Deep sleep staging management method and system based on electroencephalogram feedback

Abstract

The invention relates to the technical field of deep sleep management, in particular to a deep sleep staging management method and system based on electroencephalogram feedback. Comprising the steps that electroencephalogram feedback of a target user is collected, an actual deep sleep waveform is extracted and judged with an expected spectral domain of an ideal state transfer stage, and an unreached stage is locked; in an unreached stage, constructing a signal mode prediction model, and determining an optimal guide window stage by combining with intervention entry point migration analysis; performing wavelet basis decomposition to obtain an excellent wave band and a clutter band of the optimal guide window period; when the signal-to-noise ratio is high, crosstalk coupling analysis is executed, and white noise is generated by using the rhythm coding library for guide management; when the signal-to-noise ratio is low, light environment and temperature response regulation is planned, and deep sleep is managed in combination with a color coding chain and a signal excitation result. According to the method and the system, guide management of different decisions can be executed in the stage of entering the deep sleep of the target user, and the deep sleep quality, stability and continuity are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of deep sleep management, and in particular to a deep sleep staging management method and system based on electroencephalogram feedback. Background Art

[0002] Deep sleep is a crucial stage in the human sleep cycle, playing a significant role in physiological processes such as memory consolidation, immune function recovery, and metabolic regulation. Studies have shown that insufficient deep sleep quality may lead to cognitive decline, mood disorders, and an increased risk of various chronic diseases. In recent years, with the development of wearable devices and smart home technologies, sleep monitoring and staging methods based on electroencephalogram (EEG) signals have gradually been applied to home health management scenarios. However, traditional deep sleep staging management methods typically rely on polysomnography (PSG) to analyze the frequency domain characteristics of EEG waveforms to determine sleep stages. The calculation steps are cumbersome and the staging judgment results are prone to errors. This makes it difficult to determine the deep sleep stage based on the actual sleep state of different users.

[0003] Secondly, existing management methods usually control smart home sleep devices to change the sleep environment from the moment the user begins to sleep to promote the transition to light sleep, thereby accelerating the onset of deep sleep. However, the creation of a sleep environment throughout the entire process may make it difficult for users with poor sleep to adapt to the current sleep rhythm. For example, continuous blue light exposure during sleep may cause a decrease or disorder in melatonin levels, resulting in a significant decrease in the probability of entering deep sleep. There are also management problems caused by inappropriate intervention timing, which makes it even more difficult for users to stably enter deep sleep mode. In addition, existing methods generally have limitations in excellent wave guidance management for achieving deep sleep mode under noise with different signal-to-noise ratios, as well as poor adaptability and robustness, resulting in unstable deep sleep staging results. Summary of the Invention

[0004] The present invention overcomes the deficiencies of the prior art and provides a deep sleep staging management method and system based on EEG feedback.

[0005] In order to achieve the above object, the technical solution adopted by the present invention is:

[0006] A first aspect of the present invention provides a deep sleep staging management method based on EEG feedback, comprising the following steps:

[0007] S102: Collecting EEG feedback to extract the actual deep sleep waveform of the target user, planning the expected deep sleep spectrum range of the target user to generate a standard deep sleep waveform based on the ideal state transition of the sleep health quality data, and determining whether the actual deep sleep waveform is within the expected deep sleep spectrum range and whether it is locked in the unreached stage;

[0008] S104: If the target user has not yet reached the stage, a signal pattern prediction model is constructed to predict the current signal pattern of the target user who is about to enter deep sleep. The optimal guidance window period for deep sleep intervention is planned based on the intervention entry point migration analysis of the current signal pattern.

[0009] S106: performing filtering processing on the low-frequency components and high-frequency components of the EEG signal within the optimal guidance window period using a wavelet basis decomposition to obtain the excellent band and the noise band to which the real-time sleep waveform adheres within the optimal guidance window period;

[0010] S108: If the signal-to-noise ratio of the excellent band to the clutter band is greater than a preset signal-to-noise ratio, crosstalk coupling analysis is performed to analyze the clutter band acting on the easily disturbed waveform segment of the excellent band, and mask tampering of the easily disturbed waveform segment is checked using a rhythm coding library of a standard waveform to generate white noise audio for guiding management of the deep sleep stage, thereby obtaining a first staged management strategy;

[0011] S110: If the signal-to-noise ratio of the excellent band and the noise band is less than the preset signal-to-noise ratio, the color gamut binning interval of the light environment-temperature response control is planned to obtain a color coding chain for the noise band that deviates from the standard, the signal excitation trend is analyzed to obtain a signal excitation result, and the color coding chain and the signal excitation result are combined to manage the deep sleep stage to obtain a second stage management strategy.

[0012] Preferably, the step S102 specifically includes the following steps:

[0013] Obtaining a deep sleep staging strategy determined by a smart home sleep device for a target user and a pre-designed deep sleep management plan for the smart home sleep device to execute the deep sleep staging strategy;

[0014] By running a predetermined deep sleep management plan on the target user and monitoring the EEG response in real time when using smart home sleep devices, the EEG feedback signal of the target user during sleep is obtained;

[0015] Introducing a short-time Fourier transform algorithm to transform the EEG feedback signal in the time domain and frequency domain to generate an EEG feedback time-frequency spectrum, and extracting the actual deep sleep waveform of the target user from the EEG feedback time-frequency spectrum;

[0016] By obtaining the potential sleep state chain of the established deep sleep management plan based on the standardized sleep staging system, a sleep state hierarchical nested tree is constructed. The sleep state hierarchical nested tree is used to perform state transition trigger inference on sleep health quality data to obtain the ideal state transition path for the target user's deep sleep staging.

[0017] Based on the time domain characteristics and frequency domain characteristics of the standard deep sleep waveform, the sleep state knowledge graph is identified to obtain the oscillation rhythm of the standard deep sleep waveform under different sleep state conditions;

[0018] A naive Bayesian inference model is constructed by calculating the transient prior probability of a standard deep sleep waveform based on a collection of observed parameters of the oscillatory rhythm. The naive Bayesian inference model is used to infer the oscillation sequence points of the ideal state transition path located in the EEG feedback time spectrum, and output a steady-state conditional probability.

[0019] Combining transient prior probability and steady-state conditional probability analysis, we determine the ideal staging points in time and frequency that can produce a standard deep sleep waveform for the target user. Based on these ideal staging points, we plan deep sleep staging on the EEG feedback time-frequency spectrum to obtain the target user's expected deep sleep spectrum.

[0020] If the actual deep sleep waveform does not exist within the expected deep sleep spectrum at this moment, the target user is locked as not reaching the stage.

[0021] Preferably, the method of constructing a sleep state hierarchical nested tree by acquiring a potential sleep state chain for executing a predetermined deep sleep management plan based on a standardized sleep staging system, and using the sleep state hierarchical nested tree to perform state transition triggering reasoning on sleep health quality data to obtain an ideal state transition path for the target user's deep sleep staging specifically includes the following steps:

[0022] Based on big data, we obtain a sleep state knowledge graph and a standardized sleep staging system that meets the target user group. We then simultaneously search and identify the sleep states that occur in a chain reaction when implementing a predetermined deep sleep management plan in accordance with the standardized sleep staging system, and obtain several potential sleep state chains.

