumulation function to calculate the new sleep pressure, wherein the sleep pressure accumulation function is a weighted nonlinear combination of the age factor, the autonomic nervous system function index, the activity load and the circadian rhythm offset.

[0080] Optionally, the sleep debt update module is further configured to: Calculate a non-linear age factor based on the user's age; The Poincaré ratio of standard deviation is extracted from the heart rate variability data, and the autonomic nervous function index of heart rate variability is calculated based on the standard deviation ratio. Calculate the user's activity load based on the motion data; The system acquires the user's first historical sleep stage sequence and second historical sleep stage sequence. Based on the first historical sleep stage sequence, it determines the user's personal biological clock type. Based on the second historical sleep stage sequence, it determines the user's bedtime for the current sleep cycle. Based on the bedtime and the user's personal biological clock type, it calculates the user's sleep time deviation and calculates the circadian rhythm offset based on the sleep time deviation. The first historical sleep stage sequence is greater than the second historical sleep stage sequence. Accordingly, the sleep debt update module is also used for: Based on the age factor, the autonomic nervous system function index, the activity load, and the circadian rhythm offset, a personalized stress accumulation rate is calculated using a weighted nonlinear sub-function in the sleep stress accumulation function. Based on the personalized pressure accumulation rate and the preset time interval, the new sleep pressure is calculated using the sleep pressure sub-function in the sleep pressure accumulation function.

[0081] Optionally, the sleep stage is further used to indicate the sleep type the user is in; the sleep debt update module is further used to: Determine the sleep type indicated by the sleep stage, and calculate the corresponding debt repayment efficiency parameter based on the sleep type; Based on the debt repayment efficiency parameter and the preset time interval, the sleep repayment value is calculated.

[0082] Optionally, the physiological data includes heart rate variability data, exercise data, and respiratory pattern data; the sleep types include light sleep, deep sleep, and REM sleep; the sleep debt update module is further used for: For light sleep, a sleep fragmentation index is calculated based on the sleep stage sequence identified from the start of the current sleep cycle to the current time interval, and the corresponding debt repayment efficiency parameter is calculated based on the sleep fragmentation index. For deep sleep types, the low-frequency power ratio is extracted as a deep sleep index from the frequency domain analysis of heart rate variability data in the corresponding time interval, and the corresponding debt repayment efficiency parameter is calculated based on the deep sleep index. For REM sleep, a REM density index is calculated based on the motion data and breathing pattern data, and a corresponding debt repayment efficiency parameter is calculated based on the REM density index.

[0083] Optionally, the sleep stage recognition module is used to: The fused feature vectors are arranged in order of generation time, and a preset number of consecutive target fused feature vectors are selected from the sorting results according to a preset sliding window to form time series data, wherein the generation time of each target fused feature vector is earlier than the generation time of the remaining fused feature vectors. The time-series data is input into a recurrent neural network containing at least one layer of long short-term memory (LSTM). The input layer of the recurrent neural network receives the fused feature vector of each time interval in the time-series data. The LTM layer of the recurrent neural network learns and memorizes the dependency relationship between the fused feature vectors of the preceding and following time intervals and generates a hidden state vector for each time interval. The attention mechanism layer of the recurrent neural network calculates the importance weight of each time interval and adds the hidden state vector of each time interval to the corresponding importance weight to obtain a weighted feature vector. Based on the weighted feature vector, the fully connected layer of the recurrent neural network outputs the probability distribution of the latest time interval corresponding to multiple preset sleep stages. Based on the probability distribution corresponding to the latest time interval, the user's initial sleep stage in the latest time interval is determined from the multiple preset sleep stages; Calculate the confidence level corresponding to the initial sleep stage. If the confidence level is less than a preset confidence threshold, then determine the sleep stage of the latest time interval based on adjacent sleep stages; otherwise, directly use the initial sleep stage as the sleep stage of the latest time interval.

[0084] Optionally, the device further includes a feedback module; the feedback module is used for: Obtain the current system status, which includes the current sleep debt, the environmental data, the third historical sleep stage sequence, and the user-configured personal preferences; The current system state is input into a reinforcement learning policy network based on the Actor-Critic architecture. Based on the current system state, the probability distribution of various sleep improvement actions is output through the Actor network in the reinforcement learning policy network. The target improvement action is selected from the probability distribution through the exploration strategy in the reinforcement learning policy network. Accordingly, the feedback module is also used for: After confirming that the target improvement action has been performed, the user's physiological and environmental data are continuously monitored to obtain subsequent sleep quality assessment results; Based on the subsequent sleep quality assessment results, a reward value is calculated, wherein the reward value is a weighted combination of the degree of improvement in sleep quality, the amount of reduction in sleep debt, and the degree of disruption to sleep caused by the execution of the target improvement actions; Based on the reward value and the current system state, the state value is evaluated through the Critic network in the reinforcement learning policy network, and the network parameters of the Actor network and the Critic network are updated according to the state value to update the reinforcement learning policy network.

