Dynamic sleep monitoring and sleep cycle guide planning system and method

Abstract

The invention discloses a dynamic sleep monitoring and sleep cycle guide planning system and method, and the system comprises a data collection module which is used for synchronously collecting the signal data of a subject through a multi-channel biosensor; the feature extraction module is used for performing real-time frequency band decomposition on the signal data and extracting power spectral density features of the brain waves; the sleep stage classification module is used for segmenting the signal data, constructing a time-frequency joint distribution characteristic matrix based on the power spectrum density characteristic corresponding to the segmentation result, and segmenting the sleep stage based on the time-frequency joint distribution characteristic matrix; the index calculation module is used for calculating a multi-dimensional sleep quality evaluation index in real time based on the sleep stage classification result and realizing dynamic sleep monitoring based on the multi-dimensional sleep quality evaluation index; and the demand instruction generation module is used for receiving the user description, converting the user description into constraint conditions through semantic understanding, generating a guide instruction based on the dynamic sleep monitoring result and the constraint conditions, and completing sleep cycle guide planning.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of sleep health management, in particular to a system and method for dynamic sleep monitoring and sleep cycle guiding planning. BACKGROUND

[0002] Sleep, as a core physiological process of the human body, plays an irreplaceable role in maintaining physical and mental health. About one-third of adults worldwide are troubled by insomnia, which is characterized by sleep continuity disorders, including prolonged sleep latency, increased night wakefulness, early awakening and reduced sleep efficiency. Such disorders not only directly lead to impaired daytime functioning, but also are important risk factors for comorbidities such as cardiovascular metabolic diseases and depression. Studies have shown that the chronicity of insomnia is associated with neurobiological adaptive changes, which can induce a "mixed wakefulness state" and further disrupt sleep homeostatic regulation.

[0003] The prior art has the following limitations: The existing sleep monitoring and intervention technology has three major defects: 1. The real-time monitoring is severely insufficient. Medical-grade polysomnography (PSG) requires manual interpretation, resulting in a feedback delay of more than 6 hours. Consumer-grade devices (such as Apple Watch) have a 40% missed detection rate for key sleep stage transition events due to a low sampling rate of 1 Hz, and signal processing delays cause the intervention measures to be misaligned with the brain state, such as mistakenly applying stimulation that should be applied to the rising edge at the slow wave falling edge, which further exacerbates sleep fragmentation; 2. The prediction mechanism is completely missing. The existing technology lacks the ability to model the temporal evolution of the sleep structure throughout the night, making it impossible to predict trends such as the attenuation of deep sleep or the shortening of REM cycles 30 minutes later, resulting in the missed opportunity to intervene; 3. The demand matching is severely disconnected. Standardized intervention strategies ignore individual differences in goals, and some neuroregulation stimulation parameters are fixed, making it impossible to dynamically adapt to different goals such as "enhancing slow waves" or "suppressing wakefulness". There is no mechanism for generating personalized sleep blueprints related to age and physiological state, creating a technical bottleneck where the monitoring, prediction, and intervention modules are disconnected.

[0004] The prior art has not yet solved the problem of "monitoring → prediction → planning → guidance" optimization, and there is an urgent need for a system that integrates dynamic prediction models, personalized sleep structure planning, and predictive intervention to break through the bottleneck of sleep regulation. SUMMARY

[0005] To solve the above technical problems, the present application provides a system and method for dynamic sleep monitoring and sleep cycle guiding planning, which specifically includes: A data acquisition module for synchronously acquiring signal data of a subject through a multi-channel biological sensor; A feature extraction module for real-time frequency band decomposition of the signal data to extract power spectral density features of brain waves; The sleep stage classification module is configured to segment the signal data, construct a time-frequency joint distribution feature matrix based on power spectrum density features corresponding to the segmentation results, and segment the sleep stage based on the time-frequency joint distribution feature matrix. The index calculation module is configured to calculate a multi-dimensional sleep quality evaluation index in real time based on the sleep stage classification results, and realize dynamic sleep monitoring based on the multi-dimensional sleep quality evaluation index. The demand instruction generation module is configured to receive user descriptions and convert them into constraint conditions through semantic understanding, generate guiding instructions based on the dynamic sleep monitoring results and the constraint conditions, and complete sleep cycle guiding planning based on the guiding instructions.

