Intelligent bed low-disturbance posture adjustment method, system and device based on sleep cycle perception and snoring sound and breathing state recognition

Abstract

An intelligent bed low-disturbance posture adjustment method, system and device based on sleep cycle perception and snoring sound and breathing state recognition. The method comprises: collecting electroencephalogram signals to identify sleep stages, stability or wakefulness risk; collecting multi-source signals such as snoring sound, breathing rhythm, body movement and pressure to judge the breathing risk level; combining sleep state, breathing risk and body position to determine the intervention suitability; generating a low-disturbance adjustment strategy to gradually implement the minimum candidate action, evaluate the respiratory improvement and sleep disturbance in the observation window, and decide to keep, advance or back off; comparing the baseline window and the observation window after adjustment, combining the attribution rule to judge the intervention effect; updating the strategy according to the feedback: keeping if effective, advancing if ineffective, back off if causing wakefulness, and updating the individualized parameters. The invention takes sleep cycle perception as the constraint of bed body adjustment, realizes low-disturbance and closed-loop optimized posture adjustment, significantly reduces sleep disturbance, and improves respiratory patency and sleep quality.

Description

Technical Field

[0001] This invention relates to sleep health monitoring, smart home and smart bed control technology, and in particular to a method, system and device for low-disturbance posture adjustment of a smart bed. Background Technology

[0002] Existing smart beds, smart mattresses, smart pillows, or anti-snoring devices typically detect whether a user is snoring through microphones, pressure sensors, body motion sensors, heart rate sensors, or bed vibration sensors. Upon detecting snoring, they raise the headboard, adjust the pillow height, change localized mattress support, or use vibration to encourage the user to turn over. These solutions can alleviate snoring to some extent, but their control logic usually revolves around the single trigger condition of "whether or not snoring occurs."

[0003] During actual sleep, users exhibit varying sensitivities to bed movements, pillow changes, mattress support variations, or external cues depending on their sleep stage. Tolerance to external intervention differs between light sleep, deep sleep, and REM sleep stages. If a smart bed adjusts its height based solely on snoring intensity, sudden movement during deep or stable sleep stages could cause the user to wake up abruptly, experience mild arousal, increase body movement, or fragmented sleep.

[0004] Therefore, smart bed systems designed for sleep assistance need to simultaneously address the risks of snoring or breathing difficulties, protect sleep continuity, and ensure comfort during bed movements. Existing technologies typically lack real-time identification of sleep stages, or even if sleep monitoring is possible, the results are mostly used for sleep reports and data display, rarely as control constraints for bed posture intervention.

[0005] Furthermore, existing bed adjustment strategies are often quite fixed, such as raising the headboard to a fixed angle after reaching the snoring threshold, lacking a comprehensive assessment of the current body position, respiratory risk level, user's historical response, and intervention sensitivity. The effectiveness and comfort of headboard elevation, back support, pillow height, side tilt assist, or zoned mattress support may vary for different users. Fixed control strategies are difficult to adapt to individual differences and are also difficult to continuously optimize over long-term use.

[0006] Existing snoring detection solutions are susceptible to environmental noise, the voice of the bed partner, the user's movement, bed vibration, or sensor contact status. Relying solely on a single sound signal may lead to misjudgments or missed detections. Relying solely on pressure or body movement signals may fail to accurately distinguish between ordinary body movements, changes in respiratory rhythm, and snoring events. Therefore, there is an urgent need for a smart bed technology solution that integrates sleep stages, snoring or respiratory status, body position, and bed posture intervention into a closed-loop control logic.

[0007] It should be noted that the information disclosed in the background section above is only for understanding the background of this application, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0008] The main objective of this invention is to overcome the deficiencies in the aforementioned background technology and provide a method, system, and device for low-disturbance posture adjustment of an intelligent bed based on sleep cycle perception and snoring and breathing state recognition.

[0009] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect of the present invention, a method for low-disturbance posture adjustment of an intelligent bed based on sleep cycle perception and snoring and breathing state recognition includes the following steps: S1. Collect the EEG signals during the user's sleep process, and perform signal quality evaluation, preprocessing, artifact recognition and feature extraction on the EEG signals. Based on the extracted features, identify the user's current sleep stage, sleep stability or awakening risk, and generate sleep stage confidence. S2. Collect respiratory-related signals, body position-related signals, and body movement-related signals during the user's sleep process; the respiratory-related signals include at least one of snoring signals, respiratory rhythm signals, bed pressure signals, and bed micro-vibration signals; S3. Based on the snoring signal, respiratory rhythm signal, body position, body movement, and optional blood oxygen signal, determine the user's current respiratory risk level through multi-source signal fusion; S4. Determine the appropriateness of intervention based on the sleep state, respiratory risk level, body position, signal reliability, and body movement status; the appropriateness of intervention is used to characterize whether it is suitable to perform bed posture adjustment in the current state and the allowable adjustment intensity boundary; S5. Based on the respiratory risk level, intervention suitability, body position, and historical intervention effects, generate a low-disturbance posture adjustment strategy; the low-disturbance posture adjustment strategy includes the adjustment object, single adjustment amplitude, adjustment speed, action interval, observation window, cumulative adjustment limit, and retreat condition; S6. Control at least one posture adjustment unit of the smart bed to execute the low-disturbance posture adjustment strategy and adopt a progressive execution method: first execute the minimum candidate action that meets the current intervention suitability, evaluate breathing improvement and sleep disturbance within the observation window, and decide to maintain, advance or regress based on the evaluation results; S7. After the adjustment is completed, based on the baseline window before adjustment and the observation window after adjustment, continue to collect EEG signals, snoring signals, respiratory rhythm signals, body position signals, body movement signals and optional blood oxygen signals. Combine the signal reliability and body movement exclusion rules to determine whether the adjustment improved the respiratory state, whether it caused sleep disturbances, and whether the change can be attributed to the posture adjustment. S8. Update subsequent adjustment strategies based on feedback evaluation results: If the adjustment is effective and does not cause sleep disturbance, maintain the current posture or record the action as an effective response; if the adjustment is ineffective and does not cause sleep disturbance, proceed to the next level of low-disturbance adjustment within the allowable range of intervention suitability; if the adjustment causes an increased risk of arousal or increased body movement, reduce the amplitude of subsequent adjustments, reduce the speed of adjustments, suspend continuous adjustments, or implement a regression; and update individualized control parameters based on historical response data.

[0010] In a second aspect of the invention, an intelligent bed low-disturbance posture adjustment system based on sleep cycle sensing and snoring and breathing state recognition includes: The sleep stage recognition module is used to collect or receive the user's electroencephalogram (EEG) signals during sleep, and process the EEG signals to output sleep state, sleep stability or awakening risk, and sleep stage confidence. The snoring and breathing event sensing module is used to collect snoring signals, breathing rhythm signals, bed pressure signals, bed micro-vibration signals or optional blood oxygen signals, and output respiratory risk characteristics. The body position and pressure recognition module is used to determine the user's body position and the pressure distribution in different areas of the body; The sleep-breathing fusion judgment module is connected to the sleep stage recognition module, the snoring and breathing event perception module, and the body position and pressure recognition module, respectively, and is used to determine the respiratory risk level and intervention appropriateness based on sleep state, respiratory risk characteristics, body position, signal reliability and body movement state. The posture adjustment decision module, connected to the sleep-breathing fusion judgment module, is used to generate a low-disturbance posture adjustment strategy based on respiratory risk level, intervention suitability, body position and historical intervention effects. The strategy includes adjustment object, single adjustment amplitude, adjustment speed, action interval, observation window, cumulative adjustment limit and backtracking condition. The bed execution module, connected to the attitude adjustment decision module, includes at least one attitude adjustment unit for executing the low-disturbance attitude adjustment strategy and adopting a progressive execution method. The feedback evaluation module is connected to the sleep stage recognition module, the snoring and breathing event perception module, the body position and pressure recognition module and the bed execution module, respectively. It is used to evaluate the improvement of breathing status, sleep disturbance and change attribution based on the baseline window before adjustment and the observation window after adjustment after adjustment, and generate feedback confidence. The individualized learning module, connected to the feedback evaluation module and the attitude adjustment decision module, is used to update the individualized control parameters based on historical response data.

