An adaptive sleep-aiding control method and system for a sleep-aiding pillow based on pressure sensing

Abstract

The application provides a self-adaptive sleep-aiding control method and system based on pressure monitoring sleep-aiding pillows, relates to the technical field of sleep-aiding pillows, and aims at the problem that the prior art cannot distinguish the sleep drive state of a user, resulting in mismatch of intervention strategies. The application realizes real-time and accurate distinction of whether the user is in a sleep deficiency state or a sleep sufficiency state by constructing a sleep drive level accumulation-dissipation dynamics model and a circadian rhythm sleep tendency function and calculating the synchronism change rate (convergence rate) of the two, When the user is evaluated as being in a sleep deficiency state (natural drowsiness), the system automatically reduces or turns off the audio intervention to protect the natural sleep process; when the user is evaluated as being in a sleep sufficiency state but having difficulty falling asleep, the system actively provides sleep-aiding audio assistance to compensate for the insufficient drive. Thus, differentiated regulation based on physiological states is realized, inappropriate stimulation is avoided, and the sleep-aiding quality is significantly improved.

Description

Technical Field

[0001] This invention relates to the field of sleep aid pillow technology, specifically to an adaptive sleep aid control method and system based on pressure-sensing monitoring sleep aid pillow. Background Technology

[0002] With the rapid development of flexible sensing and artificial intelligence technologies, smart sleep-aid pillows based on pressure monitoring are gradually becoming an important direction for improving sleep quality. By embedding a pressure sensor array within the pillow, physiological signals such as head displacement, breathing fluctuations, and body movement frequency can be collected in real time. This allows for the identification of sleep stages and the activation of an acoustic module to play sleep-aiding audio, achieving non-drug and non-invasive sleep intervention.

[0003] For example, the invention patent with publication number CN110456846A discloses an artificial intelligence-based adaptive multi-sensory sleep assistance system. This system detects the user's sleep state through a pressure sensor and controls actuators such as a sound module to provide multi-sensory feedback. However, its audio control strategy only performs simple binary switch control based on the sleep state (turning off when falling asleep and turning on when waking). It fails to implement fine-grained adjustments such as gradual volume reduction and frequency changes based on the continuous evolution of the sleep state, and also lacks the ability to predict future changes in sleep stability.

[0004] For example, the invention patent with announcement number CN112704365B discloses a smart pillow and a sleep monitoring method, which determines insomnia and initiates sleep aid by comparing the time required to fall asleep with the historical average. However, this solution only focuses on abnormal sleep onset time and does not address the maintenance of sleep throughout the night or the guarantee of deep sleep. Furthermore, the pressure threshold judgment cannot distinguish between sub-stages such as light sleep and deep sleep, and the sleep aid control is also limited to simple on / off logic.

[0005] For example, the invention patent with announcement number CN119185732A discloses an auxiliary method for controlling ambient sound to improve sleep. It dynamically adjusts the ambient sound frequency and volume according to the real-time sleep state. However, its regulation is still a feedback control of "detecting the current state → adjusting the current parameters". It does not establish a predictive mechanism for future sleep evolution trends, nor does it set up an online parameter self-correction function based on nightly prediction errors.

[0006] Monitoring and intervention are disconnected, lacking refined control strategies based on the continuous evolution of sleep depth; intervention logic is mostly a post-event feedback response, unable to implement preventive interventions before micro-awakening occurs; system parameters rely on fixed thresholds or manual user settings, lacking closed-loop adaptive calibration capabilities based on individual differences and long-term use data; it cannot distinguish between the two physiological states of "insufficient sleep" and "sufficient sleep" to adapt differentiated strategies; and it lacks an adaptive response mechanism when signal quality fluctuates or the autonomic nervous system state changes, leading to a decline in long-term effectiveness.

[0007] Therefore, there is an urgent need to propose an adaptive sleep aid control method and system based on pressure-sensing monitoring sleep aid pillows. Summary of the Invention

[0008] To address the problems mentioned in the background art, the present invention provides the following technical solution: an adaptive sleep aid control method for a pressure-sensing sleep aid pillow, comprising the following steps: S1. Real-time acquisition of user's head pressure distribution, respiratory fluctuations, and body movement frequency data; after noise reduction processing, the user's current sleep state is identified and recorded as a time series; the sleep state includes wakefulness, falling asleep, light sleep, deep sleep, and wakefulness; S2. Based on the time series from step S1, the user's current sleep drive level is assessed in real time according to the sleep drive level accumulation-dissipation dynamics model. At the same time, the user's circadian rhythm phase is assessed based on the accumulated multi-night time series, and a circadian rhythm sleep tendency function with values ​​in the interval [0,1] is constructed that maps to the circadian rhythm phase. The synchronicity change rate between the sleep drive level and the circadian rhythm sleep tendency is calculated as the convergence rate. The sleep drive level accumulation-dissipation dynamics model includes an accumulation rate for controlling the speed of sleep drive accumulation and a dissipation rate for controlling the speed of sleep drive dissipation. S3. Based on the current sleep drive level, circadian rhythm phase and the convergence rate, predict the user's sleep vulnerability index within a future preset time window. The sleep vulnerability index, after saturation limitation, has a value range of [0,1]. S4. Based on the predicted sleep vulnerability index and the current sleep state, a two-dimensional coupled feedforward control method is adopted to adjust the audio output parameters of the acoustic module in advance before the user experiences micro-awakening; the two-dimensional coupling determines the strategy domain of audio modulation based on the sleep state and determines the timing and intensity of audio modulation based on the sleep vulnerability index, with joint decision-making between the two dimensions; the two-dimensional coupled feedforward control is implemented through an adaptive fuzzy inference system. S5. Compare the predicted sleep vulnerability index sequence for the night with the actual micro-awakening sequence for the night, calculate the prediction error, and update the accumulation rate and dissipation rate in step S2, as well as the model parameters used in step S3 to predict the sleep vulnerability index based on the prediction error.