[0023] Establish a state stack from the top parent state to the bottom leaf state, determine the contextual logic structure of the state stack based on several potential sleep state chains, and construct a hierarchical nested tree of sleep states based on the contextual logic structure;

[0024] Obtain the target user's sleep health log, preset multiple sets of evenly continuous health monitoring timestamps, and extract the target user's sleep health quality data at each health monitoring timestamp when the target user accepted the established deep sleep management plan through the sleep health log;

[0025] Import the sleep health quality data into the sleep state hierarchical nested tree and search the current state hierarchy to the parent state layer by layer, and traverse and calculate the current trigger matching value between each stage and the state trigger constraint premise proposed by the standardized sleep stage system;

[0026] By determining whether the trigger matching threshold is greater than the trigger matching threshold of each state trigger constraint premise, the transfer entropy weight of the state trigger constraint premise corresponding to the top-level parent state is weightedly calculated in the strategy trade-off algorithm to generate the entry channel and the exit channel;

[0027] Based on the exit logic pattern from the entry channel to the exit channel, the sleep state transition clues under the premise of sleep health quality are calculated to obtain the ideal state transition path of the target user's deep sleep stage.

[0028] Preferably, the step S104 specifically includes the following steps:

[0029] When the target user is locked and marked as not reaching the stage, the EEG management log of the smart home sleep device is obtained. The historical response inflection points of the spectrum feedback when the target user's deep sleep waveform is in EEG feedback during the past time period and the emergence timing of each historical response inflection point are extracted from the EEG management log;

[0030] A multi-layer perceptron model is introduced to construct a variational recognition encoder and variational recognition decoder based on dynamic time series, and the historical response inflection points and corresponding emergence timing data are injected into the variational recognition encoder;

[0031] After the injection, the variational recognition encoder uses a recurrent neural network to analyze the regularity of the patterns of historical response inflection points at the time of emergence. It then uses a gated recurrent unit to track the temporal variation of the patterns at the time of emergence relative to the occurrence of the historical response inflection points. This pattern regularity and temporal variation are then input into the variational recognition decoder for reconstruction and propagation, thereby constructing a signal pattern prediction model for deep sleep waveforms.

[0032] The signal pattern prediction model is used to predict the actual deep sleep waveform of the EEG feedback spectrum, outputting the current signal pattern of the actual deep sleep that is about to enter. The intervention point of the actual deep sleep waveform on the future timeline is determined based on the mode switching rhythm of the current signal pattern, which is defined as the first intervention entry point.

[0033] Obtaining an intervention point in the desired deep sleep spectrum domain, defining it as a second intervention entry point, and calculating a transfer function between the first intervention entry point and the second intervention entry point;

[0034] If the migration function is greater than the preset migration function, the jurisdiction area of ​​the first intervention entry point and the second intervention entry point is marked as the long-term window margin; if it is less than, it is marked as the short-term window margin. The optimal guidance window period for deep sleep intervention of the target user is generated based on the collaborative planning of the long-term window margin and the short-term window margin.

[0035] Preferably, the step S106 specifically includes the following steps:

[0036] Preset the wavelet basis function, and use the wavelet basis function to spectrally decompose the brain waves in the optimal guidance window period during EEG feedback to obtain low-frequency sub-signals and high-frequency sub-signals at different wavelet basis decomposition layers;

[0037] Filtering is performed based on the wavelet coefficients of the low-frequency sub-signal and the high-frequency sub-signal to extract the excellent bands and noise bands to which the real-time sleep wave is attached during the optimal guidance window.

[0038] Preferably, the step S108 specifically includes the following steps:

[0039] If the signal-to-noise ratio between the excellent band and the noise band is greater than the preset signal-to-noise ratio, the rhythm representation template of the output EEG signal corresponding to the standard deep sleep waveform is obtained based on the big data, and the rhythm representation template is encoded using the EEG signal encoding rules to obtain the rhythm encoding library of the standard deep sleep waveform;

[0040] Highlighting the real-time sleep wave in the optimal guidance window period, extracting the position pattern of the excellent band relative to the real-time sleep wave, which is defined as the first subsidiary wave distribution, and extracting the position pattern of the noise band relative to the real-time sleep wave, which is defined as the second subsidiary wave distribution;

[0041] The crosstalk coupling strength of the clutter band on the excellent band is calculated based on the co-frequency feedback data of amplitude and frequency between the excellent band and the clutter band. Based on the second subsidiary wave distribution, only the sub-coupling waveform segments in the first subsidiary wave distribution with crosstalk coupling strength greater than the preset crosstalk coupling strength are marked and defined as the easily disturbed waveform segments of the real-time sleep wave.

[0042] Construct a registration domain, in which the real-time sleep wave is superimposed and masked onto the standard deep sleep waveform for registration. At this time, the mask segment where the easily disturbed waveform segment is located is intercepted from the rhythm coding library to obtain the corrected mask sequence and the actual coding sequence of the easily disturbed coupled waveform segment;

[0043] Obtain the association management mapping table of the sleep wave frequency band and the white noise frequency band for the smart home sleep device, check the tampering mask for missing waveform frequency bands in the actual coding sequence compared to the return mask sequence, search the association management mapping table for the corresponding white noise compensation frequency band based on the tampering mask, and generate a white noise audio control decision;

[0044] The deep sleep stage of the target user is managed according to the white noise audio control decision, and the first phased management strategy is obtained.

[0045] Preferably, the step S110 specifically includes the following steps:

[0046] If the signal-to-noise ratio between the excellent band and the noise band is less than the preset signal-to-noise ratio, a signal peak model diagram of the real-time sleep wave is constructed through the signal oscilloscope modeling software;

[0047] The Wigner-Ville algorithm is introduced to calculate the energy distribution of the clutter band and output the power spectrum density of the clutter band. At the same time, the peak slope of the clutter band is obtained through scalar data analysis of the signal spike model.

[0048] Obtain the control rules for light environment-temperature creation, perform calculus estimation of the distribution probability density for each clutter band based on the power spectrum density located below the signal peak model diagram, and render according to the coding of the control rules to generate the color gamut bin intervals for different clutter bands in response to light environment-temperature control;

[0049] The excitation state trend of the signal in the clutter band is calculated by combining the peak slope with the amplitude and frequency of the clutter band to obtain the excitation steepness.

[0050] If the excitation steepness is greater than the preset excitation steepness, the color gamut bin interval corresponding to the noise band in the signal spike model diagram is marked as an enhancement trend of 1; if the excitation steepness is less than the preset excitation steepness, the color gamut bin interval corresponding to the noise band is marked as an attenuation trend of 0, and the signal excitation result is obtained;

[0051] The phase deviation of the excellent band compared to the standard deep sleep waveform is calculated by registration to obtain the phase drift value. At the same time, the deviation area between the noise band and the standard deep sleep waveform is extracted to obtain the color saturation scale of the deviation area in the color gamut bin interval;

[0052] A guidance control law is designed based on the signal excitation results, and the guidance control law is executed on the smart home sleep device according to the color saturation scale to manage the light environment and temperature creation of the target user in the deep sleep stage until the phase drift value is equal to 0, thereby obtaining the second phase management strategy.

[0053] A second aspect of the present invention provides a deep sleep staging management system based on EEG feedback, which is applied to any of the above-mentioned deep sleep staging management methods based on EEG feedback. The system specifically includes:

[0054] An EEG monitoring module, which is used to monitor the EEG signals of the target user during sleep in real time;

[0055] The EEG feedback module is responsible for feeding back the EEG signals of the target user during sleep and converting the EEG signals into an EEG feedback time-frequency spectrum using a short-time Fourier transform algorithm;

[0056] a data processing module, the data processing module being configured to perform data planning processing on an optimal guidance window period for deep sleep intervention of a target user in coordination with the long-term window margin and the short-term window margin;

[0057] An intelligent analysis module, which is responsible for determining whether the signal-to-noise ratio between the optimal band and the noise band to which the real-time sleep waveform adheres during the optimal guidance window is greater than a preset signal-to-noise ratio, thereby performing strategic analysis on the guidance management of the target user who is about to enter the deep sleep stage;

[0058] The stage management module is used to manage the parameter control of the smart home sleep device and provide corresponding deep sleep stage guidance to the target user.