[0011] In a third aspect of the invention, a smart bed low-disturbance posture adjustment device based on sleep cycle sensing and snoring and breathing state recognition includes: Bed frame or mattress body; At least one posture adjustment unit is disposed on the bed frame or mattress body for adjusting at least one area of ​​the headboard, back, waist, legs, pillow, left and right tilt mechanism or mattress zone support area; The breathing state sensing unit is used to collect snoring signals, breathing rhythm signals, bed pressure signals, or bed micro-vibration signals. EEG signal acquisition or receiving unit, used to acquire EEG signals during the user's sleep process; The controller is connected to the posture adjustment unit, the breathing state sensing unit, and the EEG signal acquisition or receiving unit, respectively, and is configured to perform the following operations: The user's current sleep stage, sleep stability, or risk of awakening is identified based on the electroencephalogram (EEG) signals. The respiratory risk level is identified based on the signals collected by the respiratory state sensing unit. The appropriateness of intervention is determined based on sleep status, respiratory risk level, body position, signal reliability, and body movement status. A low-disturbance posture adjustment strategy is generated based on respiratory risk level, intervention suitability, body position, and historical intervention effects. The strategy includes adjustment target, single adjustment amplitude, adjustment speed, observation window, and retreat conditions. The attitude adjustment unit is controlled to execute the strategy in a progressive manner, and after adjustment, signals are collected for feedback evaluation based on the baseline window before adjustment and the observation window after adjustment. The subsequent adjustment strategy and individualized control parameters will be updated based on the feedback and evaluation results.

[0012] The present invention has the following beneficial effects: This invention proposes a method, system, and device for low-disturbance posture adjustment of an intelligent bed based on sleep cycle perception and snoring and respiratory status recognition. Its core lies in directly incorporating EEG sleep stage, sleep stability, or awakening risk into the control logic of bed posture adjustment, serving as a dual constraint on the timing and intensity of adjustment. Compared to existing technologies that trigger bed movements based on a single snoring threshold or fixed vital sign threshold, this invention, after identifying abnormal snoring or respiratory status, further comprehensively judges the appropriateness of intervention based on the user's current sleep stage, respiratory risk level, body position, and historical intervention effects. Based on this, it dynamically determines whether to adjust, when to adjust, the adjustment amplitude, the adjustment speed, the observation window, and whether to revert. Simultaneously, after bed adjustment, this invention continues to collect EEG, snoring, respiratory rhythm, body movement, and optional blood oxygenation signals to determine whether the adjustment improves respiratory status and whether it induces a tendency to awaken or suspected micro-awakening. Based on the feedback results, it updates subsequent adjustment strategies, thus forming a complete closed-loop control scheme of "sleep state perception—respiratory risk assessment—intervention appropriateness determination—low-disturbance posture adjustment—feedback evaluation—individualized strategy update."

[0013] Based on this, the present invention enables the smart bed to select a more appropriate intervention time through EEG sleep stage recognition, avoiding unnecessary awakenings during deep sleep or unstable sleep stages; by integrating multi-source signals such as snoring, respiratory rhythm, body pressure, body movement, and optional blood oxygen, it significantly improves the reliability of snoring and suspected breathing difficulties risk assessment, and can make decisions based on rules such as credibility priority, multi-source consistency, and risk conservatism when signals conflict; through a progressive low-disturbance adjustment strategy and multi-regional collaborative adjustment capability, that is, first executing the minimum candidate action that meets the appropriateness of intervention, evaluating breathing improvement and sleep disturbance within the observation window, and then deciding to maintain, advance, or regress based on the evaluation results, it effectively reduces the interference of bed movements on sleep continuity, and protects the user's sleep structure to the maximum extent while alleviating the risk of snoring or breathing difficulties. In addition, this invention introduces a closed-loop learning mechanism for individual users, which can record the relationship between different sleep stages, different body positions, different bed postures and snoring and breathing states, gradually obtain the user's minimum effective adjustment range, easy awakening adjustment mode and preferred intervention strategy, and continuously update individualized control parameters through methods such as moving average, linear regression, Bayesian optimization or reinforcement learning, thereby continuously improving the effectiveness and comfort of intervention in long-term use.

[0014] In summary, this invention integrates sleep stages, respiratory risk, and low-disturbance posture adjustment into a closed-loop control system. It can dynamically adjust intervention strategies based on the user's real-time sleep status and respiratory risk in scenarios such as home sleep, health and wellness spaces, nursing spaces, and smart bedrooms. This helps reduce awakenings and sleep fragmentation caused by bed movements and improves breathing patency and sleep continuity under low-disturbance conditions.

[0015] Other beneficial effects of the embodiments of the present invention will be further described below. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of a smart bed low-disturbance posture adjustment scheme based on sleep cycle perception and snoring and breathing state recognition, according to an embodiment of the present invention.

[0017] Figure 2 This is a block diagram of the intelligent bed low-disturbance posture adjustment system based on sleep cycle perception and snoring and breathing state recognition, according to an embodiment of the present invention.

[0018] Figure 3 This is a schematic diagram of the sleep-breathing fusion judgment logic in an embodiment of the present invention.

[0019] Figure 4A This is a schematic diagram of the layout of the multi-zone posture adjustment actuator for the smart bed according to an embodiment of the present invention.

[0020] Figure 4B This is a schematic diagram of the left and right tilt adjustment mechanism of the smart bed according to an embodiment of the present invention.

[0021] Figure 4C This is a schematic diagram illustrating an example of the range of intelligent bed posture adjustment parameters according to an embodiment of the present invention.

[0022] Figure 5 This is a flowchart of a smart bed low-disturbance posture adjustment method based on sleep cycle perception and snoring and breathing state recognition, according to an embodiment of the present invention. Detailed Implementation

[0023] The embodiments of the present invention will be described in detail below. It should be emphasized that the following description is merely exemplary and not intended to limit the scope and application of the present invention.

[0024] This invention aims to overcome the sleep disturbance problem caused by existing smart beds that adjust their posture based solely on a fixed amplitude of snoring threshold. It proposes a low-disturbance posture adjustment scheme for smart beds that uses EEG sleep stages, sleep stability, and awakening risk as timing and intensity constraints for posture adjustment. This scheme integrates multi-source signals such as snoring, respiratory rhythm, and body position for respiratory risk grading and intervention suitability assessment. Through a progressive and reversible low-disturbance adjustment strategy, it minimizes body movement and awakening while improving the risk of snoring or breathing difficulties. Combined with post-adjustment feedback evaluation and individualized learning mechanisms, it continuously optimizes the adjustment amplitude, speed, and intervention timing, forming a closed-loop control that balances respiratory improvement and sleep continuity protection, significantly improving overall sleep quality throughout the night.

