Sleep assisting system based on photoacoustic magnetic vibration wave resonance

Abstract

The invention provides a sleep assisting system based on photoacoustic magnetic vibration wave resonance, and relates to the technical field of sleep assisting. The system comprises an acquisition unit, a processing unit, a stimulation unit and an evaluation feedback unit. Determining a sleep stage of the user, analyzing phases of multi-scale electroencephalogram rhythm components such as slow waves and spindle waves in a cross-scale manner, and extrapolating and predicting a phase value of a target electroencephalogram rhythm at a future moment in combination with a recent phase trajectory; taking the predicted phase as a target state, mapping a rising / falling slope of a slow wave form into an intensity change parameter of photoacoustic and magnetic vibration stimulation by constructing a mapping channel, and selecting an optimal stimulation combination for output through a decision model based on the intensity change parameter; calculating a phase locking effect index, and adjusting a phase prediction model parameter by using an optimization algorithm according to the index so as to optimize subsequent stimulation; and linkage regulation and control of an external environment are supported, and centralized monitoring of sleep states of multiple users can be realized. According to the invention, self-adaptive guidance of sleep brain rhythms is realized, and the sleep quality is effectively improved.

Description

Technical Field

[0001] This application relates to the field of sleep assistance technology, and more specifically, to a sleep assistance system based on photoacoustic-magnetic vibration wave resonance. Background Technology

[0002] Traditional sleep aids typically rely on medication or single physical stimulation, which suffers from side effects, dependence, and poor individual adaptability. Existing non-pharmacological techniques, such as using specific frequencies of sound, light pulses, or transcranial electrical / magnetic stimulation to modulate brain activity, attempt to guide or synchronize brain waves to promote sleep. However, these methods generally have shortcomings: First, they often employ open-loop, fixed-pattern stimulation, failing to dynamically adjust according to the user's real-time sleep state and brain activity, resulting in low stimulation efficiency or interference; second, the synchronization accuracy between the stimulation signal and the user's intrinsic brain rhythm is insufficient, particularly neglecting the dynamic changes in brain phase, making it difficult to achieve truly effective "phase-locked" stimulation and limiting the resonance effect; finally, the scope and depth of single-modal stimulation are limited, with weak targeted regulation of different sleep stages (such as slow waves promoting deep sleep and theta waves during REM sleep). To improve synchronization, closed-loop stimulation techniques based on event detection have emerged, for example, triggering a single sound or electrical stimulation after detecting slow-wave oscillations. These methods achieve "follow-up" intervention, but in essence, they are still "post-event" responses to EEG events that have already occurred. They have inherent systemic and physiological delays and are difficult to achieve true "phase synchronization".

[0003] To address these issues, current mainstream research is exploring closed-loop neuromodulation techniques and multisensory integrated stimulation. For example, real-time EEG analysis of sleep stages is used to switch stimulation modes accordingly. Some studies have also attempted to introduce phase prediction techniques to predict future phases of EEG rhythms in advance for early stimulation triggering. However, existing prediction methods largely rely on simple temporal extrapolation of historical data from a single rhythm, failing to fully consider the inherent neurophysiological coupling mechanisms between different EEG rhythms. This results in poor robustness and low accuracy during sleep stage transitions or rhythm instability. Furthermore, the inherent system delay from EEG signal acquisition and processing to stimulation output makes it difficult for stimulation to precisely target the "optimal time window" of the target EEG phase, limiting the real-time performance and accuracy of phase locking. Under multimodal stimulation conditions, there is no mature solution for how to make real-time, collaborative decisions on the optimal stimulation combination based on dynamic sleep states to maximize phase locking benefits and avoid intermodal interference. Finally, the system lacks real-time evaluation of stimulation effects and adaptive parameter optimization capabilities, hindering long-term iterative optimization of stimulation strategies at the individual level, thus affecting the system's long-term effectiveness and personalization level.

[0004] Therefore, overcoming system latency to achieve precise phase synchronization, conducting intelligent collaborative multimodal stimulation, and constructing an adaptive closed-loop control system that can self-optimize based on real-time effects have become key to improving the effectiveness of sleep aid technology. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this application provides a sleep aid system based on photoacoustic-magnetic vibration wave resonance, comprising:

[0006] The acquisition unit is used to acquire the user's physiological signals in real time, analyze the physiological signals, and determine the current sleep stage;

[0007] The processing unit, in response to the sleep stage being a preset target sleep stage, extracts the instantaneous phase of the target EEG rhythm from the physiological signal, and predicts the phase value of the target EEG rhythm at a preset future time through a phase prediction model.

[0008] The stimulation unit is equipped with a multimodal stimulation generator, which can generate at least one of light, sound, magnetic, and vibration signals; the stimulation unit controls the multimodal stimulation generator to output a composite stimulation signal that is phase-locked with the target EEG rhythm based on the phase value of the preset future time.

[0009] The evaluation feedback unit is used to collect feedback physiological signals after the application of the composite stimulation signal, calculate the phase-locking effect index, and adjust the parameters of the phase prediction model according to the phase-locking effect index using a preset optimization algorithm to update subsequent composite stimulation signals.

[0010] In one optional implementation, obtaining the phase value of the target EEG rhythm at a preset future time includes:

[0011] Extract at least two different time-scale EEG rhythm components associated with the current sleep stage;

[0012] A cross-scale coupling analysis was performed on the instantaneous phase sequence of the EEG rhythm component to obtain the phase evolution constraints of the target EEG rhythm component.

[0013] Based on the phase evolution constraints and the phase change trajectory of the target EEG rhythm within a preset time period, the phase value of the target EEG rhythm at the preset future time is predicted.

[0014] In one optional implementation, the extraction of at least two different time-scale EEG rhythm components associated with the current sleep stage includes:

[0015] The electroencephalogram (EEG) signal in the physiological signal was filtered in parallel using a dual-channel bandpass filter to obtain the slow wave oscillation component of the first frequency band and the sleep spindle wave component of the second frequency band, respectively.

[0016] The Hilbert transform is applied to the slow-wave oscillation component to obtain the instantaneous phase sequence of the slow-wave oscillation component analytically.

[0017] The amplitude envelope of the sleep spindle wave component is extracted, and the maximum value of the amplitude envelope within a preset time window is identified.

[0018] Based on the maximum value moment, the corresponding slow wave instantaneous phase value is indexed from the instantaneous phase sequence on the same time axis;

[0019] Calculate the phase deviation between the slow wave instantaneous phase value and the preset standard coupling phase threshold;

[0020] Based on the phase deviation, the center frequency of the slow wave oscillation component is modulated, and the modulated frequency is determined as the frequency of the target EEG rhythm.

[0021] In one optional implementation, modulating the center frequency of the slow-wave oscillation component based on the phase deviation further includes verifying the slow-wave morphology before performing modulation.

[0022] Extreme point detection is performed on the slow wave oscillation component to determine the trough and peak times within a single cycle;

[0023] Based on the signal segment from the trough to the peak, a rising edge component is constructed;

[0024] Based on the signal segment from the peak time to the next trough time, a falling edge component is constructed;

[0025] The mean of the first derivative of the rising edge component is calculated as the rising slope, and the mean of the first derivative of the falling edge component is calculated as the falling slope.