[0059] The present invention solves the technical defects existing in the background technology, and the beneficial technical effects of the present invention are:

[0060] This invention collects the target user's EEG feedback signals in real time and, in combination with sleep health quality data, constructs an ideal state transition staging model. This model accurately matches the user's actual deep sleep waveform with the desired deep sleep spectrum, and locks in the unreached stage. When detecting the unreached stage, a signal pattern prediction model and intervention entry point migration analysis are used to implement targeted intervention during the optimal guidance window. Wavelet decomposition and filtering are used to separate the optimal and noise bands, and different staging management strategies are selectively implemented based on signal-to-noise ratio differences. When the signal-to-noise ratio is high, white noise intervention is generated through crosstalk coupling analysis and rhythm coding library comparison to achieve rhythmic guidance of the deep sleep waveform. When the signal-to-noise ratio is low, the deep sleep environment is stabilized through light environment and temperature control combined with color coding chains and signal excitation trends. This invention can determine the target user's window stage of deep sleep in the actual sleep state, significantly improving deep sleep quality, extending deep sleep duration, and reducing the disruption of deep sleep rhythms by external interference, thereby stabilizing deep sleep. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, without paying any creative work, they can also obtain drawings of other embodiments based on these drawings.

[0062] Figure 1 A first method flow chart of a deep sleep staging management method based on EEG feedback is shown;

[0063] Figure 2 A second method flow chart of a deep sleep staging management method based on EEG feedback is shown;

[0064] Figure 3 The figure shows a system architecture diagram of a deep sleep staging management system based on EEG feedback. DETAILED DESCRIPTION

[0065] In order to more clearly understand the above-mentioned objects, features and advantages of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that, in the absence of conflict, the embodiments of the present application and the features therein can be combined with each other.

[0066] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Therefore, the scope of protection of the present invention is not limited to the specific embodiments disclosed below.

[0067] The first aspect of the present invention provides a deep sleep stage management method based on EEG feedback, such as Figure 1 As shown, the following steps are included:

[0068] S102: Collecting EEG feedback to extract the actual deep sleep waveform of the target user, planning the expected deep sleep spectrum range for the target user to generate a standard deep sleep waveform based on the ideal state transition of the sleep health quality data, and determining whether the actual deep sleep waveform is within the expected deep sleep spectrum range to lock the unreached stage;

[0069] S104: If the target user has not yet reached the stage, a signal pattern prediction model is constructed to predict the current signal pattern of the target user who is about to enter deep sleep. The optimal guidance window period for deep sleep intervention is planned based on the intervention entry point migration analysis of the current signal pattern.

[0070] S106: performing filtering processing on the low-frequency components and high-frequency components of the EEG signal within the optimal guidance window period using a wavelet basis decomposition to obtain the excellent band and the noise band to which the real-time sleep waveform adheres within the optimal guidance window period;

[0071] S108: If the signal-to-noise ratio of the excellent band to the clutter band is greater than a preset signal-to-noise ratio, crosstalk coupling analysis is performed to analyze the clutter band acting on the easily disturbed waveform segment of the excellent band, and mask tampering of the easily disturbed waveform segment is checked using a rhythm coding library of a standard waveform to generate white noise audio for guiding management of the deep sleep stage, thereby obtaining a first staged management strategy;

[0072] S110: If the signal-to-noise ratio of the excellent band and the noise band is less than the preset signal-to-noise ratio, the color gamut binning interval of the light environment-temperature response control is planned to obtain a color coding chain for the noise band that deviates from the standard, the signal excitation trend is analyzed to obtain a signal excitation result, and the color coding chain and the signal excitation result are combined to manage the deep sleep stage to obtain a second stage management strategy.

[0073] Preferably, the step S102 specifically includes the following steps:

[0074] Obtaining a deep sleep staging strategy determined by a smart home sleep device for a target user and a pre-designed deep sleep management plan for the smart home sleep device to execute the deep sleep staging strategy;

[0075] By running a predetermined deep sleep management plan on the target user and monitoring the EEG response in real time when using smart home sleep devices, the EEG feedback signal of the target user during sleep is obtained;

[0076] Introducing a short-time Fourier transform algorithm to transform the EEG feedback signal in the time domain and frequency domain to generate an EEG feedback time-frequency spectrum, and extracting the actual deep sleep waveform of the target user from the EEG feedback time-frequency spectrum;

[0077] By obtaining the potential sleep state chain of the established deep sleep management plan based on the standardized sleep staging system, a sleep state hierarchical nested tree is constructed. The sleep state hierarchical nested tree is used to perform state transition trigger inference on sleep health quality data to obtain the ideal state transition path for the target user's deep sleep staging.

[0078] Based on the time domain characteristics and frequency domain characteristics of the standard deep sleep waveform, the sleep state knowledge graph is identified to obtain the oscillation rhythm of the standard deep sleep waveform under different sleep state conditions;

[0079] A naive Bayesian inference model is constructed by calculating the transient prior probability of a standard deep sleep waveform based on a collection of observed parameters of the oscillatory rhythm. The naive Bayesian inference model is used to infer the oscillation sequence points of the ideal state transition path located in the EEG feedback time spectrum, and output a steady-state conditional probability.

[0080] Combining transient prior probability and steady-state conditional probability analysis, we determine the ideal staging points in time and frequency that can produce a standard deep sleep waveform for the target user. Based on these ideal staging points, we plan deep sleep staging on the EEG feedback time-frequency spectrum to obtain the target user's expected deep sleep spectrum.

[0081] If the actual deep sleep waveform does not exist within the expected deep sleep spectrum at this moment, the target user is locked as not reaching the stage.

[0082] It should be noted that smart home sleep devices include smart mattresses, smart bracelets, sleep monitoring pillows, sleep-aiding lamps, energy-saving air conditioners and automatic curtains. The standardized sleep staging system is the AASM staging standard system. In this standard sleep staging system, deep sleep belongs to the N3 stage of non-rapid eye movement sleep, which is usually entered after the transition from the N2 light sleep stage. The staging of deep and light sleep usually depends on the actual sleep state of the target user. For example, if the target user has frequent insomnia recently and has poor sleep quality, the delta wave in the deep sleep stage will be delayed or difficult to appear. However, the existing methods cannot make an idealized staging of the deep sleep stage of the target user in advance according to different sleep states, making it difficult to judge whether the actual sleep wave of the current EEG feedback has reached the deep sleep stage, which makes it easy for smart home sleep devices to miss the best staging management opportunity for entering deep sleep. To this end, this method retrieves and identifies possible sleep states with full-process dependencies by following a standardized sleep staging system through a sleep state knowledge graph retrieval. For example, if a deep sleep management plan is decided to dim the room lights, the target user's melatonin level will surge, resulting in a process of development from a shallow sleep state to a deep sleep state, thus forming a potential transition to different sleep states. Therefore, this method constructs it into a hierarchical nested tree of sleep states guided by comprehensive clues to further traverse and search the target user's sleep health quality data, thereby exploring the actual sleep transition states that may occur, and providing an accurate and highly reliable planning reasoning basis for the subsequent optimal deep sleep staging of the target user.