[0006] Optionally, in the data acquisition module, the signal data includes electroencephalogram signals, electrocardiogram signals, respiratory signals, and body movement signals. The electroencephalogram signals are collected by an 8-channel dry electrode cap, and the sampling frequency is 256 Hz. The electrocardiogram signals and the respiratory signals are obtained by a flexible conductive patch. The body movement signals are obtained by real-time monitoring using a three-axis accelerometer.

[0007] Optionally, the working process of the feature extraction module specifically includes: The signal data is subjected to real-time frequency band decomposition, and the power spectrum features of alpha waves, beta waves, theta waves, and delta waves are extracted, wherein the alpha waves are brain electrical signals of 8-12 Hz, the beta waves are brain electrical signals of 12-30 Hz, the theta waves are brain electrical signals of 4-8 Hz, and the delta waves are brain electrical signals of 0.5-4 Hz.

[0008] Optionally, the sleep stage classification module includes a segmentation processing submodule, a matrix construction submodule, and a classification submodule. The segmentation processing submodule is configured to segment the signal data in time windows of 30 seconds to obtain a sequence of standardized data frames. The matrix construction submodule is configured to perform wavelet transform on the electroencephalogram signals of the sequence of standardized data frames, construct a time-frequency joint distribution feature matrix based on the wavelet transform results and the power spectrum density features of the electroencephalogram waves, and obtain micro-scale fluctuations based on the time-frequency joint distribution feature matrix. The classification submodule is configured to fuse the micro-scale fluctuations with RR interval variation features of the electrocardiogram signals and harmonic distortion rates of the respiratory signals to obtain fusion information, output six-stage sleep labels through a support vector machine classifier based on the fusion information, and segment the sleep stage based on the sleep labels.

[0009] Optionally, in the matrix construction submodule, obtaining the micro-scale fluctuations based on the time-frequency joint distribution feature matrix specifically includes: The time-frequency joint distribution feature matrix is input into a first-level convolution layer to extract macro-rhythm features of the time-frequency joint distribution feature matrix by using 64 convolution kernels with a size of 16; The macro-rhythm features are captured to medium-scale features by using 128 convolution kernels with a size of 8 and combining a residual connection structure; The medium-scale features are analyzed to micro-scale fluctuations by using 256 convolution kernels with a size of 4.

[0010] Optionally, in the index calculation module, the multi-dimensional sleep quality evaluation indexes include a sleep efficiency, a deep sleep proportion, a REM cycle stability and a delta wave power attenuation rate.

[0011] Optionally, the content of the delta wave power attenuation rate is as follows: wherein, denotes a time point, denotes delta a frequency band power spectrum density value, and denote a mean value of a time sequence and a power value respectively, M denotes a number of data points in a fitting window.

[0012] The application further discloses a dynamic sleep monitoring and sleep cycle guidance planning method, which comprises the following steps: Step S1, synchronously collecting signal data of a subject by using a multi-channel biological sensor; Step S2, performing real-time frequency band decomposition on the signal data to extract power spectrum density features of brain waves; Step S3, segmenting the signal data, constructing a time-frequency joint distribution feature matrix based on power spectrum density features corresponding to a segmentation result, and performing segmentation of a sleep stage based on the time-frequency joint distribution feature matrix; Step S4, based on a sleep stage classification result, calculating multi-dimensional sleep quality evaluation indexes in real time, and realizing dynamic sleep monitoring based on the multi-dimensional sleep quality evaluation indexes; Step S5, receiving user descriptions and converting the user descriptions into constraint conditions by semantic understanding, generating guidance instructions based on a dynamic sleep monitoring result and the constraint conditions, and completing sleep cycle guidance planning based on the guidance instructions.

[0013] Compared with the prior art, the application has the following beneficial effects: Compared with the prior art, the application significantly improves the response speed and accuracy of sleep regulation through the innovative closed-loop monitoring guidance mechanism, realizes efficient cooperation from data acquisition to intervention decision, and achieves industry-leading levels of sleep stage recognition accuracy and cycle prediction reliability, effectively prolongs deep sleep time and reduces sleep interruption problems, and the system has continuous self-optimization capability and can dynamically adapt to individual sleep mode changes, providing a breakthrough solution for sleep health management. BRIEF DESCRIPTION OF DRAWINGS

[0014] In order to more clearly illustrate the technical solutions of the present application, the following briefly introduces the drawings needed in the embodiments. Obviously, the drawings described in the following are only some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without creative labor.