[0009] An adaptive sleep aid control system based on a pressure-sensing monitoring sleep aid pillow includes: The pressure-sensing monitoring module is used to collect data on the user's head pressure distribution, breathing fluctuations, and body movement frequency in real time; The intelligent control chip, electrically connected to the pressure-sensitive monitoring module, incorporates a noise reduction algorithm and a sleep state recognition algorithm to identify the user's current sleep state and record it as a time series. The intelligent control chip also integrates: The dual-process state assessment unit is used to assess the user's sleep drive level and circadian rhythm phase in real time based on the time series, construct a circadian rhythm sleep tendency function that maps to the circadian rhythm phase, and calculate the rate of synchronicity change between the sleep drive level and the circadian rhythm sleep tendency as the convergence rate. A sleep vulnerability prediction unit is used to predict a sleep vulnerability index within a future time window based on the sleep drive level, circadian rhythm phase, and convergence rate. The sleep vulnerability index, after being saturated and limited, has a value range of [0,1]. The audio strategy scheduling unit is used to adjust the audio output parameters of the acoustic module in advance, based on the current sleep state and the predicted sleep vulnerability index, in a two-dimensional coupled feedforward control manner, before the user experiences micro-awakening. A self-correcting unit is used to update the dynamic parameters in the dual-process state assessment unit and the model parameters in the sleep vulnerability prediction unit based on the prediction error; An acoustic module, electrically connected to the intelligent control chip, is used to play sleep-aid audio according to the control instructions of the audio strategy scheduling unit; The communication module is electrically connected to the intelligent control chip and is used to realize data interaction with the mobile application. The mobile application communicates with the smart control chip through the communication module, and is used to receive personalized parameters set by the user and display sleep quality reports.

[0010] The present invention has the following beneficial effects: (1) To address the problem of intervention strategy mismatch caused by the inability of existing technologies to distinguish the user's sleep drive state, this invention constructs a sleep drive level accumulation-dissipation dynamic model and a circadian rhythm sleep tendency function, and calculates the synchronicity change rate (convergence rate) of the two, thereby achieving real-time and accurate differentiation between whether the user is currently in a state of sleep deprivation or sleep sufficiency. When the system assesses that the user is in a state of sleep deprivation (natural drowsiness), it automatically reduces or turns off audio intervention to protect the natural sleep-on process; when the system assesses that the user is in a state of sleep sufficiency but has difficulty falling asleep, it actively provides sleep-aid audio assistance to compensate for insufficient drive. This achieves differentiated regulation based on physiological state, avoids inappropriate stimulation, and significantly improves the quality of sleep aid.

[0011] (2) To address the problem that existing feedback control cannot proactively prevent micro-awakening, this invention adopts a two-dimensional coupled feedforward control method, which uses the current sleep state and the predicted sleep vulnerability index to jointly make decisions, adjusting the audio output parameters in advance before micro-awakening occurs. Simultaneously, differentiated strategies are designed for different sleep states: in the awake state, audio attenuation is dynamically adjusted based on the convergence rate; in the sleep state, asynchronous rhythmic acoustic stimulation is used to interfere with the synchronous accumulation of vulnerability oscillations, and a smooth removal mechanism is implemented; in the light sleep state, the intervention aggressive coefficient is calculated by combining the predicted vulnerability mean and the sleep architecture quality index, and override control is introduced to prevent strategy deadlock; in the deep sleep state, audio is completely shut off, restart is determined only by conditional probability, and the lowest effective volume is obtained through individualized closed-loop calibration. This feedforward control, combined with the state refinement strategy, effectively reduces the total number of micro-awakenings throughout the night, prolongs the duration of deep sleep, and avoids damage to the sleep structure caused by inappropriate intervention.

[0012] (3) In response to the problem that the system parameters are fixed and cannot adapt to individual differences and long-term physiological changes, the present invention sets up a self-correction unit. By comparing the error between the predicted vulnerability index and the actual micro-awakening sequence each night, the backpropagation algorithm is used to synchronously update the accumulation / dissipation rate of the sleep drive model and the vulnerability prediction weighting coefficient. This allows the parameters of the dynamic model and the prediction model to gradually converge to the optimal value for the individual user as the number of days of use increases, which significantly shortens the time required for system adaptation and adjustment, and ensures the stability and effectiveness of long-term use.

[0013] (4) To address the issue of decreased intervention reliability caused by signal quality fluctuations and high sympathetic nervous system states in complex usage environments, this invention further integrates a sleep state confidence assessment unit and an autonomic nervous system balance index monitoring mechanism into the system. When the signal-to-noise ratio or body movement amplitude leads to insufficient state confidence, a conservative strategy is automatically switched to avoid erroneous regulation; when an abnormally high level of sympathetic nervous system excitation is detected, a positive bias correction is applied to the predicted sleep vulnerability index to enhance the intervention intensity in advance. The above mechanisms, in conjunction with the core sleep aid strategy, enable the system to perform precise intervention when the signal quality is good, ensure operational safety and adaptively increase the intervention level under signal fluctuations or physiological stress, and comprehensively enhance the robustness, accuracy, and environmental adaptability of the system.

[0014] Of course, any product implementing this invention does not necessarily need to achieve all of the above advantages at the same time. Attached Figure Description

[0015] Figure 1 This is a flowchart of the method of the present invention.

[0016] Figure 2 This is a flowchart of the system of the present invention.

[0017] In the diagram: 1. Pressure sensing module; 2. Intelligent control chip; 3. Acoustic module; 4. Communication module; 5. Mobile application. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Example 1 This embodiment provides an adaptive sleep aid control method based on a pressure-sensing monitoring sleep aid pillow. This method is implemented in a sleep aid pillow comprising a pressure-sensing monitoring module 1, a smart control chip 2, and an acoustic module 3. The smart control chip 2 is electrically connected to the signal output terminal of the pressure-sensing monitoring module 1 and the control input terminal of the acoustic module 3 via a flexible printed circuit board. The acoustic module 3 is a bone conduction speaker or loudspeaker, embedded in a soundproof cavity on the side of the pillow.