[0083] It should be noted that the basic flash probability of the standard deep sleep waveform when the target user's actual sleep state feedback is not perceived, that is, the transient prior probability, is calculated through the collection of observed parameters of the oscillation rhythm, thereby defining the initial possibility that the target user can enter the deep sleep stage under the premise of unclear sleep quality, and then constructing a naive Bayesian inference network that can capture the deep sleep rhythm characteristics in terms of state, avoiding the inference bias caused by relying solely on state feature information. The naive Bayesian inference model is then used to estimate the fluctuations in the oscillation sequence points during each sequential sleep state transition along the ideal state transition path, thereby calculating the probability of reaching values ​​that correspond to deep sleep rhythm characteristics during the target user's actual sleep state, namely the steady-state conditional probability. In other words, this steady-state conditional probability indicates the characteristic distribution of deep sleep waveforms captured by the naive Bayesian inference model along the ideal state transition path, reflecting the regular fitting nodes that lead to the entry into deep sleep stages from one ideal sleep state to the next ideal sleep state under actual sleep conditions. Therefore, combining the analysis of transient prior probability and steady-state conditional probability can further determine the ideal stage landing point in time and frequency that can produce a standard deep sleep waveform for the target user. Sleep stages are then planned based on the ideal stage landing point, obtaining the expected time-frequency location of the waveform when the target user enters deep sleep during their actual sleep state, namely the expected deep sleep spectral domain. If the actual deep sleep waveform does not currently exist within the expected deep sleep spectral domain, it indicates that the target user has not currently entered a deep sleep stage. This method can infer the ideal stage of deep sleep for the user based on the actual sleep situation, thereby analyzing whether the actual waveform of EEG feedback has reached the ideal deep sleep stage, improving the timing control accuracy of deep sleep staging management, and ensuring a smooth transition from light sleep to deep sleep.

[0084] Preferably, the method of constructing a sleep state hierarchical nested tree by acquiring a potential sleep state chain for executing a predetermined deep sleep management plan based on a standardized sleep staging system, and using the sleep state hierarchical nested tree to perform state transition triggering reasoning on sleep health quality data to obtain an ideal state transition path for the target user's deep sleep staging specifically includes the following steps:

[0085] Based on big data, we obtain a sleep state knowledge graph and a standardized sleep staging system that meets the target user group. We then simultaneously search and identify the sleep states that occur in a chain reaction when implementing a predetermined deep sleep management plan in accordance with the standardized sleep staging system, and obtain several potential sleep state chains.

[0086] Establish a state stack from the top parent state to the bottom leaf state, determine the contextual logic structure of the state stack based on several potential sleep state chains, and construct a hierarchical nested tree of sleep states based on the contextual logic structure;

[0087] Obtain the target user's sleep health log, preset multiple sets of evenly continuous health monitoring timestamps, and extract the target user's sleep health quality data at each health monitoring timestamp when the target user accepted the established deep sleep management plan through the sleep health log;

[0088] Import the sleep health quality data into the sleep state hierarchical nested tree and search the current state hierarchy to the parent state layer by layer, and traverse and calculate the current trigger matching value between each stage and the state trigger constraint premise proposed by the standardized sleep stage system;

[0089] By determining whether the trigger matching value is greater than the trigger matching threshold of each state trigger constraint premise, the transfer entropy weight of the state trigger constraint premise corresponding to the top-level parent state is weightedly calculated in the strategy trade-off algorithm to generate the entry channel and exit channel;

[0090] Based on the exit logic pattern from the entry channel to the exit channel, the sleep state transition clues under the premise of sleep health quality are calculated to obtain the ideal state transition path of the target user's deep sleep stage.

[0091] It should be noted that by hierarchically organizing all states in the potential sleep state chain according to the development trajectory from the top-level parent state (a certain sleep state) to the base-level leaf state (the next possible sleep state), the contextual logic information of the structured transition of the potential state is clarified. This allows the constructed sleep state hierarchical nested tree to share common behaviors and conditional judgments for similar potential sleep states, reducing the reuse of state definitions and making the exploration of idealized states based on actual sleep conditions and in accordance with the standardized system of deep sleep staging more accurate. Subsequently, by importing sleep health quality data, the state hierarchy of the sleep state hierarchical nested tree is searched layer by layer towards the parent state to find the base-level leaf state that can currently be triggered by different sleep health quality data. This ensures that the reasoning of state transitions is based on state logic conditions that can be satisfied by actual sleep conditions. The state trigger constraint premise is the trigger premise of each stage of the sleep staging criteria in the standardized sleep staging system. The current trigger match value defines the degree of adaptation of the state logic corresponding to the deep sleep stage staging criteria that can be triggered by sleep health quality data. If the trigger match value is greater than the trigger match threshold of the state trigger constraint premise, it indicates that the target user is very likely to have a corresponding sleep state in that sleep stage when a certain sleep health quality data appears. Therefore, the transfer priority of the top-level parent state is further weighted, that is, the transfer entropy weight, to generate a sleep state transfer channel from the beginning (entering the channel) to the end (leaving the channel) that continuously fits the target user's sleep health quality data. Finally, based on the exit logic pattern from the entry channel to the exit channel, the sleep transfer promotion clues that may occur under the standardized sleep staging system under the actual sleep conditions of the user can be determined, thereby improving the interpretability of the optimal deep sleep staging for different target users.

[0092] Preferably, the S104, as Figure 2 As shown, the specific steps include:

[0093] When the target user is locked and marked as not reaching the stage, the EEG management log of the smart home sleep device is obtained. The historical response inflection points of the spectrum feedback when the target user's deep sleep waveform is in EEG feedback during the past time period and the emergence timing of each historical response inflection point are extracted from the EEG management log;

[0094] A multi-layer perceptron model is introduced to construct a variational recognition encoder and variational recognition decoder based on dynamic time series, and the historical response inflection points and corresponding emergence timing data are injected into the variational recognition encoder;

[0095] After the injection, the variational recognition encoder uses a recurrent neural network to analyze the regularity of the patterns of historical response inflection points at the time of emergence. It then uses a gated recurrent unit to track the temporal variation of the patterns at the time of emergence relative to the occurrence of the historical response inflection points. This pattern regularity and temporal variation are then input into the variational recognition decoder for reconstruction and propagation, thereby constructing a signal pattern prediction model for deep sleep waveforms.

[0096] The signal pattern prediction model is used to predict the actual deep sleep waveform of the EEG feedback spectrum, outputting the current signal pattern of the actual deep sleep that is about to enter. The intervention point of the actual deep sleep waveform on the future timeline is determined based on the mode switching rhythm of the current signal pattern, which is defined as the first intervention entry point.

[0097] Obtaining an intervention point in the desired deep sleep spectrum domain, defining it as a second intervention entry point, and calculating a transfer function between the first intervention entry point and the second intervention entry point;

[0098] If the migration function is greater than the preset migration function, the jurisdiction area of ​​the first intervention entry point and the second intervention entry point is marked as the long-term window margin; if it is less than, it is marked as the short-term window margin. The optimal guidance window period for deep sleep intervention of the target user is generated based on the collaborative planning of the long-term window margin and the short-term window margin.