[0026] The ratio of the rising slope to the falling slope is calculated to obtain the waveform asymmetry index;

[0027] When the waveform asymmetry index falls within a preset sleep range, the center frequency of the slow wave oscillation component is modulated.

[0028] In one optional implementation, controlling the multimodal stimulation generator to output a composite stimulation signal that is phase-locked with the target EEG rhythm based on the phase value of the preset future time includes:

[0029] The phase value at the preset future time is taken as the target phase state of the composite stimulus signal;

[0030] The frequency of the modulated target EEG rhythm is used as the rhythm reference of the composite stimulation signal;

[0031] Construct the first mapping channel to map the rising slope to the intensity growth parameters of the acoustic and optical stimulation signals;

[0032] A second mapping channel is constructed to map the descent slope to the intensity attenuation rate of the magnetic field or vibration stimulus signal;

[0033] Based on the target phase state, the intensity growth parameter, and the intensity decay rate, an optimal stimulus combination is selected from predefined composite stimulus modes through a real-time collaborative decision-making model. The optimal stimulus combination includes: stimulus type, stimulus intensity, and waveform parameters.

[0034] The optimal stimulus combination, along with the intensity growth parameter and intensity decay rate, is synthesized into a digitally controlled waveform sequence.

[0035] Based on the digital control waveform sequence and the real-time delay estimate of each modal stimulation channel, the composite stimulation command is calculated in reverse. The composite stimulation command includes: the pre-compensation trigger time of each modal stimulus and the digital control waveform sequence.

[0036] Based on the composite stimulus command, each modal stimulus is output synchronously at the pre-compensation trigger time.

[0037] In an optional implementation, the step of selecting the optimal stimulus combination from predefined composite stimulus modalities through a real-time collaborative decision-making model further includes:

[0038] Based on the real-time collaborative decision-making model, an action set containing combinations of various stimulus modalities is established.

[0039] Based on the target phase state, the intensity growth parameter, the intensity decay rate, and the waveform asymmetry index, calculate the expected phase-locked loop benefit value for each action;

[0040] Using the expected phase-locked loop return value as the optimization objective, the action with the highest expected phase-locked loop return value is selected as the optimal stimulus combination output by the real-time collaborative decision-making model.

[0041] In one optional implementation, the step of synchronously outputting each modal stimulus at the pre-compensation trigger time based on the composite stimulus instruction includes:

[0042] The pre-compensation trigger time is determined by calculating the difference between the preset future time and the real-time delay value of each modal stimulus.

[0043] In one optional implementation, the step of collecting feedback physiological signals after applying the composite stimulation signal and calculating the phase-locking effect index includes:

[0044] The phase-locked loop performance indicators include at least: phase synchronization indicators, slow wave activity energy change indicators, and waveform asymmetry index improvement indicators;

[0045] The phase-locked loop effect index is generated by weighting and comprehensively scoring the phase synchronization index, the slow wave activity energy change index, and the waveform asymmetry index improvement index.

[0046] Specifically, based on the composite stimulation signal and the feedback physiological signal, rhythms with the same frequency as the target EEG rhythm are extracted and their phase sequences are calculated.

[0047] Calculate the synchronicity measure between two phase sequences to obtain the phase synchronization index;

[0048] The change index of slow wave activity energy is obtained by calculating the relative change in the energy of the target EEG rhythm frequency band before and after the application of the composite stimulation signal;

[0049] The improvement index of the waveform asymmetry index is obtained by calculating the degree to which the waveform asymmetry index approaches the preset sleep range before and after the application of the composite stimulation signal.

[0050] In one optional implementation, adjusting the parameters of the phase prediction model based on the phase-locked loop effect index using a preset optimization algorithm to update subsequent composite stimulus signals includes:

[0051] Construct a mapping relationship between the parameter combinations of the phase prediction model and the corresponding phase-locked loop performance indicators within the historical period;

[0052] Based on the mapping relationship, with the phase-locked loop performance index as the optimization target, a preset optimization algorithm is used to search for the optimal combination of parameters that makes the phase-locked loop performance index the best.

[0053] The phase prediction model is updated using the optimal parameter combination, and the phase prediction of the target EEG rhythm and the generation of compound stimulation signals are continued.

[0054] In an optional implementation, generating the phase-locked loop performance index further includes:

[0055] Based on the sleep stage and the phase-locked loop effect index, generate environmental control commands;

[0056] The relationship between the phase-locked loop effect index and the preset threshold is compared, and the smart home devices are controlled by the environmental control command, and a prompt message is sent to the preset monitoring terminal.

[0057] Compared with existing technologies, this invention extracts and analyzes the instantaneous phase relationship between two rhythmic components with different time scales: slow-wave oscillations and sleep spindle waves. Through cross-scale coupling analysis, it derives phase evolution constraints for prediction. This method fully utilizes the brain's inherent neural coupling mechanism as prior knowledge for prediction, effectively overcoming the problem of prediction failure when the rhythm is unstable in traditional methods. By constructing independent mapping channels, physiological characteristics are transformed into intensity growth parameters of photoacoustic stimulation and intensity decay rates of magnetic resonance stimulation, enabling external stimuli to be synchronized with the EEG rhythm in time. A multi-dimensional phase-locking effect index covering phase synchronization, slow-wave energy changes, and waveform asymmetry improvement is defined to quantitatively evaluate each intervention, thereby enabling the system to continuously adapt to the user's personalized physiological response patterns and effectively improve the quality of sleep assistance. Attached Figure Description

[0058] Figure 1 A schematic diagram of a sleep aid system based on photoacoustic-magnetic vibration wave resonance provided in an embodiment of this application;

[0059] Figure 2 A flowchart of a sleep aid based on photoacoustic-magnetic vibration wave resonance provided for an embodiment of this application;

[0060] Figure 3 A flowchart of dual-channel rhythm coupling phase prediction provided in an embodiment of this application;

[0061] Figure 4 A flowchart illustrating the dynamic evaluation and optimization of the phase prediction model provided in this application embodiment. Detailed Implementation

[0062] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0063] See Figure 1 and Figure 2 The diagram shown is a schematic of a sleep aid system based on photoacoustic-magnetic vibration wave resonance provided in an embodiment of this application. The system includes a data acquisition unit 10, a processing unit 20, a stimulation unit 30, and an evaluation feedback unit 40, wherein:

[0064] The acquisition unit 10 is used to acquire the user's physiological signals in real time and analyze the physiological signals to determine the current sleep stage;

[0065] Processing unit 20, in response to the sleep stage being a preset target sleep stage, extracts the instantaneous phase of the target EEG rhythm from the physiological signal, and predicts the phase value of the target EEG rhythm at a preset future time through a phase prediction model.

[0066] The stimulation unit 30 is equipped with a multimodal stimulation generator, which can generate at least one of light, sound, magnetic, and vibration signals; the stimulation unit controls the multimodal stimulation generator to output a composite stimulation signal that is phase-locked with the target EEG rhythm based on the phase value of the preset future time.