[0025] See Figures 1 to 5This invention provides a method for low-disturbance posture adjustment of a smart bed based on sleep cycle perception and snoring and breathing state recognition. The method includes the following steps: Step S1. Collect the EEG signals during the user's sleep process, and perform signal quality evaluation, preprocessing, artifact recognition and feature extraction on the EEG signals. Based on the extracted features, identify the user's current sleep stage, sleep stability or awakening risk, and generate sleep stage confidence.

[0026] In some embodiments, step S1, identifying the user's current sleep stage, sleep stability, or awakening risk, includes: performing EEG processing using a layered time window, wherein sleep stage identification uses a basic staging window of a first preset time length and is updated with a first sliding step size, sleep stability and awakening risk assessment use a sliding window of a second preset time length, and suspected awakening, artifacts, and sudden body movement risk identification use a short window of a third preset time length; when the confidence level of the sleep stage is higher than a preset threshold, the fine sleep stage results are used as control constraints; when the confidence level of the sleep stage is lower than a preset threshold or the EEG signal quality score is lower than a preset threshold, the sleep stage is downgraded to a coarse-grained state such as stable sleep, unstable sleep, or suspected awakening, and the posture adjustment level, single adjustment amplitude, adjustment speed, observation window, or automatic posture adjustment is temporarily suspended accordingly.

[0027] Step S2. Collect respiratory-related signals, body position-related signals, and body movement-related signals during the user's sleep process; the respiratory-related signals include at least one of snoring signals, respiratory rhythm signals, bed pressure signals, and bed micro-vibration signals.

[0028] In some embodiments, step S2, the acquisition of respiratory-related signals during the user's sleep process includes: estimating background noise, denoising, framing, and segmenting sound events in the snoring signal, and determining candidate snoring events based on short-time energy, zero-crossing rate, duration, dominant frequency range, frequency band energy, Mel-Frequency Cepstral Coefficients (MFCC), or snoring recognition probability; the optional blood oxygen signal determines the blood oxygen decline trend based on the decrease in blood oxygen saturation relative to the individual baseline, the duration of the decrease, the slope of the decrease, the minimum value, and the recovery speed.

[0029] Step S3. Based on the snoring signal, respiratory rhythm signal, body position, body movement, and optional blood oxygen signal, determine the user's current respiratory risk level through multi-source signal fusion. For example... Figure 3 As shown in the schematic diagram of the sleep-breathing fusion judgment logic, and in conjunction with the exemplary features and rules in Tables 1 to 3 (see below for details), this step is specifically implemented as follows.

[0030] In some embodiments, step S3, determining the user's current respiratory risk level through multi-source signal fusion, includes: preprocessing and feature extraction of snoring signals, respiratory rhythm signals, postural pressure signals, body movement signals, and optional blood oxygen signals, respectively. The extracted features include one or more of the following: snoring intensity, snoring duration, event frequency, dominant frequency range, frequency band energy, MFCC features, respiratory cycle, respiratory interval statistical variance, respiratory cycle coefficient of variation, respiratory amplitude change, suspected pause-like interval, pulse oximetry (SO2) decrease magnitude, decrease duration, decrease slope, minimum value, and recovery speed; normalizing each feature, outputting a comprehensive respiratory risk score using a weighted scoring or machine learning model, and classifying the respiratory risk level into low risk, medium risk, high risk, and warning risk based on a preset threshold; and adjudicating signal conflicts according to the principles of credibility priority, multi-source consistency, risk conservatism, body movement exclusion, and missing and conflict-related downgrading.

[0031] Step S4. Determine the intervention suitability based on the sleep state, respiratory risk level, body position, signal reliability, and body movement status. The intervention suitability is used to characterize whether bed posture adjustment is suitable in the current state and the allowable adjustment intensity boundary. As shown in Tables 2, 4 to 6 (see below), different sleep stages correspond to different basic intervention suitability mapping relationships with respiratory risk levels, and can be corrected according to body position, reliability, and body movement.

[0032] In some embodiments, step S4, determining the appropriateness of intervention includes: first, forming a basic intervention appropriateness mapping relationship based on sleep state and respiratory risk level, and then correcting it based on body position, signal confidence, and body movement status; wherein, the appropriateness of intervention includes at least six levels: suspending automatic intervention, no motion monitoring, allowing only fine-tuning, suitable for low-disturbance adjustment, cautious phased adjustment, and gentle reminder or terminal prompt; when the confidence level of EEG staging is low, the confidence level of snoring recognition is low, the body position recognition is uncertain, or there is conflict between multiple sources of signals, the intervention level is reduced or the observation window is extended; when the user moves frequently or is suspected of turning over, automatic posture adjustment is suspended; when the body movement increases or the risk of awakening increases after adjustment, the subsequent adjustment amplitude is reduced, the adjustment speed is reduced, or a rollback is performed.

[0033] Step S5. Based on the respiratory risk level, intervention suitability, body position, and historical intervention effects, generate a low-disturbance posture adjustment strategy; the low-disturbance posture adjustment strategy includes the adjustment object, single adjustment amplitude, adjustment speed, action interval, observation window, cumulative adjustment upper limit, and retreat conditions. Figure 4AThe smart bed is shown to have multiple adjustable zones. Tables 7 and 8 (see below for details) show the correspondence between body position and priority adjustment objects, as well as the exemplary value range of low-disturbance adjustment parameters.

[0034] In some embodiments, step S5, generating a low-disturbance posture adjustment strategy further includes: selecting candidate adjustment objects based on body position; when the user is in a supine position, priority is given to slightly raising the head of the bed, or fine-tuning the head and neck or occiput support; when the user is in a side-lying position, priority is given to fine-tuning the head and neck, occiput, or shoulder and back support. The length of the observation window is dynamically adjusted based on the adjustment object, respiratory risk level, sleep state, and historical feedback results; a shorter observation window is used for occiput or head and neck fine-tuning, and a longer observation window is used for combined back adjustment, side-tilt assistance, or mattress zone support adjustment. The method further includes establishing an adjustment action-awakening risk prediction mechanism; estimating the likelihood of candidate adjustment actions causing awakening or increased body movement based on sleep state, body position, historical action type, historical adjustment amplitude, historical adjustment speed, past awakening response, body movement changes, and individual user sensitivity; and prioritizing actions with lower expected disturbance and potential for respiratory improvement among multiple candidate strategies. Furthermore, for the deep sleep stage, the system adopts a small-amplitude, low-speed, and phased adjustment method, that is, it performs the adjustment step by step according to the preset small amplitude and low speed, and sets a longer observation window after each adjustment to assess sleep disturbances.

[0035] Step S6. Control at least one posture adjustment unit of the smart bed to execute the low-disturbance posture adjustment strategy and adopt a progressive execution method: first execute the minimum candidate action that meets the current intervention suitability, evaluate breathing improvement and sleep disturbance within the observation window, and decide whether to maintain, advance or regress based on the evaluation results.

[0036] In some embodiments, in step S6, the low-disturbance posture adjustment strategy implemented by controlling at least one posture adjustment unit of the smart bed includes: controlling at least one of the following to perform adjustment: headboard lifting mechanism, back support mechanism, lumbar support mechanism, leg support mechanism, pillow adjustment mechanism, mattress zone support mechanism, or lateral tilting assist mechanism; in the progressive execution mode, the single adjustment amplitude, adjustment speed, action duration, and action interval are determined according to the intervention suitability, respiratory risk level, sleep state, and body position, and are executed step by step within the cumulative adjustment limit.