[0099] It should be noted that when the target user's EEG feedback signal is not in deep sleep, the deep sleep waveform feedback's historical response inflection points and corresponding emergence timing data are obtained. A historical response inflection point is a sudden change in the target user's deep sleep waveform from light sleep, with a significant and sustained rise in the waveform curve. The emergence timing data represents the time when these sudden inflection points occurred. A variational recognition encoder and decoder are constructed using a multi-layer perceptron model. The variational recognition encoder integrates a recurrent neural network and a gated recurrent unit (GRU), which can capture complex and nonlinear pattern relationships in deep sleep waveforms and improve the abstract power of feature representation. The RNN analyzes the regularity of historical response inflection points at emergence timing, revealing temporal dependency chains and recurring patterns between inflection points. This allows the prediction model to identify characteristic changes in the signal pattern as the target user is about to enter deep sleep, providing a training foundation for capturing long-range dependencies. The GRU tracks the temporal regularity of emergence timing with respect to historical response inflection points, accurately capturing the dynamic characteristic changes in the signal pattern as historical response inflection points shift with emergence timing. Finally, the regular features (pattern regular features) and time features (pattern time series change features) are fused and input into the variational recognition decoder for reconstruction training, so that the signal pattern prediction model can retain the global pattern trend and reflect subtle changes in time series, thereby outputting high-fidelity prediction results, making the signal pattern prediction model more robust in predicting the signal pattern trend when the actual waveform enters the deep sleep stage.

[0100] It should be noted that the first intervention entry point represents the predicted time-frequency drop point at which the actual deep sleep waveform is about to enter the deep sleep stage under the current signal pattern, and the second intervention entry point represents the ideal time-frequency drop point for entering the desired deep sleep spectrum. The migration function reflects the signal wave span required to enable the actual deep sleep waveform to reach the second intervention entry point from the first intervention entry point, reflecting the management buffer period for the actual sleep condition to enter the ideal deep sleep stage. If the migration function is greater than the preset migration function, it indicates that the signal wave span distance from the first intervention entry point to the second intervention entry point is large, and a relatively longer management buffer time is required from the first intervention entry point to reach the second intervention entry point. This indicates that the intervention guidance timing for deep sleep within this time-frequency region should be advanced, and this jurisdiction area is marked as a long-time window margin. Otherwise, it indicates that the signal wave span distance between the two is small, the management buffer time is short, and this jurisdiction area is marked as a short-time window margin. This method can predict the signal patterns of the impending deep sleep stage based on the target user's actual sleep status, thereby planning the optimal guidance window for deep sleep intervention. The optimal guidance window is an idealized deep sleep guidance stage with the best timing (optimal) after the user enters sleep. The optimality emphasizes the best objective preference and highly appropriate node. This achieves the effect of early management of deep sleep entry, ensuring that staged management provides sufficient adjustment buffer while avoiding missing the optimal intervention opportunity, ensuring the achievement and stability of the deep sleep state, and improving deep sleep comfort.

[0101] Preferably, the step S106 specifically includes the following steps:

[0102] Preset the wavelet basis function, and use the wavelet basis function to spectrally decompose the brain waves in the optimal guidance window period during EEG feedback to obtain low-frequency sub-signals and high-frequency sub-signals at different wavelet basis decomposition layers;

[0103] Filtering is performed based on the wavelet coefficients of the low-frequency sub-signal and the high-frequency sub-signal to extract the excellent bands and noise bands to which the real-time sleep waveform adheres within the optimal guidance window.

[0104] It should be noted that the excellent band is an effective focus band that can enable the target user to highly concentrate and enter the deep sleep stage; the noise band is a background noise band that can easily distract the target user's sleep concentration and cause them to deviate from the deep sleep stage.

[0105] Preferably, the step S108 specifically includes the following steps:

[0106] If the signal-to-noise ratio between the excellent band and the noise band is greater than the preset signal-to-noise ratio, the rhythm representation template of the output EEG signal corresponding to the standard deep sleep waveform is obtained based on the big data, and the rhythm representation template is encoded using the EEG signal encoding rules to obtain the rhythm encoding library of the standard deep sleep waveform;

[0107] Highlighting the real-time sleep wave in the optimal guidance window period, extracting the position pattern of the excellent band relative to the real-time sleep wave, which is defined as the first subsidiary wave distribution, and extracting the position pattern of the noise band relative to the real-time sleep wave, which is defined as the second subsidiary wave distribution;

[0108] The crosstalk coupling strength of the clutter band on the excellent band is calculated based on the co-frequency feedback data of amplitude and frequency between the excellent band and the clutter band. Based on the second subsidiary wave distribution, only the sub-coupling waveform segments in the first subsidiary wave distribution with crosstalk coupling strength greater than the preset crosstalk coupling strength are marked and defined as the easily disturbed waveform segments of the real-time sleep wave.

[0109] Construct a registration domain, in which the real-time sleep wave is superimposed and masked onto the standard deep sleep waveform for registration. At this time, the mask segment where the easily disturbed waveform segment is located is intercepted from the rhythm coding library to obtain the corrected mask sequence and the actual coding sequence of the easily disturbed coupled waveform segment;

[0110] Obtain the association management mapping table of the sleep wave frequency band and the white noise frequency band for the smart home sleep device, check the tampering mask for missing waveform frequency bands in the actual coding sequence compared to the return mask sequence, search the association management mapping table for the corresponding white noise compensation frequency band based on the tampering mask, and generate a white noise audio control decision;

[0111] The deep sleep stage of the target user is managed according to the white noise audio control decision, and the first phased management strategy is obtained.

[0112] It should be noted that if the signal-to-noise ratio between the superior band and the noise band is greater than the preset signal-to-noise ratio, this indicates that the background noise energy is low, the delta wave activity in deep sleep is relatively high, and the target user has a clear tendency to rely on the real-time sleep waveform to enter the deep sleep stage, with only localized occasional interference. To further improve the activity smoothness of the real-time sleep waveform before entering the deep sleep stage, this method obtains a rhythm representation template of the standard deep sleep waveform. This rhythm representation template is a regular template that expresses the electrical activity pattern of the standard deep sleep waveform within a specific range. This rhythm representation template is further encoded into EEG signals to construct a rhythm encoding library of standardized templates for subsequent mask queries. Next, using the real-time sleep waveform as a reference, the relative distribution patterns of the superior band and the noise band on the real-time sleep waveform are extracted, namely the first and second subsidiary wave distributions. This can clarify the actual rhythm distribution relationship between the superior band and the noise band. Since co-frequency bands represent a high degree of crosstalk coupling between the clutter band and the superior band, the co-frequency feedback data of amplitude and frequency is used to calculate the crosstalk coupling strength of the clutter band on the superior band. The crosstalk coupling strength reflects the clutter band's ability to suppress the signal expression of the superior band. If the crosstalk coupling strength is greater than the preset crosstalk coupling strength, it indicates that the clutter band is significantly interfering with the superior band. Combining the distribution patterns of the second and first patterns, the real-time sleep waveform segment corresponding to the clutter band's gain reduction to the superior band is identified, i.e., the easily disturbed waveform segment.