[0067] The evaluation feedback unit 40 is used to collect feedback physiological signals after the application of the composite stimulation signal, calculate the phase-locking effect index, and adjust the parameters of the phase prediction model according to the phase-locking effect index using a preset optimization algorithm to update subsequent composite stimulation signals.

[0068] In specific implementation, the physiological signals include electroencephalogram (EEG) signals, electrocardiogram (ECG) signals, electromyogram (EMG) signals, etc.

[0069] For example, the EEG signals are collected by a dry electrode or hydrogel electrode cap worn on the user's head.

[0070] As an optional implementation method, a multi-channel EEG acquisition device (sampling rate ≥256Hz, resolution ≥16 bits) is used to simultaneously acquire the user's EEG signals, ECG signals, EMG signals, and EEG signals;

[0071] The data collection period covers the entire sleep cycle, and a preset sleep cycle length is continuously collected as the basic dataset. The preset sleep cycle length is set according to the actual engineering situation.

[0072] The basic dataset contains balanced samples from different age groups (18-65 years old) and different sleep quality groups (no sleep disorders, mild insomnia, moderate insomnia);

[0073] The base dataset is divided into a training set, a test set, and a validation set for training, testing, and validating the phase prediction model, respectively. As an optional implementation, a target EEG rhythm corresponding to the frequency band of the current sleep stage is extracted from the EEG signal.

[0074] In practice, noise in the EEG signal is identified and classified through spectrum analysis and waveform feature matching.

[0075] For example, a notch filter is used to specifically filter out power frequency noise while preserving the original frequency band characteristics of the EEG signal;

[0076] Based on independent component analysis, the EEG signal is decomposed, the independent components corresponding to electromyography / electroographia artifacts are separated and removed, and then the pure EEG signal is reconstructed through inverse transformation.

[0077] Random noise is removed by wavelet thresholding. Adaptive thresholding (based on dynamic calculation of noise variance) is used to process high-frequency wavelet coefficients, while the original values ​​of low-frequency coefficients are retained. Finally, the signal is recovered by inverse wavelet transform.

[0078] For short-term signal loss caused by electrode detachment, linear interpolation is used to fill in the missing signal.

[0079] After denoising, the signal-to-noise ratio of the signal is calculated to ensure that the denoising process does not destroy the waveform characteristics of key physiological rhythms such as slow waves and spindle waves.

[0080] The sleep stages are segmented according to a preset sleep segmentation standard;

[0081] As an optional implementation method, sleep stages are determined according to sleep staging standards (such as the AASM standard), resulting in sleep stages including the sleep onset stage, light sleep stage, deep sleep stage, and REM sleep stage.

[0082] The spectrum and waveform characteristics of the EEG signal in the physiological signals are extracted across the entire frequency band to determine the current sleep stage.

[0083] It should be noted that the determination of sleep stages can be achieved using existing mature sleep staging methods and is not limited to a specific algorithm or model. In one embodiment, a professionally trained sleep technician can manually interpret and label the collected EEG signals and related physiological signals according to preset sleep staging standards to obtain sleep stage labels corresponding to each time period. In another embodiment, a publicly available automatic sleep staging algorithm can also be used, such as a model combining convolutional neural networks and recurrent neural networks, to extract and classify features from the original multichannel physiological signals, thereby automatically outputting sleep stage labels equivalent to the aforementioned manual labeling. This application does not impose specific implementation methods on the sleep staging module as limiting conditions. Those skilled in the art can choose any suitable implementation scheme based on the actual system deployment environment and computing resource constraints, provided that reliable sleep stage information is obtained, without affecting the implementation of other modules of this application.

[0084] See Figure 3 The flowchart shown is a process for a dual-channel rhythm-coupled phase prediction method provided in an embodiment of this application, including:

[0085] S201: The slow wave oscillation component and the sleep spindle wave component are synchronously separated using a dual-channel filter;

[0086] S202: Perform Hilbert transform on the slow wave oscillation component to obtain its instantaneous phase sequence, extract the amplitude envelope of the spindle wave component and mark the time when its peak occurs;

[0087] S203: Map the peak time of the spindle wave to the slow wave phase sequence to obtain the actual coupled phase value, and calculate the phase deviation between this value and the preset standard coupled phase threshold.

[0088] S204: Based on phase deviation data, analyze the dynamic law of slow wave-spindle wave phase coupling and extract the constraint conditions characterizing the phase evolution trend under the current state.

[0089] S205: Continuously tracks the instantaneous phase of the slow wave oscillation component and caches the phase change sequence of the most recent few complete cycles as the recent trajectory;

[0090] S206: Combine the constraints with the recent trajectory to predict the phase value of the preset future time after compensation for the delay.

[0091] As an optional implementation, at least two different time-scale EEG rhythm components associated with the current sleep stage are extracted;

[0092] For example, the EEG rhythm components include: low-frequency slow-wave delta waves, theta waves associated with early sleep onset and REM sleep, alpha waves during wakefulness with eyes closed, and high-frequency rhythmic beta waves associated with wakefulness and cognitive activity.

[0093] In specific implementation, a dual-channel bandpass filter is used to perform parallel filtering on the electroencephalogram (EEG) signal in the physiological signal to obtain the slow wave oscillation component of the first frequency band and the sleep spindle wave component of the second frequency band, respectively.

[0094] The slow-wave oscillation component is a low-frequency, high-amplitude oscillation activity separated from the electroencephalogram (EEG) signal, and is a typical electrophysiological marker of deep sleep.

[0095] The sleep spindle wave component is a short, high-frequency oscillation sequence with a spindle-shaped appearance, separated from the electroencephalogram (EEG) signal, and is a characteristic marker of light sleep.

[0096] In specific implementation, the first frequency band refers to the frequency band from which slow-wave oscillation characteristic physiological rhythms can be extracted from the user's EEG signal;

[0097] The second frequency band refers to the frequency band that can extract the physiological rhythm of sleep spindle waves.

[0098] For example, for typical physiological characteristics of adults, the first frequency band may be about 0.5 Hz to about 4.0 Hz, and the second frequency band may be about 11.0 Hz to about 16.0 Hz;

[0099] For elderly users, since their slow wave oscillations and sleep spindles often show characteristic shifts in frequency, these can be adjusted through analysis of individual EEG spectra.

[0100] For example, a Hilbert transform is performed on the slow-wave oscillation component to obtain the instantaneous phase sequence of the slow-wave oscillation component;

[0101] The Hilbert transform converts the real signal of the slow-wave oscillation component into an analytic signal and separates the phase angle of the slow-wave oscillation component at each sampling point, forming an instantaneous phase sequence that varies with time.

[0102] Extreme point detection is performed on the slow wave oscillation component to determine the trough and peak times within a single cycle;

[0103] The single cycle refers to the time period and waveform segment that occurs from a trough to the next peak and then back to the next trough.

[0104] Based on the signal segment from the trough to the peak, a rising edge component is constructed;

[0105] Based on the signal segment from the peak time to the next trough time, a falling edge component is constructed;

[0106] In practice, the data points of the rising and falling edge components are tested using the 3σ criterion. After removing outliers that exceed the standard deviation of the mean by a preset multiple, linear least squares fitting is performed to ensure the robustness of the slope calculation.