[0015] Figure 1 The system overall flow principle diagram provided for the embodiments of the present application; Figure 2 The CNN-SVM classification model architecture diagram in the system provided for the embodiments of the present application; Figure 3 The LSTM prediction engine time sequence processing schematic diagram in the system provided for the embodiments of the present application; Figure 4 The demand accurate guidance module schematic diagram in the system provided for the embodiments of the present application; Figure 5 The method step diagram of the method of dynamic sleep monitoring and sleep cycle guidance planning provided for the embodiments of the present application. DETAILED DESCRIPTION

[0016] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the present application will be further described in detail below with reference to the drawings and specific embodiments.

[0017] Embodiment one: A dynamic sleep monitoring and sleep cycle guidance planning system, as shown in Figure 1 The system comprises: A data acquisition module for synchronously acquiring signal data of a subject through a multi-channel biosensor.

[0018] The brain electrical, electrocardiogram, respiration and body movement signals of the subject are synchronously acquired through an integrated biosensor array. The brain electrical signal acquisition adopts an 8-channel dry electrode cap covering the key points of the forehead area, central area and occipital area, and the sampling frequency is set to 256Hz. The electrocardiogram and respiration signals are acquired through flexible conductive patches, and a three-axis accelerometer simultaneously monitors the body movement intensity in real time.

[0019] The feature extraction module is configured to perform real-time frequency band decomposition on the signal data to extract power spectral density features of the brain waves. The collected electroencephalogram signal is immediately subjected to frequency band decomposition processing to extract power spectral features of alpha waves (8-12 Hz), beta waves (12-30 Hz), theta waves (4-8 Hz), and delta waves (0.5-4 Hz), thereby providing basic data support for subsequent analysis.

[0020] The sleep stage classification module is configured to divide the signal data into sleep stages. The six-stage sleep in the embodiment strictly follows the classic Rechtschaffen & Kales (R&K) standard, i.e., divided into: wakefulness period, non-rapid eye movement sleep 1-4 period, and rapid eye movement sleep period. The power spectral features obtained by the feature extraction module are used to construct a time-frequency joint distribution feature matrix, and based on the time-frequency joint distribution feature matrix, other physiological signal features are fused to perform six-stage sleep staging.

[0021] The continuous signal stream is segmented in 30-second time windows to generate a sequence of standardized data frames. The continuous wavelet transform is performed on the electroencephalogram signal of each data frame to construct a time-frequency joint distribution feature matrix. The feature matrix is input into a three-level convolutional neural network architecture: the first convolutional layer uses 64 convolutional kernels with a size of 16 to extract macro-rhythm features; the second convolutional layer uses 128 convolutional kernels with a size of 8 and combines a residual connection structure to capture mesoscale features; and the third convolutional layer uses 256 convolutional kernels with a size of 4 to analyze microscale fluctuations. After fusing the RR interval variation features of the electrocardiogram signal and the harmonic distortion rate of the respiration signal, the six-stage sleep labels are output by a support vector machine classifier to realize millisecond-level dynamic recognition of the sleep state.

[0022] The index calculation module is configured to calculate multi-dimensional sleep quality evaluation indexes in real time based on the sleep stage classification results, and to realize dynamic sleep monitoring based on the multi-dimensional sleep quality evaluation indexes.

[0023] The multi-dimensional sleep quality evaluation indexes are calculated in real time. The core indexes include sleep efficiency (Sleep Efficiency), deep sleep ratio (Deep Sleep Ratio), REM cycle stability (REM Cycle Stability), and delta wave power attenuation rate (Delta Wave Power Attenuation Rate).

[0024] (1) (2) (3) (4) (5) ​​Formula 1 is a calculation formula of sleep efficiency, i.e. the percentage of total sleep time to time in bed, wherein represents total sleep time (unit: minute), represents time in bed (unit: minute).

[0025] Formula 2 is the percentage of deep sleep, i.e. the percentage of the duration of deep sleep stages S3 and S4 to the total sleep time, wherein, and represent the time of sleep stages S3 and S4, respectively, represents total sleep time (same as formula 1).