[0113] It's important to note that these easily disturbed waveform segments are sensitive waveform segments that, when disturbed by clutter, can easily cause the real-time sleep waveform, which is maintaining a deep sleep stage, to deviate from its intended path. This can make it difficult for the target user to enter the deep sleep stage. Therefore, further eliminating these cluttered segments is necessary to guide the correct extension of the optimal waveform segments. To address this, this method overlays and masks the real-time sleep waveform onto a standard deep sleep waveform in the registration domain. Using the standard deep sleep waveform as the target constraint, a standardized coding sequence, known as a return mask sequence, is masked out that should be maintained for the easily disturbed waveform segments. This return mask sequence reveals the waveform correction answer that ensures entry into deep sleep mode when the real-time sleep waveform is disturbed by clutter. The difference between the actual coding sequence and the return mask sequence is fed back through the missing tampering mask symbol. The tampering mask symbol is the compensation basis for eliminating the clutter band and guiding the excellent band to maintain the correct trend. Therefore, based on the tampering mask, a white noise audio control decision is established to generate sleep inertia and promote a stable transition to deep sleep mode to generate sound stimulation management. Before entering deep sleep, the white noise audio is played through the smart home sleep device to stabilize the target user's delta wave toward the standard deep sleep waveform activity, stimulate and guide them to enter the deep sleep stage faster and more smoothly, and significantly improve the stability and continuity of deep sleep stage management.

[0114] Preferably, the step S110 specifically includes the following steps:

[0115] If the signal-to-noise ratio between the excellent band and the noise band is less than the preset signal-to-noise ratio, a signal peak model diagram of the real-time sleep wave is constructed through the signal oscilloscope modeling software;

[0116] The Wigner-Ville algorithm is introduced to calculate the energy distribution of the clutter band and output the power spectrum density of the clutter band. At the same time, the peak slope of the clutter band is obtained through scalar data analysis of the signal spike model.

[0117] Obtain the control rules for light environment-temperature creation, perform calculus estimation of the distribution probability density for each clutter band based on the power spectrum density located below the signal peak model diagram, and render according to the coding of the control rules to generate the color gamut bin intervals for different clutter bands in response to light environment-temperature control;

[0118] The excitation state trend of the signal in the clutter band is calculated by combining the peak slope with the amplitude and frequency of the clutter band to obtain the excitation steepness.

[0119] If the excitation steepness is greater than the preset excitation steepness, the color gamut bin interval corresponding to the noise band in the signal spike model diagram is marked as an enhancement trend of 1; if the excitation steepness is less than the preset excitation steepness, the color gamut bin interval corresponding to the noise band is marked as an attenuation trend of 0, and the signal excitation result is obtained;

[0120] The phase deviation of the excellent band compared to the standard deep sleep waveform is calculated by registration to obtain the phase drift value. At the same time, the deviation area between the noise band and the standard deep sleep waveform is extracted to obtain the color saturation scale of the deviation area in the color gamut bin interval;

[0121] A guidance control law is designed based on the signal excitation results, and the guidance control law is executed on the smart home sleep device according to the color saturation scale to manage the light environment and temperature creation of the target user in the deep sleep stage until the phase drift value is equal to 0, thereby obtaining the second phase management strategy.

[0122] It should be noted that if the signal-to-noise ratio between the superior band and the clutter band is less than the preset signal-to-noise ratio, it indicates that the real-time sleep waveform is interspersed with a significant amount of interfering noise. The signal characteristics of the superior band may be hidden or obscured, significantly weakening the target user's ability to follow the superior band's guidance and enter the N3 deep sleep mode, making it very likely that they will remain in the N2 light sleep stage. To address this issue, in order to minimize the large proportion of clutter interference and further improve the deep sleep concentration quality of the superior waveform, this method establishes a signal spike model diagram of the real-time sleep waveform. The signal spike model diagram can express the excitation rate and excitation sharpness of the clutter band signal changes using the spike slope. This method tracks the excitation trends of the rising and falling edges of the clutter band, providing a quantifiable basis for the enhancement and attenuation of clutter interference and obtaining the excitation steepness of the clutter band. Previously, the power spectral density of the clutter band energy distribution was calculated using the Wigner-Ville algorithm. The Wigner-Ville algorithm, also known as the Wigner-Ville distribution algorithm, is used to simultaneously analyze the energy distribution of a signal in both the time and frequency domains. This power spectral density intuitively displays the distribution of energy carried by the EEG signal across frequency components within the noise band, reflecting the density variation of energy. Therefore, this method uses the calculus to estimate the probability density of the energy variation of the power spectral density corresponding to each noise band, thereby forming a binned thermal density framework that provides a platform for visualizing energy density variations. The light environment-temperature control rules can minimize the interference of clutter bands on excellent bands by regulating and managing the sleep light environment and temperature. Therefore, the RGB mapping gradient coding of the clutter bands is further eliminated according to the control rules to render the thermal density architecture. Finally, the bins below each clutter band on the spike model map are projected with color codes that eliminate the corresponding power energy density, forming a color gamut bin area exclusive to each clutter band. The color gamut bin interval not only displays the energy density changes of the clutter band power spectrum in the form of thermal distribution, but also dynamically formulates a control plan for the corresponding RGB color control scale to eliminate the clutter band with power energy density changes at each location. For example, when the power spectrum of the clutter band in the 35-55 Hz frequency band changes in the color gamut bin interval, the corresponding wavelength color value of the 215-400 nm sleep blue light is suppressed, making the sleep light environment and temperature created by subsequent control more suitable for entering deep sleep mode and effectively suppressing the interference of clutter.

[0123] It should be noted that if the excitation steepness is greater than the preset excitation steepness, it indicates that the signal excitation trend in the clutter band is rising, indicating high clutter activity. Therefore, the color bin corresponding to this clutter band in the signal spike model diagram is marked as 1, indicating an increasing trend. Conversely, if the excitation steepness is lower, it indicates that the signal excitation trend is falling, indicating low clutter activity. Therefore, the color bin corresponding to this clutter band is marked as 0, indicating a decreasing trend. The signal excitation results can provide instructions and specifications for light environment and temperature management and control, precisely constraining the control intensity of clutter elimination. The color saturation scale provides the control parameter range. The phase drift value reflects the phase difference between the current optimal wave and the standardized waveform signal. Therefore, only by eliminating this phase drift value can the optimal wave smoothly guide the target user into the N3 deep sleep stage along the standard deep sleep waveform. Therefore, the light environment and temperature created by the guidance control law and color coding chain for the target user's deep sleep stage must meet the requirement of a phase drift value of 0. This method can accurately create and manage the light environment and temperature during sleep, allowing users to quickly and smoothly enter deep sleep mode, reduce sleep alertness and misleading rate, and improve deep sleep quality.

[0124] In addition, the deep sleep staging management method based on EEG feedback further includes the following steps:

[0125] Obtaining management schemes associated with the first phase management strategy and the second phase management strategy, and simultaneously obtaining candidate control parameter solutions for the first phase management strategy and the second phase management strategy;

[0126] Obtain the actual number of fault levels where the real-time sleep wave continuously generates an excellent band within the optimal guidance window period when the target user is in the unreached stage, and construct the actual fault Bezier curve of the EEG feedback response based on the actual number of fault levels;

[0127] Obtain the tolerance threshold of excellent band fault feedback through the established deep sleep management plan, and construct the tolerance fault Bezier curve based on the tolerance threshold;

[0128] A discrete optimization algorithm is introduced to calculate the misalignment function between the actual fault Bezier curve and the tolerance fault Bezier curve. A backtracking scale is preset based on the misalignment function. Based on the backtracking scale, a global optimal discrete joint is performed for each candidate control parameter solution in the discrete optimization algorithm to obtain the degree of fit of each candidate control parameter solution.

[0129] If the degree of fit is greater than the preset degree of fit, the candidate control parameter solution corresponding to the degree of fit is used to replace the current control parameter, and several feasible control parameter examples are obtained that optimize the actual fault without exceeding the tolerance fault limit;

[0130] The simulation runs and records the coordination index between each set of feasible control parameters relative to the management scheme. The management scheme with the largest coordination index is selected as the optimal management scheme and uploaded to the smart home sleep device terminal to manage the parameter control of white noise, light environment and temperature in the deep sleep stage of the target user.