[0037] Step S7. After the adjustment is completed, based on the baseline window before adjustment and the observation window after adjustment, continue to collect EEG signals, snoring signals, respiratory rhythm signals, body position signals, body movement signals, and optional blood oxygenation signals. Combine the signal reliability and body movement exclusion rules to determine whether the adjustment improved the respiratory state, whether it caused sleep disturbances, and whether the change can be attributed to the posture adjustment. Tables 9 and 10 (see below for details) provide examples of feedback evaluation indicators and attribution rules.

[0038] In some embodiments, step S7, determining whether the current adjustment improves the respiratory state, whether it causes sleep disturbances, and whether the change can be attributed to the current posture adjustment, includes: setting a baseline window before adjustment and an observation window after adjustment; comparing indicators such as changes in snoring intensity, snoring duration, respiratory cycle, respiratory interval statistical variance, EEG arousal risk, body movement intensity, and optional blood oxygenation within the two windows; and generating feedback confidence by combining time proximity rules, multi-source consistency rules, disturbance exclusion rules, signal credibility rules, and sleep stage stability rules; when the improvement of respiratory-related indicators occurs within the observation window after adjustment, is adjacent to the time of the adjustment action, is not accompanied by obvious spontaneous turning over or natural transition of sleep stages, and the signal quality meets preset conditions, the change is attributed to the current posture adjustment and marked as high-confidence feedback; otherwise, it is marked as medium-confidence or low-confidence feedback.

[0039] Step S8. Update the subsequent adjustment strategy based on the feedback evaluation results: If the adjustment is effective and does not cause sleep disturbance, maintain the current posture or record the action as an effective response; if the adjustment is ineffective and does not cause sleep disturbance, proceed to the next level of low-disturbance adjustment within the allowable range of intervention suitability; if the adjustment causes an increased risk of arousal or increased body movement, reduce the amplitude of subsequent adjustments, reduce the adjustment speed, suspend continuous adjustment, or perform a regression; and update the individualized control parameters based on historical response data. Table 11 (see below) lists the updatable individualized control parameters and examples of their update rules.

[0040] In some embodiments, step S8, updating the individualized control parameters includes: using each nightly sleep session as the basic update cycle, or periodically updating based on data from several consecutive nights; recording the sleep state, respiratory risk level, body position, signal confidence, body movement state, intervention suitability, adjustment target, adjustment amplitude, adjustment speed, observation window, and changes in each indicator before and after adjustment, as well as feedback confidence, corresponding to each intervention event; updating at least one of the following control parameters using moving average, exponentially weighted average, linear regression, logistic regression, decision tree, Bayesian optimization, or reinforcement learning methods: minimum effective adjustment amplitude, head-of-bed elevation angle threshold, occipital or head-neck adjustment parameters, lateral tilt assistance priority, easily aroused movement type, movement threshold under different sleep stages, and observation window duration; wherein, samples with low-confidence feedback, accompanied by obvious spontaneous body movement, or natural transitions in sleep stages are not included in the learning update or have their weights reduced.

[0041] In some embodiments, the method further includes safety protection and fault degradation steps: monitoring sensor health status, data reliability, actuator operating status, and communication status; when the EEG headband falls off, electrode contact remains abnormal, or the EEG signal quality is below a preset threshold, downgrading to a coarse-grained state such as stable sleep, unstable sleep, or suspected awakening, or suspending active posture adjustment based on fine sleep stages; when pressure or postural signals are abnormal, reducing the role of postural state in the adjustment strategy, and not performing tilting movements that rely on direction judgment; when the actuator detects jamming, overcurrent, overheating, or positional deviation exceeding a preset range, stopping the current action and maintaining the current position, reverting to the nearest safe posture, or restoring to a preset safe posture according to the safety strategy; adopting a graded degradation strategy, including maintaining the current posture and increasing the monitoring frequency, allowing only the smallest amplitude low-speed reversible fine adjustment, suspending all automatic mechanical adjustments, and triggering user terminal prompts or nursing assistance prompts.

[0042] In one specific embodiment of the present invention, a smart bed low-disturbance posture adjustment system based on sleep cycle sensing and snoring and breathing state recognition is also provided. See also... Figure 2 The system includes a sleep stage recognition module, a snoring and breathing event perception module, a body position and pressure recognition module, a sleep-breathing fusion judgment module, a posture adjustment decision module, a bed execution module, a feedback evaluation module, and an individualized learning module.

[0043] The sleep stage identification module collects or receives EEG signals during the user's sleep process and processes these signals to output sleep state, sleep stability or awakening risk, and sleep stage confidence. The snoring and respiratory event sensing module collects snoring signals, respiratory rhythm signals, bed pressure signals, bed micro-vibration signals, or optional blood oxygenation signals and outputs respiratory risk characteristics. The body position and pressure identification module determines the user's body position and pressure distribution in different areas of the body. The sleep-breathing fusion judgment module is connected to the sleep stage identification module, the snoring and respiratory event sensing module, and the body position and pressure identification module, respectively. It determines the respiratory risk level and intervention suitability based on sleep state, respiratory risk characteristics, body position, signal confidence, and body movement. The definitions of its core state variables are shown in Table 2 (see below for details). The classification of intervention suitability levels and basic mapping relationships can be found in Tables 4 to 6. The posture adjustment decision module is connected to the sleep-breathing fusion judgment module and is used to generate a low-disturbance posture adjustment strategy based on respiratory risk level, intervention suitability, body position, and historical intervention effects. The strategy includes the adjustment target, single adjustment amplitude, adjustment speed, action interval, observation window, cumulative adjustment upper limit, and backtracking conditions. The correspondence between the adjustment strategy and intervention suitability, the correspondence between body position and priority adjustment target, and exemplary value ranges of the low-disturbance adjustment parameters can be found in Tables 6 to 8. The bed execution module is connected to the posture adjustment decision module and includes at least one posture adjustment unit for executing the low-disturbance posture adjustment strategy using a progressive execution method. The feedback evaluation module is connected to the sleep stage identification module, snoring and respiratory event perception module, body position and pressure identification module, and bed execution module, respectively. It is used to evaluate the improvement in respiratory status, sleep disturbance, and attribution of changes based on the pre-adjustment baseline window and the post-adjustment observation window after adjustment, and to generate feedback confidence. Its feedback evaluation indicators and attribution rules can be found in Tables 9 and 10. The individualized learning module is connected to the feedback evaluation module and the attitude adjustment decision module, and is used to update the individualized control parameters based on historical response data. Examples of the updatable control parameters and update rules can be found in Table 11.

[0044] In some embodiments, the system further includes a safety protection and fault degradation module. This module is connected to the sleep stage recognition module, the snoring and breathing event sensing module, the body position and pressure recognition module, the sleep-breathing fusion judgment module, the posture adjustment decision module, and the bed execution module, respectively. It is used to monitor the health status of sensors, data reliability, actuator operating status, and communication status, and to perform pause, rollback, degraded operation, or prompt operation when an abnormality is detected.

[0045] In another specific embodiment of the present invention, a smart bed low-disturbance posture adjustment device based on sleep cycle sensing and snoring and breathing state recognition is also provided. The device includes a bed frame or mattress body, at least one posture adjustment unit, a breathing state sensing unit, an electroencephalogram (EEG) signal acquisition or reception unit, and a controller.