[0026] Formula 3 is the REM cycle stability, i.e. the standard deviation of the interval of the start time of adjacent REM periods, wherein represents the start time of the i-th REM sleep period (unit: second), represents the mean of the interval of adjacent REM periods, which is calculated as shown in formula 4, N represents the total number of REM cycles detected in a single sleep, and REM cycle stability reflects the regularity of REM sleep, and the lower the value, the higher the stability.

[0027] Formula 5 is the δ wave power attenuation rate, i.e. the slope of the decline of δ band energy over time, wherein, represents the time point (unit: second), with a continuous time window (30 seconds / frame) as the sampling point, represents the δ band power spectral density value (unit: μV² / Hz), which is obtained by real-time frequency band decomposition of electroencephalogram (S2), and represent the mean of the time series and the power value, respectively, and M in the formula represents the number of data points in the fitting window. The attenuation rate can reflect the decay speed of deep sleep intensity.

[0028] All indicators are dynamically fitted by the least squares method, refreshed every minute and presented in real time through a visual interface, providing users with intuitive sleep quality evaluation.

[0029] The demand instruction generation module is used to receive user descriptions and convert them into constraint conditions through semantic understanding, generate guidance instructions based on dynamic sleep monitoring results and the constraint conditions, and complete sleep cycle guidance planning based on the guidance instructions.

[0030] The historical sleep stage sequence is encoded into a one-hot vector time series data stream, which is input into a long short-term memory prediction model for time series analysis. The model integrates an attention weighting mechanism, dynamically focusing on key historical state nodes, and outputs the probability distribution of each sleep stage in the next 30 minutes. When the conversion probability of non-rapid eye movement sleep N3 period to REM period exceeds a set threshold, or the wakefulness probability increases significantly, the system automatically activates a phase change warning mechanism to provide decision basis for sleep cycle intervention.

[0031] The system receives a user's natural language description of sleep optimization needs (such as "extend deep sleep time"), and converts it into a mathematical constraint condition through a semantic analysis engine. A dynamic matching degree evaluation model of sleep state and target needs is constructed, and when real-time monitoring data simultaneously meet multiple conditions such as insufficient matching degree, excessive key period phase transition probability, and excessive user demand weight, a guiding instruction for θ / δ band neural oscillation is generated.

[0032] (6) (7) (8) (9) wherein, formula 6 represents a sleep state-target matching degree function, which is the main calculation formula in the dynamic matching degree evaluation of sleep state and target needs. Among them, represents the demand weight, , represents a dynamic normalization function, and the final output value is in [0, 1].

[0033] Formula 7 is a real-time index evaluation function, wherein, and are index weight coefficients, satisfying ; The negative exponential function of delta wave power attenuation rate is used to reflect that the faster the δ wave attenuates, the worse the deep sleep quality is; reflects the stability of the sleep cycle, and through this function, the influence factor of the index on the sleep target can be obtained to participate in decision-making.

[0034] Formula 8 is a prediction data evaluation function, wherein and are sigmoid steepness factors (default λ=μ=10), which control the sensitivity of probability mutation, is the wake-up probability baseline, whose value is the statistical standard deviation of the recent 60-minute wake-up probability, and are the N3 period to REM period conversion probability and the wake-up warning probability obtained at the LSTM output layer, respectively, and in this function, >80% or , the value drops sharply, the sleep stage transition and the wake-up probability increase, reflecting the influence of the prediction value on the demand decision.

[0035] Formula 9 is a constraint condition satisfaction function, wherein The mathematical constraints transformed for each natural language requirement, if the i-th natural language requirement is satisfied, then The value is 1 if it is 1, otherwise it is 0.

[0036] The dynamic matching degree assessment model calculates real-time sleep indicators, prediction data, and user needs, uses these as input parameters, and obtains results through an evaluation formula to influence decision triggers and generate intervention instructions, reflecting a holistic guidance and planning mechanism that integrates indicators, predictions, and user needs.

[0037] A closed-loop optimization mechanism is established to continuously improve system performance. The support vector machine classifier performs incremental learning to update the decision boundary every 24 hours, and the long short-term memory prediction model updates the network weight parameters every 72 hours. The system automatically records the changes in state matching degree before and after guidance, and dynamically optimizes the guidance parameter configuration based on the reinforcement learning framework to ensure that the system continuously adapts to the evolution of the user's sleep pattern.