[0131] It should be noted that since the first and second stage management strategies are designed to manage different sleep environments during the target user's actual deep sleep phase, there are numerous appropriate control parameters. By analyzing the actual fault conditions of the optimal wave and the idealized fault conditions tolerated by the management plan, a Bezier curve difference calculation is performed to obtain a misalignment function. This misalignment function expresses the degree of irrationality of the current control parameters. Based on the misalignment function, a combinatorial optimization backtracking scale is preset to conduct a discrete combination analysis of all candidate control parameter solutions towards the global optimal trend, forming a control parameter combination that optimizes the current optimal wave fault phenomenon. If the degree of fit is greater than the preset fit, it indicates that the candidate control parameter solution is more capable of maintaining the coherence of the optimal wave transmission and improving the inertia quality of the N3 deep sleep wave. Therefore, this candidate control parameter solution is preferred to replace the current control parameters. Finally, the operation of the feasible control parameter case is simulated to analyze whether the management scheme can coordinate the replaced control parameters. This method can screen the optimal management scheme of suitable and feasible control parameters for the first stage management strategy and the second stage management strategy, reduce the error probability of controlling white noise, light environment and temperature parameters during deep sleep, and improve the accuracy and guidance performance of deep sleep stage management.

[0132] The second aspect of the present invention provides a deep sleep stage management system based on EEG feedback, such as Figure 3 As shown, a deep sleep staging management method based on EEG feedback is applied to any of the above items, and the system specifically includes:

[0133] An EEG monitoring module, which is used to monitor the EEG signals of the target user during sleep in real time;

[0134] The EEG feedback module is responsible for feeding back the EEG signals of the target user during sleep and converting the EEG signals into an EEG feedback time-frequency spectrum using a short-time Fourier transform algorithm;

[0135] A data processing module, the data processing module being configured to perform data planning processing on an optimal guidance window period for deep sleep intervention of a target user in coordination with the long-term window margin and the short-term window margin;

[0136] An intelligent analysis module, which is responsible for determining whether the signal-to-noise ratio between the optimal band and the noise band to which the real-time sleep waveform adheres during the optimal guidance window is greater than a preset signal-to-noise ratio, thereby performing strategic analysis on the guidance management of the target user who is about to enter the deep sleep stage;

[0137] The stage management module is used to manage the parameter control of the smart home sleep device and provide corresponding deep sleep stage guidance to the target user.

[0138] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

Claims

1. A deep sleep staging management method based on EEG feedback, characterized in that: The following steps are involved: S102: Collecting EEG feedback to extract the actual deep sleep waveform of the target user, planning the expected deep sleep spectrum range of the target user to generate a standard deep sleep waveform based on the ideal state transition of the sleep health quality data, and determining whether the actual deep sleep waveform is within the expected deep sleep spectrum range and whether it is locked in the unreached stage; S104: If the target user has not yet reached the stage, a signal pattern prediction model is constructed to predict the current signal pattern of the target user who is about to enter deep sleep. The optimal guidance window period for deep sleep intervention is planned based on the intervention entry point migration analysis of the current signal pattern. S106: performing filtering processing on the low-frequency components and high-frequency components of the EEG signal within the optimal guidance window period using a wavelet basis decomposition to obtain the excellent band and the noise band to which the real-time sleep waveform adheres within the optimal guidance window period; S108: If the signal-to-noise ratio of the excellent band to the clutter band is greater than a preset signal-to-noise ratio, crosstalk coupling analysis is performed to analyze the clutter band acting on the easily disturbed waveform segment of the excellent band, and mask tampering of the easily disturbed waveform segment is checked using a rhythm coding library of a standard waveform to generate white noise audio for guiding management of the deep sleep stage, thereby obtaining a first staged management strategy; S110: If the signal-to-noise ratio of the excellent band and the noise band is less than the preset signal-to-noise ratio, the color gamut binning interval of the light environment-temperature response control is planned to obtain a color coding chain for the noise band that deviates from the standard, the signal excitation trend is analyzed to obtain a signal excitation result, and the color coding chain and the signal excitation result are combined to manage the deep sleep stage to obtain a second stage management strategy.

2. The deep sleep staging management method based on EEG feedback according to claim 1, characterized in that: The S102 specifically includes the following steps: Obtaining a deep sleep staging strategy determined by a smart home sleep device for a target user and a pre-designed deep sleep management plan for the smart home sleep device to execute the deep sleep staging strategy; By running a predetermined deep sleep management plan on the target user and monitoring the EEG response in real time when using smart home sleep devices, the EEG feedback signal of the target user during sleep is obtained; Introducing a short-time Fourier transform algorithm to transform the EEG feedback signal in the time domain and frequency domain to generate an EEG feedback time-frequency spectrum, and extracting the actual deep sleep waveform of the target user from the EEG feedback time-frequency spectrum; By obtaining the potential sleep state chain of the established deep sleep management plan based on the standardized sleep staging system, a sleep state hierarchical nested tree is constructed. The sleep state hierarchical nested tree is used to perform state transition trigger inference on sleep health quality data to obtain the ideal state transition path for the target user's deep sleep staging. Based on the time domain characteristics and frequency domain characteristics of the standard deep sleep waveform, the sleep state knowledge graph is identified to obtain the oscillation rhythm of the standard deep sleep waveform under different sleep state conditions; A naive Bayesian inference model is constructed by calculating the transient prior probability of a standard deep sleep waveform based on a collection of observed parameters of the oscillatory rhythm. The naive Bayesian inference model is used to infer the oscillation sequence points of the ideal state transition path located in the EEG feedback time spectrum, and output a steady-state conditional probability. Combining transient prior probability and steady-state conditional probability analysis, we determine the ideal staging points in time and frequency that can produce a standard deep sleep waveform for the target user. Based on these ideal staging points, we plan deep sleep staging on the EEG feedback time-frequency spectrum to obtain the target user's expected deep sleep spectrum. If the actual deep sleep waveform does not exist within the expected deep sleep spectrum at this moment, the target user is locked as not reaching the stage.

3. The deep sleep staging management method based on EEG feedback according to claim 2, characterized in that: The method of constructing a sleep state hierarchical nested tree by acquiring a potential sleep state chain for executing a predetermined deep sleep management plan based on a standardized sleep staging system, and using the sleep state hierarchical nested tree to perform state transition triggering reasoning on sleep health quality data, thereby obtaining an ideal state transition path for the target user's deep sleep staging, specifically includes the following steps: Based on big data, we obtain a sleep state knowledge graph and a standardized sleep staging system that meets the target user group. We then simultaneously search and identify the sleep states that occur in a chain reaction when implementing a predetermined deep sleep management plan in accordance with the standardized sleep staging system, and obtain several potential sleep state chains. Establish a state stack from the top parent state to the bottom leaf state, determine the contextual logic structure of the state stack based on several potential sleep state chains, and construct a hierarchical nested tree of sleep states based on the contextual logic structure; Obtain the target user's sleep health log, preset multiple sets of evenly continuous health monitoring timestamps, and extract the target user's sleep health quality data at each health monitoring timestamp when the target user accepted the established deep sleep management plan through the sleep health log; Import the sleep health quality data into the sleep state hierarchical nested tree and search the current state hierarchy to the parent state layer by layer, and traverse and calculate the current trigger matching value between each stage and the state trigger constraint premise proposed by the standardized sleep stage system; By determining whether the trigger matching threshold is greater than the trigger matching threshold of each state trigger constraint premise, the transfer entropy weight of the state trigger constraint premise corresponding to the top-level parent state is weightedly calculated in the strategy trade-off algorithm to generate the entry channel and the exit channel; Based on the exit logic pattern from the entry channel to the exit channel, the sleep state transition clues under the premise of sleep health quality are calculated to obtain the ideal state transition path of the target user's deep sleep stage.