[0085] It should be noted that other corresponding descriptions of the functional units involved in the multi-dimensional sleep quality assessment device provided in this application embodiment can be found in the following references. Figures 1 to 2 The corresponding descriptions in the method will not be repeated here.

[0086] This application also provides a computer device, which may specifically be a personal computer, a server, a network device, etc. Figure 5 As shown, the computer device includes a bus, a processor, memory, and a communication interface, and may also include an input / output interface and a display device. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database stores location information. The network interface allows communication with external terminals via a network connection. When the computer program is executed by the processor, it implements the steps in the various method embodiments.

[0087] Those skilled in the art will understand that Figure 5 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0088] In one embodiment, a computer-readable storage medium is provided, which may be non-volatile or volatile, having stored thereon a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0089] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0090] It should be noted that any AI models, software tools, or components not belonging to this company appearing in the embodiments of this application are merely illustrative examples and do not represent actual use. All user personal information involved in the embodiments of this application has been authorized (with the knowledge and consent) by the relevant parties or has been fully authorized by all parties, and the executing entity may obtain it through various legal and compliant means. The collection, storage, use, processing, transmission, provision, and disclosure of the information, data, and signals involved all comply with relevant laws and regulations and do not violate public order and good morals.

[0091] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0092] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0093] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A multi-dimensional assessment method for sleep quality, characterized in that, include: Continuously and synchronously collect users' physiological and environmental data; The continuously collected physiological and environmental data are subjected to time-frequency domain feature extraction according to a preset time interval. The extraction results are then fused to obtain a fused feature vector representing the user's sleep state in each time interval. Based on the fused feature vector of each time interval, a neural network model is used to identify the user's sleep stage in the corresponding time interval. Whenever a new sleep stage in a new time interval is identified, the change in the user's sleep debt in the corresponding time interval is dynamically calculated based on the sleep stage and the physiological data within the corresponding time interval. The user's current sleep debt is then updated based on the change in sleep debt. If the sleep stage indicates that the user is awake, the user's new sleep pressure is calculated as the change in sleep debt. If the sleep stage indicates that the user is asleep, the user's sleep repayment value is calculated as the change in sleep debt. The user's sleep quality assessment result is dynamically determined based on the updated current sleep debt.

2. The method according to claim 1, characterized in that, The physiological data includes the user's age, heart rate variability data, and exercise data; the user's new sleep pressure is calculated based on the following method: Personalized parameters are extracted from the physiological data within the corresponding time interval to characterize the rate of sleep pressure accumulation during the user's wakefulness. The personalized parameters include an age factor determined based on the user's age, an autonomic nervous system function index determined based on the heart rate variability data, an activity load calculated based on the exercise data, and a circadian rhythm offset determined based on the historical sleep stage sequence. The personalized parameters are input into the sleep pressure accumulation function to calculate the new sleep pressure, wherein the sleep pressure accumulation function is a weighted nonlinear combination of the age factor, the autonomic nervous system function index, the activity load and the circadian rhythm offset.

3. The method according to claim 2, characterized in that, The step of extracting personalized parameters from physiological data within the corresponding time interval to characterize the rate of sleep pressure accumulation during the user's wakefulness includes: Calculate a non-linear age factor based on the user's age; The Poincaré ratio of standard deviation is extracted from the heart rate variability data, and the autonomic nervous function index of heart rate variability is calculated based on the standard deviation ratio. Calculate the user's activity load based on the motion data; The system acquires the user's first historical sleep stage sequence and second historical sleep stage sequence. Based on the first historical sleep stage sequence, it determines the user's personal biological clock type. Based on the second historical sleep stage sequence, it determines the user's bedtime for the current sleep cycle. Based on the bedtime and the user's personal biological clock type, it calculates the user's sleep time deviation and calculates the circadian rhythm offset based on the sleep time deviation. The first historical sleep stage sequence is greater than the second historical sleep stage sequence. Accordingly, the step of inputting the personalized parameters into the sleep pressure accumulation function to calculate the new sleep pressure includes: Based on the age factor, the autonomic nervous system function index, the activity load, and the circadian rhythm offset, a personalized stress accumulation rate is calculated using a weighted nonlinear sub-function in the sleep stress accumulation function. Based on the personalized pressure accumulation rate and the preset time interval, the new sleep pressure is calculated using the sleep pressure sub-function in the sleep pressure accumulation function.