[0107] The standard deviation of the preset multiple is determined based on the actual engineering accuracy requirements.

[0108] The mean of the first derivative of the rising edge component is calculated as the rising slope, and the mean of the first derivative of the falling edge component is calculated as the falling slope.

[0109] In a specific implementation, the data point sequences of the rising edge component and the falling edge component are fitted with linear least squares, and the slope of the resulting fitted line is used as the rising slope and the falling slope to avoid noise.

[0110] The ratio of the rising slope to the falling slope is calculated to obtain the waveform asymmetry index.

[0111] As an optional implementation, the waveform asymmetry index is mapped to the [0, 1] interval using the min-max normalization method.

[0112] In specific implementation, the amplitude envelope of the sleep spindle wave component is extracted, and the maximum value of the amplitude envelope within a preset time window is identified;

[0113] The amplitude envelope is a curve that can outline the overall trend of the oscillation amplitude of the sleep spindle wave component;

[0114] The amplitude envelope is extracted to identify the occurrence and peak value of the spindle wave.

[0115] Based on the maximum value moment, the corresponding slow wave instantaneous phase value is indexed from the instantaneous phase sequence on the same time axis;

[0116] Calculate the phase deviation between the slow wave instantaneous phase value and the preset standard coupling phase threshold;

[0117] Based on the phase deviation, the center frequency of the slow wave oscillation component is modulated, and the modulated frequency is determined as the frequency of the target EEG rhythm.

[0118] For example, by analyzing historical sleep EEG big data, the typical phase position of the most frequently occurring spindle wave event within the slow wave oscillation cycle is statistically determined, and the phase value is set as the standard coupled phase threshold.

[0119] In specific implementation, the center frequency is the average frequency at which the slow wave oscillation component dominates during the current time period.

[0120] A cross-scale coupling analysis was performed on the instantaneous phase sequence of the EEG rhythm components to derive the phase evolution constraints of the target EEG rhythm components.

[0121] As an optional implementation, phase-amplitude coupling is used as the coupling type between slow wave oscillations and sleep spindles, that is, the instantaneous phase of the slow wave modulates the amplitude change of the sleep spindle, which conforms to the physiological coupling mechanism of brain electrical rhythms in the sleep state.

[0122] The coupling strength is quantized using the modulation index, which takes values ​​in the range of [0, 1].

[0123] For example, if the modulation index is greater than or equal to a preset modulation threshold, it is determined to be a strong coupling state; if the modulation index is less than the preset modulation threshold, it is determined to be a weak coupling state. The preset modulation threshold is set according to actual coupling requirements. As an optional implementation, the target EEG rhythm center frequency obtained by real-time modulation is multiplied by twice pi to convert it into the corresponding fundamental angular frequency.

[0124] For example, the modulation index can be determined based on the joint distribution between the slow wave phase and the spindle wave amplitude envelope. Specifically, a complete slow wave oscillation period can be divided into several preset phase intervals according to the phase. Within each phase interval, the average value and / or frequency of the spindle wave amplitude envelope in the corresponding time period are statistically analyzed to obtain the distribution of the spindle wave amplitude in each phase interval.

[0125] Furthermore, this distribution is compared with an ideal uniform distribution with equal weights in each phase interval to obtain the deviation of the spindle wave from the slow wave phase under the current state. This deviation, after normalization, is used as the value of the modulation index. Specifically, when the spindle wave is basically uniformly distributed across each phase interval and the phase-amplitude coupling is weak, the modulation index tends to be a smaller value; when the spindle wave is clearly concentrated in a few phase intervals and the phase-amplitude coupling is significant, the modulation index tends to be a larger value.

[0126] The coupling strength between the slow wave and the spindle wave at the current moment is calculated using the modulation index formula; a preset empirical coefficient is used as the coupling coefficient, which can be obtained by fitting training data.

[0127] In specific implementation, by substituting the phase evolution constraint equation, the difference between the real-time coupling strength and the standard coupling strength threshold is calculated, the difference is multiplied by the phase deviation, and the product result from the previous step is multiplied by the coupling coefficient to obtain the dynamic frequency correction amount.

[0128] The dynamic frequency correction is then added to the fundamental angular frequency to obtain the instantaneous phase change rate of the target EEG rhythm at the current moment.

[0129] The dynamic frequency correction amount is a dynamic angular frequency fine-tuning value, which is determined by the aforementioned deviation.

[0130] The instantaneous phase change rate is not constant, but is dynamically determined by the fundamental frequency, the difference between the current coupling strength and the standard strength, and the deviation between the current phase and the ideal phase.

[0131] For example, the phase evolution constraint equation is used to describe the empirical relationship between the instantaneous phase of the target EEG rhythm and time. The instantaneous phase change rate of the target EEG rhythm at the current moment can be understood as a superposition of several parts: one part is the base change rate calculated from the current center frequency, reflecting the natural oscillation speed of the target EEG rhythm without any coupling modulation; another part is a coupling correction amount obtained based on the difference between the current coupling strength and the standard coupling strength threshold. When the actual coupling strength is lower than the standard strength, this correction amount drives the instantaneous phase change rate to adjust towards strengthening coupling; when the actual coupling strength is higher than the standard strength, this correction amount drives the instantaneous phase change rate to adjust towards weakening coupling; it may also include a phase correction amount obtained based on the deviation between the current phase value and the standard coupling phase threshold, used to gradually approach the desired coupling phase of the target EEG rhythm. Each correction amount is multiplied by a coefficient obtained by fitting training data and then weighted and superimposed with the base change rate to obtain the instantaneous phase change rate at the current moment.

[0132] In this way, while ensuring the basic frequency of the target EEG rhythm remains stable, the phase evolution trajectory can be finely adjusted based on the coupling state and phase deviation.

[0133] As an optional implementation, the instantaneous phase sequence of the target EEG rhythm over the most recent few cycles is continuously cached to form its phase change trajectory within a preset time period;

[0134] Based on the phase evolution constraints and the phase change trajectory of the target EEG rhythm within a preset time period, the phase value of the target EEG rhythm at the preset future time is obtained by extrapolation through the phase prediction model.

[0135] The preset time period is dynamically set according to the prediction accuracy requirements.

[0136] For example, the length of the preset time period is the time corresponding to N complete cycles of the target EEG rhythm, where N is an integer greater than or equal to 2. For example, when the target rhythm is a slow wave oscillation (about 0.5-4Hz), the preset time period can be set to cover 2 to 5 slow wave cycles, that is, an EEG signal segment of about 2 to 10 seconds.

[0137] As an optional implementation, a hybrid model combining a long short-term memory network and an attention mechanism is used as the phase prediction model, with the input being a time step and a feature dimension; the time step is a slow wave period, and the feature dimension includes at least: waveform asymmetry index, coupling strength, phase deviation, and center frequency.

[0138] The hidden layer of the phase prediction model contains at least two LSTM layers, with a Dropout layer added after each layer to prevent overfitting; the output of the LSTM layer is connected to an attention layer, which assigns weights to features at different time steps (higher weights to historical periodic features with higher importance).