4. The deep sleep staging management method based on EEG feedback according to claim 1, characterized in that: The S104 specifically includes the following steps: When the target user is locked and marked as not reaching the stage, the EEG management log of the smart home sleep device is obtained. The historical response inflection points of the spectrum feedback when the target user's deep sleep waveform is in EEG feedback during the past time period and the emergence timing of each historical response inflection point are extracted from the EEG management log; A multi-layer perceptron model is introduced to construct a variational recognition encoder and variational recognition decoder based on dynamic time series, and the historical response inflection points and corresponding emergence timing data are injected into the variational recognition encoder; After the injection, the variational recognition encoder uses a recurrent neural network to analyze the regularity of the patterns of historical response inflection points at the time of emergence. It then uses a gated recurrent unit to track the temporal variation of the patterns at the time of emergence relative to the occurrence of the historical response inflection points. This pattern regularity and temporal variation are then input into the variational recognition decoder for reconstruction and propagation, thereby constructing a signal pattern prediction model for deep sleep waveforms. The signal pattern prediction model is used to predict the actual deep sleep waveform of the EEG feedback spectrum, outputting the current signal pattern of the actual deep sleep that is about to enter. The intervention point of the actual deep sleep waveform on the future timeline is determined based on the mode switching rhythm of the current signal pattern, which is defined as the first intervention entry point. Obtaining an intervention point in the desired deep sleep spectrum domain, defining it as a second intervention entry point, and calculating a transfer function between the first intervention entry point and the second intervention entry point; If the migration function is greater than the preset migration function, the jurisdiction area of ​​the first intervention entry point and the second intervention entry point is marked as the long-term window margin; if it is less than, it is marked as the short-term window margin. The optimal guidance window period for deep sleep intervention of the target user is generated based on the collaborative planning of the long-term window margin and the short-term window margin.

5. The deep sleep staging management method based on EEG feedback according to claim 1, characterized in that: The S106 specifically includes the following steps: Preset the wavelet basis function, and use the wavelet basis function to spectrally decompose the brain waves in the optimal guidance window period during EEG feedback to obtain low-frequency sub-signals and high-frequency sub-signals at different wavelet basis decomposition layers; Filtering is performed based on the wavelet coefficients of the low-frequency sub-signal and the high-frequency sub-signal to extract the excellent bands and noise bands to which the real-time sleep wave is attached during the optimal guidance window.

6. The deep sleep staging management method based on EEG feedback according to claim 1, characterized in that: The S108 specifically includes the following steps: If the signal-to-noise ratio between the excellent band and the noise band is greater than the preset signal-to-noise ratio, the rhythm representation template of the output EEG signal corresponding to the standard deep sleep waveform is obtained based on the big data, and the rhythm representation template is encoded using the EEG signal encoding rules to obtain the rhythm encoding library of the standard deep sleep waveform; Highlighting the real-time sleep wave in the optimal guidance window period, extracting the position pattern of the excellent band relative to the real-time sleep wave, which is defined as the first subsidiary wave distribution, and extracting the position pattern of the noise band relative to the real-time sleep wave, which is defined as the second subsidiary wave distribution; The crosstalk coupling strength of the clutter band on the excellent band is calculated based on the co-frequency feedback data of amplitude and frequency between the excellent band and the clutter band. Based on the second subsidiary wave distribution, only the sub-coupling waveform segments in the first subsidiary wave distribution with crosstalk coupling strength greater than the preset crosstalk coupling strength are marked and defined as the easily disturbed waveform segments of the real-time sleep wave. Construct a registration domain, in which the real-time sleep wave is superimposed and masked onto the standard deep sleep waveform for registration. At this time, the mask segment where the easily disturbed waveform segment is located is intercepted from the rhythm coding library to obtain the corrected mask sequence and the actual coding sequence of the easily disturbed coupled waveform segment; Obtain the association management mapping table of the sleep wave frequency band and the white noise frequency band for the smart home sleep device, check the tampering mask for missing waveform frequency bands in the actual coding sequence compared to the return mask sequence, search the association management mapping table for the corresponding white noise compensation frequency band based on the tampering mask, and generate a white noise audio control decision; The deep sleep stage of the target user is managed according to the white noise audio control decision, and the first phased management strategy is obtained.

7. The deep sleep staging management method based on EEG feedback according to claim 1, characterized in that: The S110 specifically includes the following steps: If the signal-to-noise ratio between the excellent band and the noise band is less than the preset signal-to-noise ratio, a signal peak model diagram of the real-time sleep wave is constructed through the signal oscilloscope modeling software; The Wigner-Ville algorithm is introduced to calculate the energy distribution of the clutter band and output the power spectrum density of the clutter band. At the same time, the peak slope of the clutter band is obtained through scalar data analysis of the signal spike model. Obtain the control rules for light environment-temperature creation, perform calculus estimation of the distribution probability density for each clutter band based on the power spectrum density located below the signal peak model diagram, and render according to the coding of the control rules to generate the color gamut bin intervals for different clutter bands in response to light environment-temperature control; The excitation state trend of the signal in the clutter band is calculated by combining the peak slope with the amplitude and frequency of the clutter band to obtain the excitation steepness. If the excitation steepness is greater than the preset excitation steepness, the color gamut bin interval corresponding to the noise band in the signal spike model diagram is marked as an enhancement trend of 1; if the excitation steepness is less than the preset excitation steepness, the color gamut bin interval corresponding to the noise band is marked as an attenuation trend of 0, and the signal excitation result is obtained; The phase deviation of the excellent band compared to the standard deep sleep waveform is calculated by registration to obtain the phase drift value. At the same time, the deviation area between the noise band and the standard deep sleep waveform is extracted to obtain the color saturation scale of the deviation area in the color gamut bin interval; A guidance control law is designed based on the signal excitation results, and the guidance control law is executed on the smart home sleep device according to the color saturation scale to manage the light environment and temperature creation of the target user in the deep sleep stage until the phase drift value is equal to 0, thereby obtaining the second phase management strategy.

8. A deep sleep stage management system based on EEG feedback, characterized in that: The system is applied to implement a deep sleep staging management method based on EEG feedback as described in any one of claims 1 to 7, and specifically comprises: An EEG monitoring module, which is used to monitor the EEG signals of the target user during sleep in real time; The EEG feedback module is responsible for feeding back the EEG signals of the target user during sleep and converting the EEG signals into an EEG feedback time-frequency spectrum using a short-time Fourier transform algorithm; a data processing module, the data processing module being configured to perform data planning processing on an optimal guidance window period for deep sleep intervention of a target user in coordination with the long-term window margin and the short-term window margin; An intelligent analysis module, which is responsible for determining whether the signal-to-noise ratio between the optimal band and the noise band to which the real-time sleep waveform adheres during the optimal guidance window is greater than a preset signal-to-noise ratio, thereby performing strategic analysis on the guidance management of the target user who is about to enter the deep sleep stage; The stage management module is used to manage the parameter control of the smart home sleep device and provide corresponding deep sleep stage guidance to the target user.

Images