4. The method according to claim 1, characterized in that, The sleep stage also indicates the sleep type the user is in; the user's sleep repayment value is calculated based on the following method: Determine the sleep type indicated by the sleep stage, and calculate the corresponding debt repayment efficiency parameter based on the sleep type; Based on the debt repayment efficiency parameter and the preset time interval, the sleep repayment value is calculated.

5. The method according to claim 4, characterized in that, The physiological data includes heart rate variability data, exercise data, and breathing pattern data; the sleep types include light sleep, deep sleep, and REM sleep. The step of calculating the corresponding debt repayment efficiency parameter based on the sleep type includes: For light sleep, a sleep fragmentation index is calculated based on the sleep stage sequence identified from the start of the current sleep cycle to the current time interval, and the corresponding debt repayment efficiency parameter is calculated based on the sleep fragmentation index. For deep sleep types, the low-frequency power ratio is extracted as a deep sleep index from the frequency domain analysis of heart rate variability data in the corresponding time interval, and the corresponding debt repayment efficiency parameter is calculated based on the deep sleep index. For REM sleep, a REM density index is calculated based on the motion data and breathing pattern data, and a corresponding debt repayment efficiency parameter is calculated based on the REM density index.

6. The method according to claim 1, characterized in that, The method of identifying the user's sleep stage within a corresponding time interval using a neural network model based on the fused feature vector for each time interval includes: The fused feature vectors are arranged in order of generation time, and a preset number of consecutive target fused feature vectors are selected from the sorting results according to a preset sliding window to form time series data, wherein the generation time of each target fused feature vector is earlier than the generation time of the remaining fused feature vectors. The time-series data is input into a recurrent neural network containing at least one layer of long short-term memory (LSTM). The input layer of the recurrent neural network receives the fused feature vector of each time interval in the time-series data. The LTM layer of the recurrent neural network learns and memorizes the dependency relationship between the fused feature vectors of the preceding and following time intervals and generates a hidden state vector for each time interval. The attention mechanism layer of the recurrent neural network calculates the importance weight of each time interval and adds the hidden state vector of each time interval to the corresponding importance weight to obtain a weighted feature vector. Based on the weighted feature vector, the fully connected layer of the recurrent neural network outputs the probability distribution of the latest time interval corresponding to multiple preset sleep stages. Based on the probability distribution corresponding to the latest time interval, the user's initial sleep stage in the latest time interval is determined from the multiple preset sleep stages; Calculate the confidence level corresponding to the initial sleep stage. If the confidence level is less than a preset confidence threshold, then determine the sleep stage of the latest time interval based on adjacent sleep stages; otherwise, directly use the initial sleep stage as the sleep stage of the latest time interval.

7. The method according to claim 1, characterized in that, The method further includes: Obtain the current system status, which includes the current sleep debt, the environmental data, the third historical sleep stage sequence, and the user-configured personal preferences; The current system state is input into a reinforcement learning policy network based on the Actor-Critic architecture. Based on the current system state, the probability distribution of various sleep improvement actions is output through the Actor network in the reinforcement learning policy network. The target improvement action is selected from the probability distribution through the exploration strategy in the reinforcement learning policy network. Accordingly, the method further includes: After confirming that the target improvement action has been performed, the user's physiological and environmental data are continuously monitored to obtain subsequent sleep quality assessment results; Based on the subsequent sleep quality assessment results, a reward value is calculated, wherein the reward value is a weighted combination of the degree of improvement in sleep quality, the amount of reduction in sleep debt, and the degree of disruption to sleep caused by the execution of the target improvement actions; Based on the reward value and the current system state, the state value is evaluated through the Critic network in the reinforcement learning policy network, and the network parameters of the Actor network and the Critic network are updated according to the state value to update the reinforcement learning policy network.

8. A multi-dimensional sleep quality assessment device, characterized in that, include: The data acquisition module is used to continuously and synchronously collect the user's physiological and environmental data; The sleep stage identification module is used to extract time-frequency domain features from the continuously collected physiological data and environmental data according to a preset time interval, fuse the extraction results to obtain a fused feature vector representing the user's sleep state in each time interval, and identify the user's sleep stage in the corresponding time interval through a neural network model based on the fused feature vector of each time interval. The sleep debt update module is used to dynamically calculate the change in the user's sleep debt in the corresponding time interval based on the sleep stage and the physiological data in the corresponding time interval whenever a new sleep stage is identified, and update the user's current sleep debt according to the change in sleep debt. If the sleep stage indicates that the user is awake, the user's new sleep pressure is calculated as the change in sleep debt; if the sleep stage indicates that the user is asleep, the user's sleep repayment value is calculated as the change in sleep debt. The sleep quality assessment module is used to dynamically determine the user's sleep quality assessment result based on the updated current sleep debt.

9. A storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 7.

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

Images