[0046] The posture adjustment unit is mounted on the bed frame or mattress body and is used to adjust at least one of the following areas: headboard, backrest, lumbar region, legs, pillow, left and right tilt mechanism, or mattress zone support area. Figures 4A to 4C The demonstration showcased the main adjustable areas, adjustment mechanisms, and adjustment methods of the smart bed. The system utilizes multiple actuators to achieve coordinated control of areas such as the headboard, backrest, lumbar region, legs, pillow, left and right tilt, and mattress zone support / airbags, enabling low-disturbance, personalized sleep posture adjustment. Figure 4AAs shown, the support surface of the smart bed is divided into seven support zones along the longitudinal direction of the user's body, including the pillow support zone 1, shoulder and back support zone 2, lumbar support zone 3, hip and buttock support zone 4, thigh support zone 5, calf support zone 6, and foot support zone 7. Different colors are used to distinguish each support zone in the diagram. The specific boundaries, lengths, and areas of each zone can be adjusted according to mattress size, user body type, ergonomic requirements, and product structure, and are not intended to limit the scope of protection of this invention. The occipital support area 1 is equipped with an occipital support actuator 8, used to adjust the support height, angle, firmness, or local pressure of the corresponding positions of the user's head and neck; the shoulder and back support area 2 is equipped with a shoulder and back support actuator 9, used to adjust the support height, angle, firmness, or local pressure of the corresponding positions of the user's shoulders and upper back, and can also participate in the overall headboard lifting or back support adjustment; the lumbar support area 3 is equipped with a lumbar support actuator 10, used to adjust the support height, support curvature, firmness, or local pressure of the corresponding positions of the lumbar region; the hip and buttock support area 4 is equipped with a hip and buttock support actuator 11, used to adjust the pelvis, buttocks, and hips. The corresponding local height, support stiffness, or pressure distribution at each position is adjusted to improve postural stability and disperse local pressure. Thigh support area 5 is equipped with a thigh support actuator 12 to adjust the support height, angle, stiffness, or local pressure at the corresponding position of the thigh. Lower leg support area 6 is equipped with a lower leg support actuator 13 to adjust the support height, angle, stiffness, or local pressure at the corresponding position of the lower leg, and can work in conjunction with thigh support area 5 to raise or lower the leg. Foot support area 7 is equipped with a foot support actuator 14 to adjust the support height, angle, stiffness, or local pressure at the corresponding positions of the ankle and heel. The occipital support actuator 8, shoulder and back support actuator 9, lumbar support actuator 10, hip and buttock support actuator 11, thigh support actuator 12, lower leg support actuator 13, and foot support actuator 14 can be implemented using electric push rods, linear actuators, screw lifting mechanisms, linkage mechanisms, motor-driven hinges, airbag arrays, partitioned lifting plates, flexible support plates, or combinations thereof. The aforementioned actuators can be set up independently, or one actuator can link two or more adjacent support zones; multiple airbags, lifting units or flexible support units can also be set up in the same support zone. Figure 4A The seven support zones shown represent the spatial division of the mattress's support surface. Figure 4B and Figure 4C The headboard lift, back support, lumbar support, leg lift, pillow height adjustment, left and right tilt, mattress zone firmness adjustment, and sit-up mode shown are all posture adjustment mechanisms. Different posture adjustments can be achieved independently by a single support zone, or through the coordinated operation of multiple support zones and the bed frame mechanism. For example... Figure 4BAs shown, the device also includes a left-right tilt adjustment mechanism. The left-right tilt adjustment can change the relative support height of the left and right sides through the support actuators on the left and right sides of the bed, the zoned airbags, the zoned lifting plates, or the bed frame tilt drive mechanism. The exemplary total mechanical tilt angle range for both left and right tilts is 0° to 10°, used to guide the user from a supine position to a semi-reclining or slightly lateral position in a low-speed, small-amplitude manner when the user snores while lying on their back or has abnormal breathing rhythm. Figure 4C As shown, the exemplary total mechanical adjustment range for each posture adjustment mode can be set as follows: headboard elevation or descent angle range of 0°–70°, back support elevation or descent angle range of 0°–60°, lumbar support elevation or descent height range of 0–80 mm, leg elevation or descent angle range of 0°–45°, pillow height adjustment range of 0–100 mm, and left / right tilt adjustment angle range of 0°–10°. The firmness, local height, or airbag pressure of each support zone can be expressed as a relative level, airbag pressure value, support stiffness value, or a percentage increase or decrease relative to the individual comfort baseline, such as low, medium, high levels, or preset pressure ranges. Furthermore, the device can be configured with a sit-up mode achieved through the coordinated operation of multiple zones such as the shoulder and back support area, lumbar support area, thigh support area, and calf support area, as well as the bed frame mechanism, to achieve one-button sit-up or comfortable standing. The above total mechanical adjustment range is used to indicate the adjustability of each adjustment mechanism and does not represent the completion of all adjustments in one go during sleep intervention. During actual sleep intervention, the system performs interventions in stages within the total mechanical adjustment range, based on sleep status, respiratory risk level, intervention suitability, body position, and historical feedback results, according to preset single adjustment amplitude, adjustment speed, and cumulative adjustment limit.

[0047] The respiratory state sensing unit is used to collect snoring signals, respiratory rhythm signals, bed pressure signals, or bed micro-vibration signals. The EEG signal acquisition or receiving unit is used to acquire EEG signals during the user's sleep process. The controller is connected to the posture adjustment unit, the respiratory state sensing unit, and the EEG signal acquisition or receiving unit, and is configured to perform the following operations: identify the user's current sleep stage, sleep stability, or awakening risk based on the EEG signals; identify the respiratory risk level based on the signals collected by the respiratory state sensing unit; determine the appropriateness of intervention based on sleep state, respiratory risk level, body position, signal reliability, and body movement; generate a low-disturbance posture adjustment strategy based on the respiratory risk level, intervention appropriateness, body position, and historical intervention effects, the strategy including the adjustment object, single adjustment amplitude, adjustment speed, observation window, and rollback conditions, such as the exemplary adjustment parameters listed in Table 8; control the posture adjustment unit to execute the strategy in a progressive manner, and continue to collect signals for feedback evaluation based on the pre-adjustment baseline window and the post-adjustment observation window after adjustment; and update subsequent adjustment strategies and individualized control parameters based on the feedback evaluation results.

[0048] In some embodiments, the posture adjustment unit includes at least one of the following: an electric actuator, a linear actuator, a lead screw lifting mechanism, a linkage mechanism, an airbag array, a partitioned lifting plate, or a flexible support sheet. The EEG signal acquisition or receiving unit is integrated into a flexible fabric headband, a flexible silicone headband, an elastic webbing, a periauricular wearing structure, or a occipital contact structure, and includes flexible dry electrodes, semi-dry electrodes, fabric electrodes, flexible conductive rubber electrodes, gel electrodes, or low-pressure contact electrodes. Furthermore, the device also includes a safety protection unit connected to the controller for monitoring the actuator status and sensor health status, and performing degraded operation in case of abnormalities.

[0049] The intelligent bed low-disturbance posture adjustment scheme proposed in this invention uses EEG sleep stage, sleep stability, and awakening risk as constraints on the timing and intensity of bed posture adjustment. It dynamically determines the respiratory risk level and intervention appropriateness through multi-source signal fusion, and adopts a progressive and reversible low-disturbance adjustment strategy to minimize body movement and awakening while alleviating snoring or breathing difficulties. Furthermore, it combines post-adjustment feedback evaluation and individualized closed-loop learning mechanisms to continuously optimize the adjustment amplitude, speed, and intervention timing, thereby forming an adaptive control that takes into account both respiratory improvement and sleep continuity protection, significantly improving the quality of sleep throughout the night.

[0050] The following further describes the implementation, working principle, and advantages of specific embodiments of the present invention.