[0139] The input features are used to construct training samples by sliding a window according to the time step. Each sample corresponds to a phase at a future time and the mean squared error is used as the loss function.

[0140] For example, the Adam optimizer (learning rate 0.001) is used with an early stopping strategy.

[0141] The output layer is a fully connected layer that outputs the phase value at a preset future time. In a specific implementation, the length of the preset future time is configured to cover the total system delay time from physiological signal acquisition to stimulus output.

[0142] As an optional implementation, when the waveform asymmetry index falls within a preset sleep range, the center frequency of the slow wave oscillation component is modulated.

[0143] In specific implementation, the preset sleep interval is determined by analyzing all slow waves in the sleep EEG data that are clearly marked as stable deep sleep periods, calculating the waveform asymmetry index of each slow wave, and determining the statistical range of the concentrated distribution of these waveform asymmetry index values.

[0144] For example, the range covered by a certain multiple of the standard deviation of the waveform asymmetry index value is taken as the preset sleep interval.

[0145] In practice, all slow-wave samples labeled as stable deep sleep are extracted from the training set, and the normalized waveform asymmetry index of each sample is calculated.

[0146] The KS test was used to verify that the exponential distribution conforms to a normal distribution (P > 0.05), and the mean and standard deviation of the distribution were calculated.

[0147] In practice, the KS test is the Kolmogorov-Smirnov test, a nonparametric statistical test method used to determine whether a set of sample data conforms to a specific theoretical distribution (such as the normal distribution). The P-value is used to quantify the degree of inconsistency between the observed data and the null hypothesis.

[0148] For example, if the p-value calculated by the KS test is greater than a preset level (e.g., 0.05), it indicates that the difference between the current data and the normal distribution is not statistically significant, that is, the data follows a normal distribution.

[0149] The preset sleep interval can be defined as a range of values ​​centered on the mean, extending to both sides by a preset number of standard deviations.

[0150] The waveform asymmetry index calculated in real time is compared with the preset sleep interval corresponding to the current sleep stage. If the index falls within the preset interval, it is double-confirmed that the current sleep state is stable and typical (such as a stable deep sleep period), and then the phase prediction is officially performed.

[0151] Conversely, if the index deviates from the preset range, it indicates that the current sleep state may be unstable or atypical, and the modulation strategy should be adjusted.

[0152] The phase value at the preset future time is taken as the target phase state of the composite stimulus signal;

[0153] The frequency of the modulated target EEG rhythm is used as the rhythm reference of the composite stimulation signal.

[0154] As an optional implementation, a first mapping channel is constructed to map the rising slope to the intensity growth parameters of the acoustic and optical stimulation signals;

[0155] In a specific implementation, the first mapping channel controls the sound-emitting device and the light-emitting device in the stimulation unit to output synchronous sound waves and modulated light signals within the time window corresponding to the rising edge component, and the rate of change of the sound pressure level of the sound wave and the rate of change of the intensity of the modulated light signal are both positively correlated with the rising slope; the modulated light signal is a light pulse sequence obtained by controlling the driving current of the light-emitting device so that the intensity (brightness) of its output light changes according to the waveform synchronized with the target EEG rhythm.

[0156] A second mapping channel is constructed to map the falling slope to the intensity attenuation rate of the magnetic field or vibration stimulation signal; the second mapping channel controls the magnetic field or vibration generating device in the stimulation unit to output an oscillating magnetic field or vibration within the time window corresponding to the falling edge component, and the attenuation rate of the intensity of the oscillating magnetic field or vibration over time is positively correlated with the falling slope.

[0157] In specific implementation, the first mapping channel and the second mapping channel are a set of predefined data processing rules and algorithms. They convert the raw slope value extracted from the EEG signal into an intensity change parameter with clear physical meaning that can be directly used to control external stimulation devices in real time. Specifically, the first mapping channel subtracts the historical minimum value of the parameter recorded in the system from the real-time extracted raw upward slope value, then divides it by the difference between the historical maximum and minimum values, normalizing it to the numerical range of [0, 1].

[0158] As an optional implementation, the final intensity growth parameter is calculated using different mathematical functions based on the numerical range of the normalized slope value:

[0159] When the normalized slope value is between zero and a preset first threshold, a quadratic function is used for calculation. The square value of the normalized slope value is calculated by taking the normalized slope value as the independent variable and then multiplying it by a preset acoustic or optical fundamental coefficient. This makes the intensity growth start smoothly when the slope is extremely low.

[0160] When the normalized slope value is between the first threshold and the preset second threshold, a linear function is used for calculation. The normalized slope value is directly multiplied by a preset acoustic or optical linear coefficient to ensure that the intensity growth and the slope maintain a stable proportional relationship.

[0161] When the normalized slope value is greater than the second threshold, the natural logarithm function is used for calculation. The normalized slope value is increased by one and then the natural logarithm is taken. Then it is multiplied by a preset acoustic or optical logarithm coefficient, so that when the slope is extremely high, the rate of intensity growth slows down, and the output is prevented from exceeding the safety threshold.

[0162] The first threshold and the second threshold are determined based on the actual slope.

[0163] The final outputs are the rate of change of sound pressure level and the rate of change of brightness, which are used as parameters for intensity growth.

[0164] The second mapping channel takes the absolute value of the original descent slope value extracted in real time, and then uses the same maximum-minimum method as the first channel to normalize it to a fixed numerical range of [0, 1].

[0165] The magnetic field stimulation intensity decay rate adopts the natural exponential decay function. The normalized descent slope value is multiplied by a preset magnetic field decay coefficient, and the negative of the product is taken as the exponent to calculate the natural constant e raised to the power of the exponent.

[0166] The calculation result represents the remaining proportion of the magnetic field strength per unit time, and its value is between [0, 1].

[0167] The actual decay rate is obtained by subtracting the remaining percentage from one. This function ensures that the decay rate is positively correlated with the rate of decline and that the decay process is smooth.

[0168] The vibration stimulus intensity attenuation rate adopts the hyperbolic tangent exponential attenuation function. The normalized descent slope value is multiplied by a preset vibration attenuation coefficient; the hyperbolic tangent function value of the product is calculated; and the negative of the hyperbolic tangent value is used as the exponent to calculate the natural constant e raised to the power of the exponent.

[0169] The actual attenuation rate is calculated in the same way as that for magnetic field stimulation.

[0170] As an optional implementation, the real-time collaborative decision-making model combines the optimal stimulus combination (including stimulus type, intensity, and waveform parameters) output from the predefined composite stimulus modal with the rising slope and falling slope robustly calculated from the EEG signal, and generates the digital control waveform sequence corresponding to each modality (light, sound, magnetism, vibration) in real time in the processing unit.

[0171] The predefined composite stimulus modality is a parameter database that stores pre-designed combinations of stimulus parameters.

[0172] As an optional implementation method, a database of the correspondence between slope intervals and stimulation parameters is established. The rising slope and falling slope are divided into three intervals: low, medium, and high according to historical physiological response characteristics. Each interval corresponds to a preset range of stimulation parameters, that is, low slope corresponds to low intensity stimulation, medium slope corresponds to medium intensity stimulation, and high slope corresponds to high intensity stimulation within the safety threshold.