[0038] Example 2: like Figure 1 The diagram shows the overall flowchart of this invention. First, the subject's EEG, ECG, respiration, and body movement signals are simultaneously collected using a multi-channel biosensor. The EEG signals are decomposed in real time to extract power spectral density features. After the multimodal signal stream is segmented into data frames within a fixed 30-second time window, the data frames are input into a CNN-SVM cascaded model for sleep stage classification. Simultaneously, an LSTM time-series prediction engine is used to predict the future sleep cycle evolution path. A dynamic decision matrix is ​​constructed based on the user's sleep optimization needs. When the triggering conditions are met, a guided intervention strategy is generated to achieve closed-loop optimization and regulation of sleep structure.

[0039] like Figure 2 The diagram illustrates the specific process of the sleep stage classification module of this invention. After receiving a 30-second data frame of multimodal signal, continuous wavelet transform is performed on the EEG signal to generate a time-frequency joint distribution feature matrix. This matrix is ​​then input into a pre-trained four-layer convolutional neural network to extract features. Simultaneously, the standard deviation of the RR interval of the ECG signal, the harmonic distortion rate of the respiratory signal, and the acceleration variance of the body motion signal are calculated. After fusing EEG and other modal features, the feature vector is input into a support vector machine classifier using a radial basis function kernel. An adaptive weight adjustment mechanism is used to solve the sample imbalance problem, and finally, the six-stage sleep classification result is output with a classification delay of less than 50 milliseconds.

[0040] Convolutional neural networks employ transfer learning for optimization: The network's underlying parameters are pre-trained based on the Sleep-EDF public dataset. The network structure includes three feature extraction modules: Module 1 consists of 64 convolutional kernels of size 16 with a stride of 4; Module 2 consists of 128 convolutional kernels of size 8 with added cross-layer residual connections; Module 3 consists of 256 convolutional kernels of size 4. Finally, a 128-dimensional feature vector is output through a global average pooling layer. The training process uses an adaptive gradient clipping technique to limit the gradient norm to no more than 0.5 to prevent overfitting.

[0041] like Figure 3 The diagram shows the flow of the sleep cycle prediction module of this invention. The historical sleep sequence processing unit is responsible for encoding the historical sleep stage sequence into a one-hot vector time-series stream. The LSTM prediction engine unit dynamically focuses on key historical states through a long short-term memory network with 128 hidden units combined with an attention weight allocation module. The feature weighted fusion unit performs weighted fusion of the LSTM output features and the fully connected layer features, and finally outputs the probability distribution function of each sleep stage in the next 30 minutes.

[0042] like Figure 4 The diagram illustrates the flow of the planning and guidance module of this invention. The real-time indicator receiver receives real-time sleep indicators such as delta wave power attenuation rate and REM cycle stability. The prediction data receiver receives sleep stage prediction data. The user demand parser transforms user demands into mathematical constraints. The multi-source decision-making collaboration unit constructs a dynamic guidance decision matrix based on the aforementioned multi-source data. The trigger condition judgment unit determines whether to generate an intervention strategy based on preset conditions. When the trigger conditions are met, the intervention strategy generator generates guidance instructions for the theta / delta frequency band neural oscillations and implements the intervention through the acoustic module or other output modules. The construction and use of the dynamic guidance decision matrix are as follows: (10) (11) (12) First, by integrating real-time sleep metrics, predicted data, and user needs, the matrix representation in Formula 10 is obtained. Then, the weights in the sleep dynamic matching function are used to calculate the weighted decision matrix in Formula 10 for each row of F. Matrix T in Formula 12 is the decision factor used as the trigger condition. T is then compared with... By splicing , which serves as the final decision matrix.

[0043] If and only if When the intervention condition is triggered, the intervention strategy "θ / δ band neural oscillation modulation" is given; if and only if When T is the intervention condition, the intervention condition is triggered, and the intervention strategy "deep sleep enhancement" is given; when T is other conditions, normal detection is maintained.

[0044] Embodiment three: A dynamic sleep monitoring and sleep cycle guidance planning method, as shown in Figure 5 The method comprises the following steps: Step S1, synchronously collecting signal data of a subject through a multi-channel biosensor.

[0045] Step S2, performing real-time frequency band decomposition on the signal data, and extracting power spectral density features of brain waves.

[0046] Step S3, segmenting the signal data, constructing a time-frequency joint distribution feature matrix based on the power spectral density features corresponding to the segmentation results, and performing segmentation of sleep stages based on the time-frequency joint distribution feature matrix.