[0051] This invention proposes a low-disturbance posture adjustment scheme for a smart bed based on sleep cycle sensing and snoring and breathing state recognition. See the overall schematic diagram below. Figure 1This solution directly incorporates EEG sleep stages, sleep stability, and awakening risk into the control logic of bed posture adjustment, serving as a dual constraint on the timing and intensity of adjustment. Specifically, the system acquires EEG signals during the user's sleep process through an EEG acquisition module to identify the user's current sleep stage, sleep stability, or awakening risk; it identifies snoring, abnormal breathing rhythms, or suspected sleep apnea risk through snoring, mattress pressure, bed micro-vibrations, postural pressure, and optional blood oxygenation signals; and it determines the respiratory risk level and intervention suitability by integrating sleep state, respiratory risk characteristics, postural state, signal reliability, and body movement status through a sleep-breathing fusion judgment module. Based on this, the posture adjustment decision module generates a low-disturbance posture adjustment strategy that includes parameters such as the adjustment target, adjustment amplitude, adjustment speed, adjustment sequence, action interval, observation window, and retraction conditions, based on the respiratory risk level, intervention suitability, postural state, and historical intervention effects. The bed execution module performs progressive and retractable posture adjustments to the headboard, back, waist, legs, pillow, mattress zone support areas, or tilt mechanism. After adjustment, the feedback evaluation module continues to monitor snoring, respiratory rhythm, EEG arousal response, body movement, and optional blood oxygenation changes to determine the intervention effect and update individualized adjustment parameters.

[0052] This invention integrates user sleep state, snoring and breathing state, body position, and signal reliability into the intelligent bed posture adjustment control logic. This means that bed adjustment is no longer triggered solely by a single snoring threshold, but rather dynamically determines whether, when, and at what intensity to intervene based on the user's current sleep state and respiratory risk level. Through this closed-loop control, the system can determine sleep stage, sleep stability, and awakening risk in real time based on EEG signals, and combine multi-source signals to form a comprehensive judgment result, thereby generating a low-disturbance adjustment strategy. This technical concept establishes a complete control chain of "sleep state perception—respiratory risk assessment—intervention appropriateness determination—low-disturbance posture adjustment—feedback evaluation—individualized strategy update," unifying sleep staging, respiratory risk, and posture adjustment into closed-loop control, rather than using sleep staging only for sleep reports or snoring recognition only as a single action trigger condition.

[0053] This invention further enables smart beds to minimize the impact of bed movements on sleep continuity while alleviating the risks of snoring or breathing difficulties. Upon detecting snoring or breathing difficulties, the system first determines the user's current sleep stage and the appropriateness of intervention, then decides whether to adjust, when to adjust, at what amplitude and speed, and whether to use phased adjustments. Furthermore, the invention's closed-loop learning mechanism, designed for individual users, records the relationship between different sleep stages, body positions, bed postures, and snoring and breathing states, gradually obtaining the user's minimum effective adjustment amplitude, easily awakened adjustment methods, and optimal intervention strategies, thereby improving comfort and effectiveness in long-term use.

[0054] The following section provides a detailed description of the overall system structure, functions of each module, and their interrelationships.

[0055] System Overall Structure

[0056] The intelligent bed system of the present invention may include a sleep stage recognition module, a snoring and breathing event sensing module, a body position and pressure recognition module, a sleep-breathing fusion judgment module, a posture adjustment decision module, a bed execution module, a feedback evaluation module, and a personalized learning module. These modules can be integrated into the intelligent bed controller or distributed across a sleep headband, intelligent mattress, bed frame controller, mobile application, edge computing terminal, or cloud platform. In a further embodiment, the system may also include a safety protection and fault degradation module, used to monitor the health status of sensors, data reliability, actuator status, and communication status, and to perform pause, rollback, degraded operation, or prompt in abnormal situations.

[0057] like Figure 2 As shown, the system includes an EEG acquisition module, a snoring and breathing event perception module, a body position pressure recognition module, a sleep-breathing fusion judgment module, a posture adjustment decision module, a bed execution module, a feedback evaluation module, and a personalized learning module. The modules form a closed-loop connection relationship from sleep state perception, breathing risk judgment, posture adjustment decision to feedback evaluation and strategy update.

[0058] Sleep stage recognition module

[0059] The sleep stage recognition module is used to collect users' electroencephalogram (EEG) signals in a low-intrusive manner in a home sleep environment, and outputs the user's current sleep stage, sleep stability, awakening risk, and sleep stage confidence level based on the EEG signals. The output of this module is not simply used for sleep report display, but serves as a control constraint for bed posture adjustment, participating in the determination of intervention trigger thresholds, adjustment timing, adjustment amplitude, adjustment speed, observation window, conditions for suspending intervention, and conditions for regression.

[0060] The sleep stage recognition module may include an EEG acquisition unit, a signal preprocessing unit, an artifact recognition unit, an EEG feature extraction unit, a sleep state recognition unit, and a signal quality evaluation unit. The EEG acquisition unit can be integrated into a flexible fabric headband, a flexible silicone headband, an elastic webbing, a periauricular wearing structure, a occipital contact structure, or other low-pressure structures suitable for overnight wear. The EEG electrodes can be flexible dry electrodes, semi-dry electrodes, fabric electrodes, flexible conductive rubber electrodes, gel electrodes, or low-pressure contact electrodes. If specific commercially available electrodes or self-developed electrodes are used, their model, material, contact area, contact impedance range, and fixing method can be further specified in the embodiments.

[0061] EEG acquisition electrodes can be placed on the user's forehead, frontal side, temporal side, periauricular region, top of the head, or near the mastoid process. Preferably, the acquisition electrodes can be placed at one or more of Fp1, Fp2, Fpz, F7, F8, Cz, or their approximate locations; the reference electrode can be placed behind the ear, near the mastoid process, on the frontal side, or in a stable contact position within the headband; the ground electrode can be placed on the inner side of the headband, forehead, periauricular region, or other stable contact areas. The electrode positions can be arranged with reference to the international 10-20 system, or equivalent adjustments can be made based on home sleep comfort, headband structure, and electrode contact stability. For example, the system can use a low-channel arrangement combining forehead acquisition electrodes and a behind-the-ear reference electrode, or a multi-channel combination of forehead, frontal side, and periauricular region to improve the robustness of sleep stage recognition and awakening risk assessment.

[0062] Regarding acquisition parameters, the EEG signal sampling rate can be no less than 100 Hz, preferably no less than 200 Hz. The system can perform bandpass filtering and power frequency notch processing on the acquired EEG signals. The bandpass filtering range can be 0.3–35 Hz, 0.5–35 Hz, or 0.5–40 Hz, and the power frequency notch frequency can be 50 Hz or 60 Hz. To balance the stability of sleep stage recognition and the real-time performance of bed control, the system uses a layered time window for EEG processing. Sleep stage recognition can use a 30-second basic staging window, with a 5–10 second sliding step to update the sleep stage and stage confidence; sleep stability and awakening risk assessment can use a 5–10 second sliding window; and the identification of suspected awakening, artifacts, and sudden body movement risks can use a 0.5–2 second short window. The output delay for sleep stage identification results can be 5–30 seconds, the output delay for sleep stability and awakening risk results can be 1–5 seconds, and the output delay for suspected awakening or artifact detection results can be 0.5–2 seconds. Bed posture adjustment decisions should prioritize the joint assessment of sleep stage identification results and short-window awakening risk results, rather than relying solely on the stage results at a single moment.