[0020] Its operation process is as follows: Step S1: The intelligent control chip 2 receives head pressure distribution data, respiratory fluctuation waveform data, and body acceleration data from the pressure-sensing monitoring module 1 in real time at a sampling frequency of 128Hz. These data include values ​​from at least 64 pressure sensing points on the head. The intelligent control chip 2's built-in wavelet transform-based denoising algorithm (using a Daubechies 4th-order wavelet basis with a decomposition level of 5) processes the raw data. After denoising, the hidden Markov model running on the intelligent control chip 2 identifies the user's current sleep state (including wakefulness, falling asleep, light sleep, deep sleep, and wakefulness) in 30-second time windows, and records the identified state sequence as time-series data, storing it in the intelligent control chip 2's built-in 4Mbit non-volatile memory. This continuous monitoring and identification process enables continuous, low-latency monitoring of the user's sleep state.

[0021] Step S2: Based on the sleep state time series generated in step S1, the dual-process state evaluation unit in the intelligent control chip 2 evaluates the user's current sleep drive level in real time. The evaluation is based on the sleep drive level accumulation-dissipation dynamics model:

[0022] The meanings of each item are as follows: S(t): Sleep drive level at time t, dimensionless, initialized to 0, and restricted to an interval in the calculation. When the value is below 0, take 0; when the value is above 0, take 0. Pick ; Maximum sleep drive level, with a value of 1.0; t: The time from the current moment to the last awake state, in seconds, is obtained through the internal timer of the intelligent control chip 2. The moment of the "last awake state" is given by the awake state in the sleep state sequence identified in step S1. : Cumulative rate, which controls how quickly the level of sleep drive rises. The initial value is set at 3600s, based on population statistics from 100 healthy adults. e: natural constant; : Sum of all micro-awakenings that occurred that night, with the subscript i indicating the i-th micro-awakening; The intensity of the i-th micro-awakening is dimensionless, ranging from 0 to 1, and is calculated from body acceleration data using the following formula: ; This is the peak body acceleration. Maximum acceleration (g is the acceleration due to gravity); The time of the i-th micro-awakening, in seconds; Dissipation rate controls how quickly the sleep drive level recovers after each micro-awakening, with an initial value set at 900s, also derived from population statistics.

[0023] Simultaneously, the dual-process state assessment unit, based on the accumulated sleep state time series of 7 consecutive nights, fits the user's circadian rhythm phase using a cosine function model. The model expression is:

[0024] in: y: Normalized circadian rhythm signal (representing diurnal fluctuations in body temperature or alertness), dimensionless, range [-0.5, 0.5]; t: Time, in seconds, starting from midnight each day; 86400: The number of seconds in a day (24h × 3600s), used to convert time into degrees; φ: Circadian rhythm phase, in rad, obtained by fitting historical data, and considered a constant value during a single night of sleep; : The phase angle of the cosine function.

[0025] Subsequently, a normalized circadian rhythm sleep tendency function was constructed. , Mapping it to the interval [0,1] aligns with the range of sleep drive level S(t). To avoid the problem of the absolute value function being non-differentiable, the synchronization index is defined using the continuously differentiable difference of squares form.

[0026] λ(t)=1-[S(t)-C(t)]².

[0027] The synchronization index reaches its maximum value of 1 when S and C are perfectly synchronized; the greater the difference, the smaller λ becomes. The convergence rate v is defined as the derivative of λ with respect to the dimensionless time τ:

[0028] v = dλ / dτ, where τ = t / 3600s. Expanding, we get...

[0029] This expression is continuously differentiable everywhere. A positive value represents that the driving force and rhythmic tendency are becoming synchronized, while a negative value represents a divergence. Its unit is h⁻¹.

[0030] In practical discrete system implementations, the above continuous model is discretized at 30-second intervals. When a micro-awakening event is detected, the consumption item... This introduces an instantaneous step. To ensure the computational continuity and differentiability of the synchronization exponent λ(t) and the convergence rate v, this embodiment applies a first-order exponential smoothing to S(t) between every two adjacent computation steps, with a smoothing time constant of 30s. This makes S(t) a continuous function in the time domain, thus making the derivatives of λ(t) and v valid in the computational domain. Furthermore, in the specific code implementation, v can also be directly expressed in discrete-domain difference form. The two calculations are mathematically equivalent and both can accurately reflect the dynamics of approaching or diverging.

[0031] The above calculation process transforms the abstract sleep dynamics into quantifiable physiological parameters.

[0032] Step S3: Based on the current sleep drive level S(t), circadian rhythm phase φ, and convergence rate v calculated in step S2, the sleep vulnerability prediction unit in the intelligent control chip 2 uses a weighted model to predict the user's sleep vulnerability index V within the next 60 minutes (preset time window). The prediction formula is:

[0033] The meanings of each item are as follows: α, β, γ: Adjustable weighting coefficients, which are reparameterized to ensure that α+β+γ=1 is always satisfied. The initial values ​​are 0.5, 0.3, and 0.2, respectively. 1-S(t): The degree of insufficient sleep drive level. The closer S(t) is to 0 (insufficient drive), the larger this term is, and the higher the vulnerability. The ideal sleep phase is π / 2 rad. : The absolute deviation of the current phase from the ideal sleep phase, in rad; π: Pi, used to normalize phase deviation to the [0,1] interval; v: the convergence rate calculated in step S2, in h⁻¹; The reference convergence rate is taken as 0.2h⁻¹, which is used to make v dimensionless; :Pick The maximum value of -0.9 ensures that the denominator 1 + max(·) is always greater than 0.1, avoiding division by zero or negative denominators; The third overall term: the greater the convergence rate (the greater v), the smaller this term, meaning the more stable the sleep; when the convergence rate is too negative, this term approaches 1, and vulnerability increases.

[0034] After obtaining V, a saturation constraint is applied: if V>1, it is set to 1; if V<0, it is set to 0, ensuring that the vulnerability index is strictly within the range of [0,1]. The higher the V value, the more vulnerable the sleep is and the more prone it is to micro-arousals.

[0035] This predictive step enables a forward-looking quantitative assessment of future sleep stability.