[0173] For example, for acoustic stimuli, the rising slope corresponds to the increase in sound pressure level; a larger slope results in a smoother increase in sound pressure level, avoiding overload. For optical stimuli, the rising slope corresponds to the brightness adjustment rate; a smaller slope results in a slower adjustment rate, ensuring gentle triggering. For magnetic field stimuli, the falling slope corresponds to the decay rate of the magnetic field strength; a larger slope results in faster decay. For vibration stimuli, the falling slope corresponds to the reduction in vibration amplitude; a smaller slope results in slower decay. The real-time collaborative decision-making model establishes an action set containing combinations of various stimulus modes.

[0174] Based on the current state parameters, including the target phase state, the intensity growth parameter, the intensity decay rate, and the waveform asymmetry index, the expected phase-locked loop benefit value for each action is calculated.

[0175] As an optional implementation, the current state parameter and the starting state parameter of each record in the historical database are normalized respectively, so that all parameter values ​​are converted to a unified dimension between 0 and 1.

[0176] In practice, the weights of each state parameter are determined by the analytic hierarchy process (AHP). The judgment matrix is ​​constructed with the goal of improving the phase-locked loop effect as the target layer and the target phase state as the criterion layer. The relative importance of each criterion is determined by expert scoring (1-9 scale method).

[0177] The eigenvalue decomposition of the judgment matrix yields the weight vectors of each criterion.

[0178] For example, a consistency ratio of <0.1 is required to ensure the rationality of weight allocation.

[0179] For each candidate stimulus action in the action set, retrieve all relevant records of the action that has been performed from the historical database to form a set of records to be analyzed;

[0180] For each historical record in the set, calculate the absolute difference between the current normalized state parameter and the corresponding parameter value in the historical record. Then, multiply the difference of each parameter by its importance weight determined in advance by the analytic hierarchy process to obtain the weighted difference value of the parameter. Finally, add up the weighted difference values ​​of all parameters to obtain a comprehensive difference value that represents the overall difference between the current state and the historical state.

[0181] As an alternative implementation, the composite difference value is processed using a negative exponential transformation function, which maps it to a similarity score.

[0182] In specific implementation, the negative exponential conversion function is calculated with the natural constant e as the base and the product of the negative comprehensive difference value and a preset normal decay coefficient as the exponent. This function ensures that when the comprehensive difference value is zero, the similarity score is one, indicating complete similarity. As the comprehensive difference value increases, the similarity score decreases exponentially from the beginning and approaches zero.

[0183] Finally, the similarity score corresponding to each historical record is multiplied by the actual phase-locking effect index obtained after performing the action to obtain a series of weighted effect values; all weighted effect values ​​are added together and then divided by the sum of all similarity scores, and the quotient is the expected phase-locking benefit value of the candidate action.

[0184] Using the expected phase-locked loop return value as the optimization objective, the action with the highest expected phase-locked loop return value is selected as the optimal stimulus combination output by the real-time collaborative decision-making model.

[0185] The generated digital control waveform sequence is sent to the corresponding mode digital-to-analog converter in real time and converted into a continuously changing analog voltage signal.

[0186] For example, the light-emitting device may be a programmable light-emitting diode array or a low-intensity light therapy module, the emission wavelength, intensity and pulse frequency of which are modulated by a microcontroller;

[0187] The generating device can be a small speaker or bone conduction headphones, driven by an audio decoding chip and a power amplifier, and can output sound waves with specific frequency, envelope and sound pressure level;

[0188] The magnetic field generating device can be a coil array that generates alternating or specific waveform weak magnetic fields, and its magnetic field strength, frequency and timing can be controlled by a current driving circuit;

[0189] Vibration generating devices can be linear resonant actuators or miniature eccentric motors, integrated into mattresses, pillows or wearable devices, and their vibration intensity and mode can be controlled by pulse width modulation signals;

[0190] Based on the optimal stimulus combination and the real-time delay estimates of each modal stimulus channel, the composite stimulus command is calculated in reverse. The composite stimulus command includes: the pre-compensation trigger time of each modal stimulus and the optimal stimulus combination.

[0191] Based on the composite stimulus instruction, each modal stimulus is output synchronously at the pre-compensation trigger time;

[0192] For example, the multimodal stimulation generator independently and synchronously controls the aforementioned devices through their respective drive circuits according to the composite stimulation command.

[0193] As an optional implementation, a synchronous test command is issued during the stimulation phase, and additional sensors (such as photodetectors, microphones, magnetometers, and accelerometers) are used to measure the time elapsed from the issuance of the command to the actual intensity of each modal stimulus reaching the preset intensity at the user's perceived location.

[0194] The delay times obtained from these measurements (including circuit response, signal transmission, device startup time, etc.) are stored as real-time delay estimates for each channel.

[0195] When generating the composite stimulus command, the pre-compensation trigger time is obtained by calculating the preset future time and the real-time delay estimate to ensure that the stimulus can reach the expected intensity and achieve synchronization.

[0196] In practice, the pre-compensation trigger time is determined by calculating the difference between the preset future time and the real-time delay value of each modal stimulus.

[0197] The phase-locked loop performance indicators include at least: phase synchronization indicators, slow wave activity energy change indicators, and waveform asymmetry index improvement indicators.

[0198] The phase-locked loop (PLL) performance index is generated by weighting and comprehensively scoring the phase synchronization index, the slow wave activity energy change index, and the waveform asymmetry index improvement index.

[0199] The change index of slow wave activity energy is obtained by dividing the growth rate of walk activity energy by the maximum positive growth rate observed from all user data during the training phase to obtain a preliminary ratio value. This ratio value is then smoothly compressed and mapped to the interval between 0 and 1 using the Sigmoid function as its normalization result.

[0200] The growth rate of the slow wave activity energy is obtained by extracting the slow wave activity energy of the target frequency band within the same preset time period from the EEG signals before and after the stimulation is applied; and by subtracting the average energy value before the stimulation from the average energy value after the stimulation, and then dividing by the average energy value before the stimulation.

[0201] The improvement index of the waveform asymmetry index is calculated by subtracting the absolute value of the difference between the index value before stimulation and the ideal value from the absolute value of the difference between the index value after stimulation and the ideal value, and then dividing the result by the absolute value of the difference between the index value before stimulation and the ideal value.

[0202] For example, the improvement index of the waveform asymmetry index is greater than 0 to indicate improvement, less than 0 to indicate deterioration, and equal to 0 to indicate no change;

[0203] As an optional implementation, the normalization process uses a piecewise function. For original values ​​less than or equal to 0, they are directly mapped to a normalized value of 0. For original values ​​greater than 0, they are divided by 1 and added to the original value for calculation, so that the normalized value increases monotonically as the original improved value increases and gradually approaches 1.

[0204] As an optional implementation method, a dynamic adaptive weight allocation method is adopted. Based on the physiological regulation goals of the current sleep stage, a set of basic weights are assigned to each sub-indicator. Secondly, by periodically analyzing the correlation between historical data of each indicator and the user's subjective sleep quality, a personalized efficacy coefficient reflecting the individual response characteristics is calculated. Finally, the basic weights are multiplied by the corresponding efficacy coefficients and normalized to obtain a set of dynamic weight values ​​that sum to 1, which are used for the weighted summation of the indicators.