[0063] The signal preprocessing unit filters, detrends, normalizes amplitude, marks bad segments, and processes abnormal channels in the raw EEG signal. The artifact recognition unit identifies body movement artifacts, poor electrode contact artifacts, eye movement artifacts, electromyography artifacts, transient high-amplitude abnormal signals, and environmental electromagnetic interference. The system can determine the reliability of EEG signals within a given time window based on the EEG signal amplitude range, variance variation, effective frequency band energy ratio, power frequency interference intensity, electrode contact status, inter-channel consistency, number of consecutive abnormal windows, and body movement signal synchronization. For time windows marked as low-quality, the system can reduce the weight of their sleep stage identification results in control decisions or exclude them from bed adjustment criteria.

[0064] The EEG feature extraction unit is used to extract time-domain, frequency-domain, and time-frequency-domain features from the preprocessed EEG signals. These features may include one or more of the following: delta wave power (0.5–4 Hz), theta wave power (4–8 Hz), alpha wave power (8–13 Hz), σ wave power (12–16 Hz), beta wave power (13–30 Hz), relative power of each frequency band, frequency band power ratio, power spectral density, spectral entropy, sample entropy, waveform amplitude statistics, spectral change rate, and sleep spindle wave-related features. Specifically, delta wave features can be used to assist in identifying deep sleep stages; theta and alpha wave features can be used to assist in determining light sleep, wakefulness, or arousal trends; σ wave features can be used to assist in determining stable sleep states; and beta wave and high-frequency activity changes can be used to assist in determining suspected arousal, electromyographic interference, or sleep instability trends.

[0065] The sleep state recognition unit can employ rule-based discrimination models, support vector machines, random forests, gradient boosting trees, convolutional neural networks, recurrent neural networks, lightweight Transformer models, or combinations thereof to classify or score EEG features. The output of the sleep state recognition unit can include one or more of the following: wakefulness, light sleep stage, deep sleep stage, REM sleep stage, stable sleep state, unstable sleep state, suspected awakening state, suspected micro-awake state, sleep stability index, awakening risk level, and sleep stage confidence level.

[0066] The signal quality evaluation unit is used to generate EEG signal quality scores and sleep stage confidence scores. EEG signal quality scores can be determined based on electrode contact status, signal-to-noise ratio, effective bandwidth energy ratio, abnormal amplitude ratio, artifact window ratio, and multi-channel consistency. Sleep stage confidence scores can be jointly determined by the classification model output probability, continuous window staging consistency, bandwidth feature stability, and artifact ratio. When the sleep stage confidence score is higher than a preset threshold, the system can use the results of fine sleep stages such as wakefulness, light sleep, deep sleep, and REM sleep as posture adjustment control constraints.

[0067] When the confidence level of a sleep stage falls below a preset threshold, or the EEG signal quality score falls below a preset threshold, the system does not directly adopt the detailed sleep stage results, but downgrades them to coarse-grained states such as stable sleep, unstable sleep, and suspected arousal. Specifically, stable sleep indicates relatively stable EEG characteristics, minimal body movement, and a low risk of arousal; unstable sleep indicates increased fluctuations in EEG characteristics, increased body movement, or frequency band changes; and suspected arousal indicates an increased risk of arousal, enhanced alpha / beta activity, increased body movement, or continuous changes in staging results. After downgrading, the system can reduce the posture adjustment level, decrease the amplitude of a single adjustment, reduce the adjustment speed, extend the observation window, or temporarily suspend automatic posture adjustment.

[0068] Through the above-mentioned electrode arrangement, acquisition parameters, signal preprocessing, artifact recognition, feature extraction, sleep state recognition, and signal quality evaluation, the sleep stage recognition module can reduce the impact of EEG signal quality fluctuations on bed control in a home sleep environment, avoid directly using low-reliability sleep staging results as the core control basis, and thus reduce the risk of misintervention, missed intervention, or user awakening due to incorrect intervention.

[0069] Snoring and Breathing Event Sensing Module

[0070] The snoring and breathing event sensing module is used to acquire snoring sounds, breathing rhythms, breathing intervals, bed micro-vibrations, chest and abdominal movements, body movements, or other signals related to breathing status during the user's sleep process, and to assist in identifying snoring, abnormal breathing rhythms, suspected hypoventilation, or suspected breathing difficulties. In a preferred embodiment, the module includes at least a snoring sound acquisition unit and a bed pressure or micro-vibration acquisition unit. The snoring sound acquisition unit can be a microphone, bone conduction sensor, occipital vibration sensor, or headboard acoustic sensor; the bed pressure or micro-vibration acquisition unit can be located in the mattress, bed frame, bed legs, occipital area, or back support area.

[0071] In a preferred embodiment, the system comprehensively assesses the risk of snoring and respiratory status by considering snoring intensity, snoring frequency characteristics, snoring duration, respiratory rhythm fluctuations, abnormal respiratory intervals, body movement patterns, and body position. The snoring signal can first undergo background noise estimation, noise reduction, framing, and sound event segmentation. Then, candidate snoring events are identified based on short-term energy, zero-crossing rate, duration, dominant frequency range, band energy, MFCC characteristics, or snoring recognition probability. The system can identify sound segments with sound intensity or energy exceeding the environmental noise baseline, duration meeting preset conditions, and spectral characteristics matching a snoring pattern as snoring events. The aforementioned snoring threshold can be a device preset value or calibrated based on bedroom background noise, microphone location, and the user's historical snoring baseline. For example, the system can identify sound segments with sound intensity exceeding a preset threshold and lasting for several seconds as candidate snoring events, but this threshold is not the sole limitation of the invention.

[0072] In the enhanced implementation, the system can also connect to a blood oxygen sensor, a heart rate sensor based on photoplethysmography (PPG), a wrist or ring-type blood oxygen device, a millimeter-wave radar respiratory detection module, a chest and abdominal exercise belt, or an inertial measurement unit (IMU) attitude sensor. In this invention, the blood oxygen signal serves as an optional enhancement signal for respiratory risk grading, improving the reliability of the grading. Blood oxygen can be obtained through a fingertip pulse oximeter, wrist pulse oximeter, ring pulse oximeter, or wristband PPG module. Its function is to improve the reliability of risk level judgment based on the downward trend of blood oxygen when the system detects increased snoring, abnormal breathing intervals, or unstable respiratory rhythms. For example, continuous snoring accompanied by abnormal breathing intervals can be classified as medium risk; if further accompanied by a downward trend in blood oxygen, it can be classified as high risk, allowing the system to adopt a higher-level but still low-disturbance attitude intervention strategy. The downward trend in blood oxygen can be determined based on the magnitude of the decrease in SpO2 relative to the individual baseline, the duration of the decrease, the slope of the decrease, the minimum value, and the recovery speed. The system can use a preset blood oxygen decline threshold as an initial judgment condition. For example, when SpO2 declines to a preset ratio and continues for a preset time, it is considered an event of increased blood oxygen risk. Alternatively, it can establish an individual blood oxygen baseline based on the user's historical data from multiple nights and dynamically adjust the threshold.

[0073] Body position and pressure recognition module

[0074] The body position and pressure recognition module is used to determine the user's current sleeping posture and the pressure distribution in different areas of the body. This module can be implemented through a mattress pressure array, zoned pressure sensors, zoned support unit status changes, bed micro-motion sensors, IMU posture sensors, or pillow pressure sensors.

[0075] The body position and pressure recognition module can output whether the user is in a supine, left lateral, right lateral, prone, or semi-lateral position, and can also output the pressure distribution of the head, neck, shoulders, back, waist, hips, and legs. Body position information is used to help determine the appropriate bed intervention method. For example, when the user is supine and snoring continuously, the system can prioritize slightly raising or tilting the head of the bed; when the user is already in a lateral position but still snoring, the system can prioritize fine-tuning the head and neck angle or back support; when the user moves frequently, the system can reduce the frequency of active adjustments to avoid cumulative interference.