[0036] Step S4: Based on the current sleep state identified in Step S1 and the sleep vulnerability index V predicted in Step S3, the audio strategy scheduling unit in the intelligent control chip 2 adopts a two-dimensional coupled feedforward control method to adjust the audio output parameters of the acoustic module 3 in advance through an adaptive fuzzy inference system. This adaptive fuzzy inference system takes the current sleep state (encoded as: awake state = 0, falling asleep state = 1, light sleep state = 2, deep sleep state = 3, awake state = 4) and the V value (divided into four intervals: [0, 0.25] low vulnerability, [0.25, 0.5) medium-low vulnerability, [0.5, 0.75) medium-high vulnerability, [0.75, 1] ​​high vulnerability) as dual inputs, and the volume adjustment step size (in dB) and frequency offset (in Hz) as outputs. The membership function of the adaptive fuzzy inference system adopts a bell-shaped membership function, and its fuzzy rule base includes the following rule examples:

[0037] Rule 1: If the sleep state is light sleep (encoded value 2) and V is medium to high vulnerability ([0.5, 0.75)), then the volume adjustment step size is −0.5dB and the frequency offset is +15Hz; Rule 2: If the sleep state is light sleep (encoded value 2) and V is highly vulnerable ([0.75,1]), then the volume adjustment step size is −1.0dB and the frequency offset is +30Hz; Rule 3: If the sleep state is deep sleep (encoded value 3), the volume adjustment step is 0dB (audio output is off) and the frequency offset is 0Hz.

[0038] The consequent parameters of the above rules are iteratively updated by the self-correcting unit using the backpropagation algorithm based on the nightly prediction error.

[0039] The audio policy scheduling unit executes differentiated audio control strategies for each of the five sleep states: When the user is awake: The audio policy scheduling unit calculates the dimensionless convergence rate in step S2. The frequency reduction slope and volume decay rate of the modulated audio output parameters are adjusted. When v > 0.2h⁻¹, it is determined to be accelerated, and the synchronous acceleration adjustment is set, that is, the frequency reduction slope is set to −0.1Hz / min, and the volume attenuation rate is set to −0.3dB / min; when When v < 0.05h⁻¹, it is considered a slowdown. Adjustments are paused, and the duration of the current audio output parameter is extended by 5 minutes. The convergence rate is continuously monitored. When the convergence rate returns to the normal range (e.g., ...), the adjustment is stopped. When ), resume the adjustment process.

[0040] When the user is asleep: The audio policy scheduling unit calculates the time gradient of the sleep vulnerability index V. If an oscillation pattern with a period in the range of 0.05Hz to 0.1Hz is detected by Fast Fourier Transform, it indicates that sleep vulnerability is fluctuating periodically and there is a risk of micro-arousals. At this time, rhythmic auditory stimulation is generated to interfere with this oscillation pattern: a sound pulse train with a Delta wave (0.5-4Hz) as the carrier and an envelope frequency close to the detected oscillation pattern frequency but with random phase jitter is output and superimposed on the current audio output through acoustic module 3. This asynchronous rhythmic stimulation reduces the oscillation amplitude by increasing the complexity of sensory input and disrupting the synchronization accumulation of endogenous micro-arousal oscillations. The oscillation amplitude is defined as the envelope extracted from V after a 0.05-0.1Hz bandpass filter. When it is below a preset recovery threshold for five consecutive cycles (e.g., envelope peak-to-peak value < 0.1), a linear fade-out process is triggered, smoothly attenuating the gain of the rhythmic stimulation to zero within 30 seconds before removal, avoiding auditory discomfort caused by signal abrupt changes.

[0041] When the user is in a light sleep state: The audio policy scheduling unit predicts the vulnerability evolution trajectory within a 30-minute time window and takes the average value as... The sleep architecture quality index Q (defined as the ratio of deep sleep duration to total sleep duration, with Q = max(actual ratio, 0.05) set to avoid division by zero) was calculated for the night. An intervention radical coefficient was used. As a decision-making basis, when R is higher than the preset aggressive threshold of 1.5, the current audio output parameters are locked (i.e., the current volume and frequency remain unchanged); when R is lower than the preset mild threshold of 0.6, the audio output is gradually reduced at a rate of −0.2dB / min until it is turned off; when R is in the transition range of 0.6 to 1.5, the current audio output parameters are maintained and the prediction window is shortened to 15 minutes. In addition, to avoid policy deadlock, a flight control is introduced: when the sleep vulnerability index V exceeds the preset emergency threshold of 0.9, regardless of the R value, the audio volume is immediately and slowly increased (step +0.3dB / min) and switched to low-frequency Theta wave sleep aid sound until V falls back below 0.9 or reaches the safe volume limit. Once V falls back below 0.9, the system immediately recalculates the current R value and smoothly switches to the lock, maintain, or reduce strategy according to the R value to avoid frequent policy jumps.

[0042] When the user is in deep sleep: The audio policy scheduling unit shuts off the audio output of acoustic module 3. Simultaneously, it continuously monitors the sleep vulnerability index V, combined with the sleep state transition frequency. (Unit: h⁻¹, counting the number of times the sleeper transitions from deep sleep to light sleep or wakefulness within the last hour) Calculate the conditional probability of micro-awakening:

[0043] in: V: Current Sleep Vulnerability Index; Sleep state transition frequency, in hours (h⁻¹); ΔT: An empirical constant, taken as 1h, to make the exponential range... Dimensionless; exp: Exponential function.

[0044] when When the probability exceeds a preset threshold of 0.6, the audio output is restarted at the lowest effective volume. The initial value of the lowest effective volume can be set to 25dB, and its individualized calibration is achieved through closed-loop feedback: based on whether micro-awakening is triggered after each audio restart, the volume value is iteratively adjusted until it eventually converges to an individualized optimal value that can effectively maintain the auditory masking effect without triggering awakening.

[0045] When the user is awake: The audio strategy scheduling unit assesses the physiological conditions for falling back asleep, including the current sleep drive level S(t) > 0.6 and circadian rhythm phase deviation. When physiological conditions are met, the optimal sleep onset time window is calculated (i.e., the time interval within the predicted next 20 minutes where the sleep vulnerability index V is below 0.4), and sleep guidance audio (pink noise with a frequency of 1Hz to 8Hz) is started 5 minutes before the arrival of this window.