[0205] In practice, a set of basic weights are assigned to the three indicators according to the preset priority rules;

[0206] For example, when the goal of the current sleep stage is to induce and enhance deep sleep, the importance of the indicators is ranked as follows: slow wave activity energy change indicators > phase synchronization indicators > waveform asymmetry index improvement indicators.

[0207] When the current sleep stage goal is to stabilize and maintain circadian rhythm synchronization, the ranking is adjusted to phase synchronization index > waveform asymmetry index improvement index > slow wave activity energy change index.

[0208] The weights are assigned according to a preset basic weight ratio, which is set according to the specific goals of the project. The basic weights of the three indicators are added together to form 1, and the initial basic weight ratio is 1.

[0209] As an optional implementation, the normalized values ​​of the three indicators after each intervention, as well as the subjective sleep quality score (e.g., a scale of 0-10) provided by the user the following morning, are continuously stored and periodically (e.g., after every 7 days of accumulated effective sleep) are rolled over.

[0210] For each indicator, calculate the Pearson correlation coefficient between its numerical sequence over the past period and the user's subjective sleep quality score sequence over the same period.

[0211] The calculated correlation coefficients are shifted and scaled to ensure they are positive, resulting in a personalized efficacy coefficient for each indicator. The larger the coefficient, the stronger the correlation between the historical changes of the indicator and the improvement in sleep quality felt by the user.

[0212] The basic weights of each indicator are multiplied by the personalized performance coefficients, and then the three products are summed to obtain the final phase-locked loop (PLL) performance index. In practice, the phase prediction model is adjusted based on whether the sleep stage has changed and whether the PLL performance index exceeds a preset threshold.

[0213] See Figure 4 The diagram shown is a flowchart of the dynamic evaluation and optimization of the phase prediction model provided in this application embodiment, including:

[0214] The parameters of the phase prediction model are adjusted according to the phase-locked effect index using a preset optimization algorithm in order to update the subsequent composite stimulus signal;

[0215] Construct a mapping relationship between the parameter combinations of the phase prediction model and the corresponding phase-locked loop performance indicators within the historical period;

[0216] Based on the mapping relationship, with the phase-locked loop performance index as the optimization target, a preset optimization algorithm is used to search for the optimal combination of parameters that makes the phase-locked loop performance index the best.

[0217] The phase prediction model is updated using the optimal parameter combination, and the phase prediction of the target EEG rhythm and the generation of compound stimulation signals are continued.

[0218] As an alternative implementation, a Bayesian optimization algorithm is used to construct a probabilistic model based on historical data, and to balance deep search in parameter regions with known good performance with trial exploration in parameter regions that have not been fully explored in the parameter space, thereby recommending a new parameter combination that is optimal for improving phase-locked loop performance.

[0219] As an optional implementation, a GPR surrogate model is trained based on the current training data, and the L-BFGS algorithm is used to solve for the maximum value of the EI function. The optimal candidate parameter combination is then searched from the parameter combinations.

[0220] The candidate parameter combination and the measured values ​​are added to the training data to update the surrogate model. This process is repeated until the number of iterations reaches the preset number or the objective function value converges, resulting in a new parameter combination.

[0221] In practice, the loss function of Bayesian optimization is defined as the mean square error between the predicted value and the measured value of the surrogate model.

[0222] The measured values ​​are the actual phase-locking effect data obtained by quantifying the feedback physiological signals after the parameter combination is actually applied. The predicted values ​​are the estimates of the phase-locking effect of the new parameter combination made by the surrogate model based on the mapping relationship between historical parameter combinations and corresponding measured values.

[0223] A new round of intervention trials will be conducted based on the new parameter combination, and the trial results will be added to the historical database to update the probability model.

[0224] Based on the sleep stage and the phase-locked loop effect index, generate environmental control commands;

[0225] The relationship between the phase-locked loop performance index and the preset threshold is compared, and the smart home devices are controlled by the environmental control commands.

[0226] In practice, the system automatically controls smart home devices such as lights and temperature based on real-time sleep stages, and dynamically fine-tunes environmental parameters or sends status prompts to the monitoring terminal based on whether the phase-locked loop effect indicators meet the standards.

[0227] In practice, each user's sleep stage determination model, phase prediction model, historical performance data, and personal environmental settings (such as sleep temperature and lighting preferences) are stored and loaded independently.

[0228] As an optional implementation, the family guardianship function is supported, but this requires strict access control.

[0229] For example, the primary user, acting as an "administrator," can authorize a specific relative's account to act as a "guardian."

[0230] The "guardian" can view the sleep stage summary, phase-locked effect index trend and system prompts of the monitored person through their terminal application. With the prior authorization of the "administrator", they can perform limited operations, such as manually adjusting environmental equipment or confirming prompts, but they have no right to modify core intervention model parameters or personal in-depth physiological data.

[0231] In practice, all operations are logged to ensure traceability and privacy.

[0232] If the technical solution disclosed herein involves personal information, the product using this technical solution has clearly informed the user of the personal information processing rules and obtained the user's voluntary consent before processing the personal information. If the technical solution disclosed herein involves sensitive personal information, the product using this technical solution has obtained the user's separate consent before processing the sensitive personal information, and also meets the requirement of "express consent". For example, at personal information collection devices such as cameras, clear and prominent signs are set up to inform users that they have entered the scope of personal information collection and that personal information will be collected. If an individual voluntarily enters the collection scope, it is deemed that they have agreed to the collection of their personal information; or on the personal information processing device, with clear signs / information informing users of the personal information processing rules, authorization is obtained from the individual through pop-up information or by asking the individual to upload their personal information; wherein, the personal information processing rules may include information such as the personal information processor, the purpose of personal information processing, the processing method, and the types of personal information processed.

[0233] Those skilled in the art will understand that in the system described above in the specific implementation, the order in which the steps are written does not imply a strict execution order and does not constitute any limitation on the implementation process. The specific execution order of each step should be determined by its function and possible internal logic. It should be understood that determining B based on A does not mean determining B solely based on A; B can also be determined based on A and / or other information.

[0234] In the description of this specification, the terms "exemplary," "for example," "specifically," etc., refer to a specific feature, structure, material, or characteristic described in connection with that embodiment or example, which is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

Claims

1. A sleep aid system based on photoacoustic-magnetic vibration wave resonance, characterized in that, The system includes: The acquisition unit is used to acquire the user's physiological signals in real time, analyze the physiological signals, and determine the current sleep stage; The processing unit, in response to the sleep stage being a preset target sleep stage, extracts the instantaneous phase of the target EEG rhythm from the physiological signal, and predicts the phase value of the target EEG rhythm at a preset future time through a phase prediction model. The stimulation unit is equipped with a multimodal stimulation generator, which can generate at least one of light, sound, magnetic, and vibration signals; the stimulation unit controls the multimodal stimulation generator to output a composite stimulation signal that is phase-locked with the target EEG rhythm based on the phase value of the preset future time. The evaluation feedback unit is used to collect feedback physiological signals after the application of the composite stimulation signal, calculate the phase-locking effect index, and adjust the parameters of the phase prediction model according to the phase-locking effect index using a preset optimization algorithm to update subsequent composite stimulation signals.