[0047] The sleep stage classification is realized through the following process: performing continuous wavelet transform on the electroencephalogram signal of each data frame to generate a time-frequency joint distribution feature matrix, inputting the matrix into a pre-trained four-layer convolutional neural network; fusing the RR interval standard deviation of the electrocardiogram signal, the harmonic distortion rate of the respiratory signal, and the acceleration variance feature of the body motion signal; inputting the fusion feature vector into a support vector machine classifier using a radial basis kernel function, solving the sample imbalance problem through an adaptive weight adjustment mechanism, and finally outputting the sleep stage classification result, with a classification delay of less than 50 milliseconds, and updating the decision hyperplane every 24 hours to adapt to individual physiological changes.

[0048] Step S4, based on the sleep stage classification result, real-time calculation of multi-dimensional sleep quality evaluation indexes, and realization of dynamic sleep monitoring based on the multi-dimensional sleep quality evaluation indexes.

[0049] The sleep quality evaluation indexes include two types of basic indexes and advanced dynamic indexes: The basic indexes include sleep efficiency (total sleep time / total bed time x 100%) and deep sleep proportion (S3+S4 duration / total sleep time x 100%); The advanced dynamic indexes include REM cycle stability (standard deviation of the interval between adjacent REM period start times), delta wave power attenuation rate (slope of the power-time curve in the delta frequency band), and stage transition entropy (quantification of sleep stage transition chaos based on Shannon entropy); All indexes are refreshed every 60 seconds and displayed in real time through a visual interface, and the delta wave power attenuation rate is calculated by fitting the power-time curve using the least squares method.

[0050] Step S5, receiving user description and converting into constraint conditions through semantic understanding, generating guidance instructions based on dynamic sleep monitoring results and the constraint conditions, and completing sleep cycle guidance planning based on the guidance instructions.

[0051] Sleep cycle prediction is achieved by the following techniques: The historical sleep stage sequence is encoded into a one-hot vector time series stream, which is input into a long short-term memory network prediction engine containing 128 hidden units. The network integrates an attention weight distribution module to dynamically focus on key historical states. The output is a probability distribution function for each sleep stage within the next 30 minutes. When the probability of N3 to REM phase transition is greater than 80% or the probability of wakefulness increases by more than 3 standard deviations from the baseline value, a phase transition tracking warning is activated. The prediction engine uses an asymmetric loss function, imposing a 5 times penalty weight on REM phase prediction errors, and updates the network weights in full every 72 hours to track sleep pattern drift.

[0052] Accurate guidance strategy generation is achieved through multi-source decision-making collaboration: Based on real-time sleep indicators (including delta wave power decay rate, REM cycle stability, stage transition entropy), sleep stage prediction data (especially N3→REM phase transition probability and wakefulness probability increase value), and mathematical constraint conditions generated by user demand analysis (objective function and constraint inequality), a dynamic guidance decision matrix is constructed. When the following conditions are met simultaneously, accurate intervention guidance is triggered: (a) sleep state-target matching degree M(t)<0.7; (b) N3→REM phase transition probability>80% and delta wave power decay rate>0.2μV / s; (c) "suppress abnormal phase transition" coefficient in user demand weight matrix>0.5; When one or several of the above conditions are met, guidance strategies targeting different effects are given on the computer interface, and a demand-driven closed-loop system is realized through intervention strategies.

[0053] The above-described embodiments are only descriptions of the preferred modes of the present application and do not limit the scope of the present application. Without departing from the design spirit of the present application, various modifications and improvements to the technical solutions of the present application made by those of ordinary skill in the art shall fall within the scope of protection determined by the claims of the present application.

Claims

1. A system for dynamic sleep monitoring and sleep cycle guidance programming, the system comprising: The system comprises: a data acquisition module for synchronously acquiring signal data of a subject through a multi-channel biosensor; a feature extraction module for real-time frequency band decomposition of the signal data to extract power spectral density features of brain waves; a sleep stage classification module for segmenting the signal data, constructing a time-frequency joint distribution feature matrix based on power spectral density features corresponding to the segmentation results, and segmenting sleep stages based on the time-frequency joint distribution feature matrix; an index calculation module for real-time calculation of multi-dimensional sleep quality evaluation indexes based on sleep stage classification results, and dynamic sleep monitoring based on the multi-dimensional sleep quality evaluation indexes; a demand instruction generation module for receiving user descriptions and converting them into constraint conditions through semantic understanding, generating guide instructions based on dynamic sleep monitoring results and the constraint conditions, and completing sleep cycle guide planning based on the guide instructions.