[0046] The audio modulation strategies for the five sleep states mentioned above share a self-correcting loop: the prediction error in step S5 simultaneously updates the cumulative rate in step S2. Dissipation rate The rule consequent parameters of the adaptive fuzzy inference system in step S4.

[0047] This two-dimensional coupled feedforward control is executed before the user experiences micro-awakening, enabling early intervention for potential sleep disturbances.

[0048] Step S5: After the night's sleep ends, the intelligent control chip 2 compares the sleep vulnerability index sequence predicted in step S3 with the actual micro-arousal sequence parsed from the body motion acceleration data for the night, and calculates the root mean square prediction error. This error is defined as the root mean square difference between the predicted vulnerability index and the actual micro-arousal intensity (marked as 1 for micro-arousals and 0 for others). Based on this error, the backpropagation algorithm is used to jointly update the dynamic parameters in step S2. , And the weighting coefficients α, β, and γ in the prediction model of step S3. The update rule is:

[0049] Similarly, update β and γ, and perform a normalization operation after each update. This ensures that the sum of the weighting coefficients is always 1.

[0050] in: η: Learning rate, taken as 0.01, dimensionless; , The partial derivatives of the error with respect to the dynamic parameters are obtained through finite differences; The partial derivative of the error with respect to the weighting coefficients is obtained by inversely calculating the expression for V using the chain rule.

[0051] If after the update Then set it to 600s, if The time limit is set to 300 seconds to maintain physiological rationality. This self-correction step forms a closed self-correction loop, which fully propagates the error gradient to all parameters to be optimized, allowing the model parameters to gradually personalize as the usage time increases.

[0052] The technical problem solved by this embodiment: Current sleep aid systems employ a uniform audio intervention strategy for all users, failing to differentiate interventions based on each user's real-time sleep physiological state. When users are sleep-deprived, their sleep drive level is high, exhibiting a strong tendency to fall asleep naturally under normal sleep conditions. In this case, applying sleep-aid audio can interfere with their natural sleep-on process, inducing unnecessary micro-awakenings. Conversely, when users are well-rested, their sleep drive level is low, facing typical problems such as difficulty falling asleep and weak sleep maintenance, urgently requiring active sleep aids to compensate for their insufficient sleep drive. Existing technologies lack the ability to accurately identify and differentiate between these two real-time physiological states, failing to adapt intervention strategies based on whether the user is in a state of "natural drowsiness" or "needs active sleep aids," resulting in poor or even counterproductive sleep aid intervention effects.

[0053] The working principle of this embodiment: This embodiment uses the pressure-sensing monitoring module 1 to collect real-time head pressure distribution data, respiratory fluctuation waveform data, and body motion acceleration data. Combined with the wavelet transform-based noise reduction algorithm and hidden Markov model built into the intelligent control chip 2, continuous and low-latency recognition of sleep state is achieved. The dual-process state assessment unit performs two assessments simultaneously: quantifying the current sleep drive level based on the sleep drive level accumulation-dissipation dynamics model, which directly reflects whether the user is currently in a state of insufficient or sufficient sleep; and determining the user's physiological time window by fitting the circadian rhythm phase using a cosine function model based on continuous multi-night sleep time series. Furthermore, a circadian rhythm sleep tendency function is constructed, using a smoothed continuous synchronization index λ and its derivative convergence rate v to accurately quantify the convergence or divergence dynamics between sleep drive and circadian rhythm, comprehensively determining whether the user is currently in a state of natural drowsiness or requires active sleep aid. The sleep vulnerability prediction unit predicts the sleep vulnerability index V within the future time window based on the above multi-dimensional physiological parameters, and ensures that V∈[0,1] through saturation constraints. The audio strategy scheduling unit uses the current sleep state and V as dual inputs, employing a two-dimensional coupled feedforward control method to pre-adjust audio output parameters before the user experiences micro-awakening: when the sleep drive level is high and the circadian rhythm phase is close to the ideal sleep-onset phase, indicating the user is in a state of natural drowsiness due to insufficient sleep, the system automatically reduces intervention intensity or shuts off audio output to protect the natural sleep-onset process; when the sleep drive level is low or the circadian rhythm phase deviates from the ideal sleep-onset phase, indicating the user is in a state of difficulty falling asleep despite sufficient sleep, the system actively outputs or maintains sleep-aiding audio to compensate for insufficient sleep drive. Differentiated audio regulation strategies are implemented for the five sleep states: audio output is completely shut off during deep sleep and protective monitoring is performed using conditional probability; during light sleep, intervention intensity is adaptively adjusted according to the vulnerability evolution trend and equipped with overdrive control and a smooth exit mechanism; during the sleep-onset phase, asynchronous rhythmic sensory stimulation is used to interfere with the synchronization of micro-awakening oscillations; and during the wakefulness and awakening phases, precise adjustments are made based on convergence states. The self-correcting unit compares the predicted V sequence with the actual micro-awakening sequence and uses the backpropagation algorithm to jointly update the dynamic parameters and the weighting coefficients of the prediction model, so that the model parameters are gradually personalized as the usage time increases.

[0054] The effect achieved in this embodiment is: First, by jointly assessing sleep drive levels and circadian rhythm phases, it achieves accurate real-time differentiation between sleep-deprived and sleep-sufficient states. When sleep-deprived, it protects the natural sleep onset process and avoids unnecessary audio stimulation; when sleep-sufficient, it provides proactive sleep aids to compensate for sleep onset difficulties caused by insufficient sleep drive levels. Second, through dual-dimensional coupled feedforward control, it intervenes before micro-awakening occurs, effectively preventing micro-awakening and its cascading spread, reducing the total number of micro-awakenings throughout the night, and improving sleep continuity and deep sleep duration. Third, through differentiated regulation strategies for five sleep states, it completely shuts off audio output during deep sleep and uses conditional probability for protective monitoring to avoid inappropriate interventions that could disrupt deep sleep, establishing an optimal balance between sleep stability protection and potential disturbance suppression; during light sleep, it employs relax control and a smooth exit mechanism to avoid strategy deadlock and oscillation; and during sleep onset, rhythmic sensory stimulation avoids the limitations of infrasound. Fourth, a robust self-correcting loop converges the sleep drive model parameters and vulnerability prediction weights from group statistics to the individual user's optimal values, reducing the number of adjustment nights required to adapt the system to different users and ensuring long-term effectiveness. In summary, this embodiment provides protective strategies when sleep is insufficient and proactive strategies when sleep is sufficient, based on the user's real-time sleep drive level and circadian rhythm phase, achieving precise and differentiated sleep aids throughout the entire sleep cycle, significantly improving sleep quality and user experience.