2. The sleep aid system based on photoacoustic-magnetic vibration wave resonance according to claim 1, characterized in that, The phase value of the target EEG rhythm at a preset future time includes: Extract at least two different time-scale EEG rhythm components associated with the current sleep stage; A cross-scale coupling analysis was performed on the instantaneous phase sequence of the EEG rhythm component to obtain the phase evolution constraints of the target EEG rhythm component. Based on the phase evolution constraints and the phase change trajectory of the target EEG rhythm within a preset time period, the phase value of the target EEG rhythm at the preset future time is predicted.

3. The sleep assistance system based on photoacoustic-magnetic vibration wave resonance according to claim 2, characterized in that, The extraction of at least two different time-scale EEG rhythm components associated with the current sleep stage includes: The electroencephalogram (EEG) signal in the physiological signal was filtered in parallel using a dual-channel bandpass filter to obtain the slow wave oscillation component of the first frequency band and the sleep spindle wave component of the second frequency band, respectively. The Hilbert transform is applied to the slow-wave oscillation component to obtain the instantaneous phase sequence of the slow-wave oscillation component analytically. The amplitude envelope of the sleep spindle wave component is extracted, and the maximum value of the amplitude envelope within a preset time window is identified. Based on the maximum value moment, the corresponding slow wave instantaneous phase value is indexed from the instantaneous phase sequence on the same time axis; Calculate the phase deviation between the slow wave instantaneous phase value and the preset standard coupling phase threshold; Based on the phase deviation, the center frequency of the slow wave oscillation component is modulated, and the modulated frequency is determined as the frequency of the target EEG rhythm.

4. The sleep aid system based on photoacoustic-magnetic vibration wave resonance according to claim 3, characterized in that, Modulating the center frequency of the slow-wave oscillation component based on the phase deviation also includes verifying the slow-wave morphology before performing modulation: Extreme point detection is performed on the slow wave oscillation component to determine the trough and peak times within a single cycle; Based on the signal segment from the trough to the peak, a rising edge component is constructed; Based on the signal segment from the peak time to the next trough time, a falling edge component is constructed; The mean of the first derivative of the rising edge component is calculated as the rising slope, and the mean of the first derivative of the falling edge component is calculated as the falling slope. The ratio of the rising slope to the falling slope is calculated to obtain the waveform asymmetry index; When the waveform asymmetry index falls within a preset sleep range, the center frequency of the slow wave oscillation component is modulated.

5. The sleep aid system based on photoacoustic-magnetic vibration wave resonance according to claim 4, characterized in that, The step of controlling the multimodal stimulation generator to output a composite stimulation signal that is phase-locked with the target EEG rhythm based on the phase value of the preset future time includes: The phase value at the preset future time is taken as the target phase state of the composite stimulus signal; The frequency of the modulated target EEG rhythm is used as the rhythm reference of the composite stimulation signal; Construct the first mapping channel to map the rising slope to the intensity growth parameters of the acoustic and optical stimulation signals; A second mapping channel is constructed to map the descent slope to the intensity attenuation rate of the magnetic field or vibration stimulus signal; Based on the target phase state, the intensity growth parameter, and the intensity decay rate, an optimal stimulus combination is selected from predefined composite stimulus modes through a real-time collaborative decision-making model. The optimal stimulus combination includes: stimulus type, stimulus intensity, and waveform parameters. The optimal stimulus combination, along with the intensity growth parameter and intensity decay rate, is synthesized into a digitally controlled waveform sequence. Based on the digital control waveform sequence and the real-time delay estimate of each modal stimulation channel, the composite stimulation command is calculated in reverse. The composite stimulation command includes: the pre-compensation trigger time of each modal stimulus and the digital control waveform sequence. Based on the composite stimulus command, each modal stimulus is output synchronously at the pre-compensation trigger time.

6. The sleep assistance system based on photoacoustic-magnetic vibration wave resonance according to claim 5, characterized in that, The step of selecting the optimal stimulus combination from predefined composite stimulus modalities through a real-time collaborative decision-making model also includes: Based on the real-time collaborative decision-making model, an action set containing combinations of various stimulus modalities is established. Based on the target phase state, the intensity growth parameter, the intensity decay rate, and the waveform asymmetry index, calculate the expected phase-locked loop benefit value for each action; Using the expected phase-locked loop return value as the optimization objective, the action with the highest expected phase-locked loop return value is selected as the optimal stimulus combination output by the real-time collaborative decision-making model.

7. The sleep aid system based on photoacoustic-magnetic vibration wave resonance according to claim 5, characterized in that, The synchronous output of each modal stimulus at the pre-compensation trigger time based on the composite stimulus command includes: The pre-compensation trigger time is determined by calculating the difference between the preset future time and the real-time delay value of each modal stimulus.

8. The sleep aid system based on photoacoustic-magnetic vibration wave resonance according to claim 4, characterized in that, The process of collecting feedback physiological signals after applying the composite stimulation signal and calculating phase-locked effect indices includes: The phase-locked loop performance indicators include at least: phase synchronization indicators, slow wave activity energy change indicators, and waveform asymmetry index improvement indicators; The phase-locked loop effect index is generated by weighting and comprehensively scoring the phase synchronization index, the slow wave activity energy change index, and the waveform asymmetry index improvement index. Specifically, based on the composite stimulation signal and the feedback physiological signal, rhythms with the same frequency as the target EEG rhythm are extracted and their phase sequences are calculated. Calculate the synchronicity measure between two phase sequences to obtain the phase synchronization index; The change index of slow wave activity energy is obtained by calculating the relative change in the energy of the target EEG rhythm frequency band before and after the application of the composite stimulation signal; The improvement index of the waveform asymmetry index is obtained by calculating the degree to which the waveform asymmetry index approaches the preset sleep range before and after the application of the composite stimulation signal.

9. The sleep aid system based on photoacoustic-magnetic vibration wave resonance according to claim 1, characterized in that, The step of adjusting the parameters of the phase prediction model according to the phase-locked loop effect index using a preset optimization algorithm to update subsequent composite stimulation signals includes: Construct a mapping relationship between the parameter combinations of the phase prediction model and the corresponding phase-locked loop performance indicators within the historical period; Based on the mapping relationship, with the phase-locked loop performance index as the optimization target, a preset optimization algorithm is used to search for the optimal combination of parameters that makes the phase-locked loop performance index the best. The phase prediction model is updated using the optimal parameter combination, and the phase prediction of the target EEG rhythm and the generation of compound stimulation signals are continued.

10. The sleep aid system based on photoacoustic-magnetic vibration wave resonance according to claim 8, characterized in that, The generation of the phase-locked loop performance index also includes: Based on the sleep stage and the phase-locked loop effect index, generate environmental control commands; The relationship between the phase-locked loop effect index and the preset threshold is compared, and the smart home devices are controlled by the environmental control command, and a prompt message is sent to the preset monitoring terminal.

Images