2. The system for dynamic sleep monitoring and sleep cycle guidance programming of claim 1, wherein, In the data acquisition module, the signal data includes electroencephalogram signals, electrocardiogram signals, respiratory signals, and body movement signals; The electroencephalogram signals are collected by an 8-channel dry electrode cap with a sampling frequency of 256 Hz; The electrocardiogram signals and the respiratory signals are obtained through flexible conductive patches; The body movement signals are obtained by real-time monitoring using a three-axis accelerometer.

3. The system for dynamic sleep monitoring and sleep cycle guidance programming of claim 1, wherein, The workflow of the feature extraction module specifically includes: Real-time frequency band decomposition of the signal data to extract power spectral features of alpha waves, beta waves, theta waves, and delta waves, wherein the alpha waves are brain electrical signals of 8-12 Hz, the beta waves are brain electrical signals of 12-30 Hz, the theta waves are brain electrical signals of 4-8 Hz, and the delta waves are brain electrical signals of 0.5-4 Hz.

4. The system for dynamic sleep monitoring and sleep cycle guidance programming of claim 1, wherein, The sleep stage classification module includes a segmentation processing submodule, a matrix construction submodule, and a classification submodule; The segmentation processing submodule is used for segmenting the signal data in time windows of 30 seconds to obtain a sequence of standardized data frames; The matrix construction submodule is used for wavelet transform of the electroencephalogram signals of the standardized data frame sequence, construction of a time-frequency joint distribution feature matrix based on the wavelet transform results and the power spectral density features of the brain waves, and obtaining micro-scale fluctuations based on the time-frequency joint distribution feature matrix; The classification submodule is used for fusing the micro-scale fluctuations with RR interval variation features of the electrocardiogram signals and harmonic distortion rates of the respiratory signals to obtain fusion information, outputting six-stage sleep labels through a support vector machine classifier based on the fusion information, and segmenting sleep stages based on the sleep labels.

5. The system for dynamic sleep monitoring and sleep cycle guidance programming of claim 4, wherein, In the matrix construction submodule, obtaining micro-scale fluctuations based on the time-frequency joint distribution feature matrix specifically includes: Inputting the time-frequency joint distribution feature matrix into a first convolutional layer to extract macro-rhythm features of the time-frequency joint distribution feature matrix using 64 convolutional kernels with a size of 16; Capturing mesoscale features using 128 convolutional kernels with a size of 8 and combining a residual connection structure; Analyzing micro-scale fluctuations using 256 convolutional kernels with a size of 4.

6. The system for dynamic sleep monitoring and sleep cycle guidance programming of claim 1, wherein, In the index calculation module, the multi-dimensional sleep quality evaluation indexes include sleep efficiency, deep sleep proportion, REM cycle stability, and delta wave power attenuation rate.

7. The system for dynamic sleep monitoring and sleep cycle guidance programming of claim 6, wherein, The content of the delta wave power attenuation rate is specifically: wherein, denotes a time point, denotes Delta band power spectral density value, and denote the mean of the time series and power values, respectively, M denotes the number of data points within the fitting window.

8. A method of dynamic sleep monitoring and sleep cycle guidance programming, the method applying the system of any one of claims 1-7, characterized in that, The method comprises: Step S1, synchronously collecting signal data of a subject through a multi-channel biosensor; Step S2, performing real-time frequency band decomposition on the signal data, and extracting power spectrum density features of brain waves; Step S3, segmenting the signal data, constructing a time-frequency joint distribution feature matrix based on the power spectrum density features corresponding to the segmentation results, and performing segmentation of sleep stages based on the time-frequency joint distribution feature matrix; Step S4, based on the sleep stage classification result, real-time calculation of multi-dimensional sleep quality evaluation indexes, and dynamic sleep monitoring based on the multi-dimensional sleep quality evaluation indexes; Step S5, receiving user description and converting it into constraint conditions through semantic understanding, generating guidance instructions based on the dynamic sleep monitoring results and the constraint conditions, and completing sleep cycle guidance planning based on the guidance instructions.

Images