[0055] Example 2 This embodiment describes an adaptive sleep aid control system based on a pressure-sensing monitoring sleep aid pillow that implements the above method. The system includes a pressure-sensing monitoring module 1, an intelligent control chip 2, an acoustic module 3, a communication module 4, and a mobile application 5.

[0056] Regarding installation and connection: The pressure monitoring module 1 consists of a 128×128 dot matrix thin-film piezoresistive sensor, embedded between the upper layer of sponge and the lower layer of memory foam in the pillow body. It is electrically connected to the analog-to-digital conversion input pin of the intelligent control chip 2 via a flexible flat ribbon cable. The intelligent control chip 2 is an ESP32-S3 microcontroller, whose general-purpose input / output pins are electrically connected to the control terminal of the acoustic module 3 and the serial data input terminal of the communication module 4, respectively. The acoustic module 3 is a bone conduction speaker or loudspeaker, embedded in the sound insulation cavity on the side of the pillow. The communication module 4 is integrated inside the ESP32-S3 microcontroller and supports 2.4GHz band Wi-Fi and Bluetooth 5.0 dual-mode protocols. The mobile application 5 runs on the user's iOS or Android operating system smart device and establishes a bidirectional signal connection with the communication module 4 via 2.4GHz band Wi-Fi.

[0057] In terms of operational processes and functional implementation: After the system is powered on, the mobile application 5 receives personalized parameters input by the user (such as preferred audio type) and transmits them to the intelligent control chip 2 via the communication module 4. The pressure sensing monitoring module 1 collects pressure distribution data at a rate of 128Hz and transmits it to the intelligent control chip 2. The intelligent control chip 2 integrates a wavelet transform-based noise reduction algorithm and a sleep state recognition algorithm to identify sleep states and form a time series according to the method in Example 1. The intelligent control chip 2 also integrates the following units:

[0058] Dual-process state assessment unit: Assess sleep drive levels and circadian rhythm phases in real time using the method described in step S2 of Example 1, and construct a circadian rhythm sleep tendency function. After smoothing S(t), the convergence rate is calculated using the continuously differentiable synchronization exponent λ(t) = 1 - [S(t) - C(t)]². The sleep drive level S(t) was restricted to the [0,1] interval, and the micro-arousal intensity was... according to

[0059] Normalization.

[0060] Sleep vulnerability prediction unit: Predicts the sleep vulnerability index V for the next 60 minutes using the method in step S3 of Example 1, employing a formula with denominator protection. , where α, β, γ are adjustable weighting coefficients, continuously optimized by self-correcting units, and after calculation, a saturation constraint V=max(0, min(1, V)) is applied to ensure that V is stable in the [0,1] interval.

[0061] Audio Strategy Scheduling Unit: Following the method in step S4 of Embodiment 1, the audio output parameters of acoustic module 3 are adjusted using a two-dimensional coupled feedforward control approach based on sleep state and vulnerability index. This audio strategy scheduling unit incorporates an adaptive fuzzy inference system, whose bell-shaped membership function parameters and fuzzy rule consequent parameters are iteratively updated by the self-correcting unit based on nightly prediction errors using a backpropagation algorithm. This audio strategy scheduling unit also executes regulation according to the five sleep state differentiation strategies in step S4 of Embodiment 1, including convergence rate modulation in the waking state, asynchronous rhythmic acoustic stimulation interference and smooth fade-out in the sleep state, and intervention aggression coefficient decision in the light sleep state (using a ratio with a lower limit threshold Q=max(actual ratio, 0.05)). It is equipped with an overdrive control mechanism with V>0.9, which recalculates R and smoothly switches strategies when V falls back, as well as the minimum effective volume closed-loop calibration in deep sleep (where ΔT in the conditional probability formula is 1h) and the optimal sleep window calculation in wakefulness.

[0062] Self-correcting unit: Jointly updates dynamic parameters according to the method in step S5 of Example 1. , The prediction model is weighted by coefficients α, β, and γ, and a dimensionless learning rate η = 0.01 is used. After the update, a lower bound constraint is applied. , The updated weighted coefficients are normalized to ensure that their sum is always 1.

[0063] In addition, the intelligent control chip 2 also integrates a sleep state confidence assessment unit. This unit calculates the confidence value (range 0-100%) of the current sleep state result based on the signal-to-noise ratio (SNR) of the data collected by the pressure-sensing monitoring module 1 (measured using an AD8232 analog front-end chip) and body acceleration data (measured using a BMI160 accelerometer). When the user violently tosses and turns, causing the SNR to drop below 20dB and the confidence level to fall below the preset 70% threshold, the audio strategy scheduling unit suspends feedforward control and switches to a conservative strategy that maintains the current audio output parameters unchanged until the SNR recovers and the confidence level rises above 70%.

[0064] Simultaneously, the dual-process state assessment unit also extracts cardiac impact recording signals from the 128Hz pressure-sensitive signal, obtains beat-by-beat interval sequences through a heartbeat detection algorithm, and then calculates heart rate variability; respiratory rate variability is extracted from the same signal using bandpass filtering from 0.1Hz to 0.4Hz. The balance index of the autonomic nervous system is calculated based on heart rate variability. (The ratio of low-frequency power (0.04-0.15Hz) to high-frequency power (0.15-0.4Hz). When this ratio deviates from the preset individualized baseline range (e.g., above the baseline mean + 1 standard deviation), it indicates that the autonomic nervous system balance is biased towards sympathetic dominance. In this case, a positive bias correction is applied to the predicted sleep vulnerability index, with a correction amount of [missing value]. Furthermore, the modified V is restricted to the [0,1] range, thereby enhancing the intervention intensity of audio regulation in advance.

[0065] The technical problems solved in this embodiment are as follows: First, the signal quality collected by the pressure-sensitive monitoring module 1 is easily affected by user movement. When the user tosses and turns violently, the signal-to-noise ratio drops sharply, leading to a decrease in the reliability of sleep state recognition and audio feedforward control. The system may then execute inappropriate audio adjustments based on incorrect judgments. Second, the user's autonomic nervous system state is not static. Under high stress or anxiety, the sympathetic nervous system excitability increases, and sleep vulnerability increases. However, existing systems lack the ability to perceive and adaptively respond to this physiological state in real time, resulting in a deviation between the predicted results and the actual sleep vulnerability, and a mismatch between the intervention intensity and the actual needs.

[0066] The working principle of this embodiment is as follows: In addition to integrating a dual-process state assessment unit, a sleep vulnerability prediction unit, an audio strategy scheduling unit, and a self-correction unit to execute the method described in Embodiment 1, the intelligent control chip 2 also includes a sleep state confidence assessment unit and an autonomic nervous system balance index monitoring and correction mechanism. The sleep state confidence assessment unit acquires the signal-to-noise ratio of the data collected by the pressure-sensitive monitoring module 1 and the user's body acceleration data in real time, calculates the confidence value of the current sleep state recognition result, and when the signal-to-noise ratio drops below a preset threshold, the audio strategy scheduling unit suspends the two-dimensional coupled feedforward control and switches to a conservative strategy that keeps the current audio output parameters unchanged until the signal quality recovers and the confidence rises above the threshold. The autonomic nervous system balance index monitoring and correction mechanism extracts the intercardia sequence from the 128Hz high-resolution pressure-sensitive signal and calculates the heart rate variability, and combines it with the respiratory rate variability to obtain the sympathetic-parasympathetic balance index. When the index exceeds the threshold, a positive bias correction is applied to the predicted sleep vulnerability index, thereby enhancing the intervention intensity of audio regulation in advance under high-pressure conditions.

[0067] The effects achieved in this embodiment are as follows: First, by using a sleep state confidence assessment unit to monitor signal-to-noise ratio and body motion acceleration data in real time, the system automatically switches to a conservative strategy when signal quality deteriorates, preventing inappropriate audio control based on erroneous state judgments and improving the system's robustness and safety in complex usage environments. Second, by linking the real-time extraction of the autonomic nervous system balance index with sleep vulnerability prediction correction, the system automatically enhances intervention intensity in a state of high sympathetic nervous system excitation, improving the accuracy and effectiveness of differentiated sleep aid strategies. Third, the synergistic effect of the above two mechanisms enables the system to accurately execute differentiated sleep aid strategies when signal quality is good, ensure operational safety when signal quality fluctuates, and adaptively adjust intervention intensity when the user's autonomic nervous system state changes, comprehensively ensuring the reliable, safe, and effective execution of precise differentiated sleep aid strategies at the system level. This embodiment improves the intelligence level, operational reliability, and environmental adaptability of the sleep aid control system, providing users with a safer and more accurate sleep aid experience.

Claims

1. An adaptive sleep-aid control method for a pressure-sensing sleep-aid pillow, characterized in that, Includes the following steps: S1. Real-time acquisition of user's head pressure distribution, respiratory fluctuations, and body movement frequency data; after noise reduction processing, the user's current sleep state is identified and recorded as a time series; the sleep state includes wakefulness, falling asleep, light sleep, deep sleep, and wakefulness; S2. Based on the time series from step S1, the user's current sleep drive level is assessed in real time according to the sleep drive level accumulation-dissipation dynamics model. At the same time, the user's circadian rhythm phase is assessed based on the accumulated multi-night time series, and a circadian rhythm sleep tendency function with values ​​in the interval [0,1] is constructed that maps to the circadian rhythm phase. The synchronicity change rate between the sleep drive level and the circadian rhythm sleep tendency is calculated as the convergence rate. The sleep drive level accumulation-dissipation dynamics model includes an accumulation rate for controlling the speed of sleep drive accumulation and a dissipation rate for controlling the speed of sleep drive dissipation. S3. Based on the current sleep drive level, circadian rhythm phase and the convergence rate, predict the user's sleep vulnerability index within a future preset time window. The sleep vulnerability index, after saturation limitation, has a value range of [0,1]. S4. Based on the predicted sleep vulnerability index and the current sleep state, a two-dimensional coupled feedforward control method is adopted to adjust the audio output parameters of the acoustic module (3) in advance before the user experiences micro-awakening. The two-dimensional coupling determines the strategy domain for audio modulation based on sleep state and determines the timing and intensity of audio modulation based on sleep vulnerability index, with joint decision-making between the two dimensions; the two-dimensional coupling feedforward control is implemented through an adaptive fuzzy inference system. S5. Compare the predicted sleep vulnerability index sequence for the night with the actual micro-awakening sequence for the night, calculate the prediction error, and update the accumulation rate and dissipation rate in step S2, as well as the model parameters used in step S3 to predict the sleep vulnerability index based on the prediction error.

2. The method according to claim 1, characterized in that: The two-dimensional coupled feedforward control described in step S4 executes differentiated audio modulation strategies for each of the five sleep states: When the user is awake: the frequency reduction slope and volume decay rate of the audio output parameters are modulated at the convergence rate of step S2; when the convergence rate increases, the adjustment is accelerated synchronously; when the convergence rate decreases, the adjustment is paused and the duration of the current audio output parameters is extended, and the convergence rate is continuously monitored; when the convergence rate returns to the normal range, the adjustment process is resumed. When the user is asleep: calculate the time gradient of the sleep vulnerability index; if a periodic oscillation pattern is detected, generate a rhythmic acoustic stimulus to interfere with the oscillation pattern and superimpose it on the current audio output; remove the rhythmic acoustic stimulus when the oscillation amplitude decays to below a preset recovery threshold. When the user is in a light sleep state: predict the vulnerability evolution trajectory within a future time window and calculate the sleep architecture quality index for the current night. Use the ratio of the predicted mean of the vulnerability evolution trajectory to the sleep architecture quality index for the current night as a coefficient for the degree of intervention aggression. When the coefficient is higher than a preset aggressive threshold, lock the current audio output parameters. When the coefficient is lower than a preset mild threshold, gradually reduce the audio output until it is turned off. When the coefficient is in the transition range, maintain the current audio output parameters and shorten the prediction window. The sleep architecture quality index for the current night is the ratio of deep sleep duration to total sleep duration and has a preset lower limit threshold. When the sleep vulnerability index exceeds a preset emergency threshold, prioritize the execution of enhanced intervention strategies, regardless of the intervention aggression coefficient. When the sleep vulnerability index falls back below the preset emergency threshold, smoothly transition to the corresponding strategy based on the recalculated intervention aggression coefficient. When the user is in a deep sleep state: turn off the audio output; continuously monitor the sleep vulnerability index and calculate the conditional probability of micro-awakening based on the sleep state transition frequency; when the conditional probability exceeds the preset probability threshold, restart the audio output at the lowest effective volume. When the user is awake: assess the physiological conditions for falling asleep again, calculate the optimal time window for falling asleep when the physiological conditions are met, and start the sleep guidance audio in advance before the window arrives; The audio control strategies for the five sleep states share a self-correcting loop: the prediction error in step S5 simultaneously updates the dynamic parameters in step S2 and the parameters of the adaptive fuzzy inference system in step S4.

3. The method according to claim 2, characterized in that: The decision basis for the two-dimensional coupled feedforward control described in step S4 is a two-dimensional decision matrix. The row dimension corresponds to five sleep states, and the column dimension corresponds to the classification interval of the sleep vulnerability index. Each element in the matrix defines the target value, control direction, and constraint conditions of the audio output parameters under the corresponding combination.

4. The method according to claim 2, characterized in that: The minimum effective volume used to restart the audio output in the deep sleep state in step S4 is determined based on the individualized calibration of the acoustic module (3); iteratively adjusted according to the feedback results of whether micro-awakening actually occurs after each audio restart, so that the minimum effective volume converges to the individualized optimal value that just maintains the auditory masking effect and does not trigger awakening.

5. The method according to claim 1, characterized in that: The initial values ​​of the accumulation rate and dissipation rate mentioned in step S2 are set based on population statistics and are gradually personalized during multi-night use through step S5.

6. The method according to claim 1, characterized in that: The duration of the preset time window mentioned in step S3 is 30 minutes to 120 minutes.

7. An adaptive sleep aid control system based on a pressure-sensing monitoring sleep aid pillow, characterized in that, include: The pressure monitoring module (1) is used to collect data on the user's head pressure distribution, breathing fluctuations and body movement frequency in real time; The intelligent control chip (2) is electrically connected to the pressure-sensitive monitoring module (1), and has a built-in noise reduction algorithm and sleep state recognition algorithm to identify the user's current sleep state and record it as a time series; the intelligent control chip (2) also integrates: The dual-process state assessment unit is used to assess the user's sleep drive level and circadian rhythm phase in real time based on the time series, construct a circadian rhythm sleep tendency function that maps to the circadian rhythm phase, and calculate the rate of synchronicity change between the sleep drive level and the circadian rhythm sleep tendency as the convergence rate. A sleep vulnerability prediction unit is used to predict a sleep vulnerability index within a future time window based on the sleep drive level, circadian rhythm phase, and convergence rate. The sleep vulnerability index, after being saturated and limited, has a value range of [0,1]. The audio strategy scheduling unit is used to adjust the audio output parameters of the acoustic module (3) in advance in a two-dimensional coupled feedforward control manner based on the current sleep state and the predicted sleep vulnerability index, before the user has experienced micro-awakening. A self-correcting unit is used to update the dynamic parameters in the dual-process state assessment unit and the model parameters in the sleep vulnerability prediction unit based on the prediction error; The acoustic module (3) is electrically connected to the intelligent control chip (2) and is used to play sleep-aid audio according to the control instructions of the audio strategy scheduling unit; The communication module (4) is electrically connected to the intelligent control chip (2) and is used to realize data interaction with the mobile application (5); The mobile application (5) communicates with the smart control chip (2) through the communication module (4) to receive personalized parameters set by the user and display sleep quality reports.

8. The system according to claim 7, characterized in that: The intelligent control chip (2) also integrates a sleep state confidence assessment unit, which is used to calculate the confidence of the current sleep state result based on the signal-to-noise ratio, body movement amplitude and frequency of the data collected by the pressure monitoring module (1); when the confidence is lower than the preset confidence threshold, the audio strategy scheduling unit suspends the execution of feedforward control and switches to a conservative strategy that keeps the current audio output parameters unchanged until the confidence recovers to above the preset confidence threshold.

9. The system according to claim 7, characterized in that: The audio strategy scheduling unit has a built-in adaptive fuzzy inference system, which takes the current sleep state encoding value and sleep vulnerability index as dual inputs and the volume adjustment step size and frequency offset of the audio output as outputs. The membership function parameters and fuzzy rule consequent parameters of the adaptive fuzzy inference system are iteratively updated by the self-correction unit based on the nightly prediction error using the backpropagation algorithm.

10. The system according to claim 7, characterized in that: The dual-process state assessment unit is also used to monitor the respiratory rate variability and heart rate variability extracted from the pressure-sensitive signal in real time, and to calculate the balance index of the autonomic nervous system. When the balance index shows that the sympathetic nerve excitation exceeds the preset sympathetic threshold, the sleep vulnerability prediction unit applies a positive bias correction that is positively correlated with the excessive sympathetic nerve excitation to the currently predicted sleep vulnerability index, so as to enhance the intervention of audio modulation in advance.

Images