[0076] Sleep-Respiration Fusion Assessment Module

[0077] The sleep-breathing fusion assessment module integrates sleep stages, snoring patterns, respiratory rhythm, body position, body movement, and optional blood oxygenation signals to generate a respiratory risk level and intervention suitability. Intervention suitability serves as an intermediate control variable connecting sleep stage identification results with bed posture adjustment strategies, transforming the user's current sleep state, respiratory risk status, and body position into executable adjustment constraints.

[0078] This module does not directly trigger bed movements based on a single snoring threshold, but rather makes a comprehensive judgment according to the process of "multi-source signal input - preprocessing - feature extraction - normalization - risk level determination - intervention suitability mapping - signal conflict adjudication". Figure 3 The sleep-breathing fusion judgment logic is shown, which is used to represent the sleep state, respiratory risk characteristics, postural state, signal confidence and body movement state respectively after preprocessing and feature extraction of EEG, snoring, respiratory rhythm, body position pressure, body movement and optional blood oxygen signals, and further outputs respiratory risk level and intervention appropriateness.

[0079] Input signal preprocessing and feature extraction

[0080] The sleep-breathing fusion judgment module receives inputs from the sleep stage recognition module, the snoring and breathing event perception module, the body position and pressure recognition module, the body movement detection module, and the optional blood oxygenation module. Each input signal undergoes preprocessing, feature extraction, and reliability evaluation before entering the fusion judgment process. The preprocessing methods and feature indicators for various signals are shown in Table 1.

[0081] Table 1: Preprocessing methods and characteristic indicators of multi-source signals

[0082] Core state variables

[0083] Based on the output results of the above multivariate signals, the sleep-breathing fusion judgment module generates at least the following core state variables: sleep state S, respiratory risk level R, body position P, signal confidence Q, and body movement state M. The definitions of each variable are shown in Table 2.

[0084] Table 2: Core State Variables and Their Control Functions

[0085] Among them, sleep state S is not limited to the standard sleep stages in the medical diagnostic sense. Its main function is to provide control constraints for bed posture adjustment. When the quality of EEG signals and the confidence of sleep stages meet the preset conditions, the system can use fine sleep stages such as light sleep, deep sleep and REM sleep to participate in the judgment; when the quality of EEG signals is insufficient or the confidence of the stage is low, the system can downgrade to coarse-grained states such as stable sleep, unstable sleep or suspected arousal.

[0086] Feature normalization

[0087] To enable features from different types of sensors to be included in the same fusion judgment process, the system can normalize snoring risk features, respiratory rhythm risk features, and optional blood oxygen risk features. Normalization methods can include threshold grading, maximum and minimum normalization, deviation normalization relative to the individual baseline, probabilistic output, or probability output from a machine learning model.

[0088] In one implementation, the system normalizes snoring risk, respiratory rhythm risk, and blood oxygenation risk into three levels: low, medium, and high, or into risk scores ranging from 0 to 1. The snoring risk score can be determined by snoring intensity, duration, frequency of events, frequency band energy, and snoring recognition probability; the respiratory rhythm risk score can be determined by the statistical variance of respiratory intervals, the coefficient of variation of the respiratory cycle, the decrease in respiratory amplitude, and the suspected pause-like interval; the blood oxygenation risk score can be determined by the magnitude of SpO2 decrease, the duration of the decrease, the slope of the decrease, and the recovery speed.

[0089] Respiratory Risk Level Determination Rules

[0090] Based on the above normalization results, the sleep-breathing fusion judgment module can use weighted scoring to obtain a comprehensive respiratory risk score. The weights can be preset or dynamically adjusted based on sensor availability, signal reliability, and user historical data. When the blood oxygen signal is available and reliable, snoring risk, respiratory rhythm risk, and blood oxygen risk all participate in the fusion; when the blood oxygen signal is unavailable or has low reliability, the system can re-normalize the snoring risk and respiratory rhythm risk; when the snoring signal has low reliability, the system reduces the role of snoring risk in the fusion judgment. In addition to weighted scoring, the sleep-breathing fusion judgment module can also use decision trees, state machines, Bayesian models, random forests, support vector machines, neural networks, or other machine learning models to output respiratory risk levels or respiratory risk probabilities. All of the above models can be used to implement the risk classification logic of this invention and are not intended to limit the scope of protection of this invention.

[0091] The respiratory risk level R can be determined by a comprehensive respiratory risk score and rule corrections. The respiratory risk level includes at least low risk, medium risk, and high risk; in a preferred embodiment, it also includes a warning risk level. Exemplary rules for determining the risk level are shown in Table 3.

[0092] Table 3: Exemplary rules for determining respiratory risk level

[0093] In one implementation, if a comprehensive risk score within the range of 0 to 1 is used, states below a first threshold can be classified as low risk, states between the first and second thresholds as medium risk, and states above the second threshold as high risk. If the comprehensive risk score remains above the second threshold and does not improve after multiple levels of low-perturbation adjustment, it is considered a risk warning. The first and second thresholds can be determined using factory preset values, population data statistics, commonly used clinical experience ranges, or user-specific baseline calibration values.

[0094] Thresholds can be determined in the following ways: First, use the factory-preset thresholds of the device as initial parameters; second, establish individual snoring baselines, respiratory rhythm baselines, and blood oxygen baselines in the early stages of sleep or in sleep data from multiple consecutive nights; third, make dynamic adjustments based on sensor sensitivity, mattress characteristics, user body type, and historical feedback results.

[0095] Signal conflict adjudication rules

[0096] Signals from different sensors may conflict with each other. To improve the reliability of fusion judgment, the system makes decisions based on signal confidence Q and multi-source consistency when generating respiratory risk levels, rather than letting any single signal directly determine bed movement.

[0097] In one implementation, the system can make decisions according to the following principles.

[0098] First, the principle of prioritizing reliability. If the quality score, recognition confidence, or continuous stability of a certain input signal is lower than a preset threshold, the system reduces the weight of that signal in risk grading. If the snoring signal is disturbed by environmental noise, bed friction noise, the voice or body movement of the bed partner, the system will not directly trigger bed posture adjustment based on the sound event; if the blood oxygen signal has poor contact, low waveform quality, or abnormal abrupt changes, the system will not directly raise the risk level based on the blood oxygen change.

[0099] Second, the principle of multi-source consistency. If at least two respiratory-related signals among snoring, respiratory rhythm, bed micro-vibration, body position pressure, body movement, and optional blood oxygenation signals are temporally adjacent and both point to an abnormal state, the system can increase the respiratory risk level; if only a single signal is abnormal, while other signals are stable or contradictory, the system can maintain the original risk level, determine it as medium risk, extend the observation window, or wait for more signals to confirm.

[0100] Third, the principle of risk conservatism. If risk signals such as a credible downward trend in blood oxygen, abnormal respiratory intervals, or a decrease in respiratory amplitude persist, even if the snoring signal is not obvious, the system can still raise the respiratory risk level based on the degree of respiratory rhythm abnormality and the quality of blood oxygen signal; if the snoring signal is strong but blood oxygen is stable and the respiratory rhythm is not obviously abnormal, the system generally does not directly determine it as high risk, but can determine it as medium risk or extend the observation window.

[0101] Fourth, the principle of excluding bodily movements. If a sound event occurs simultaneously with turning over, sudden changes in the center of pressure, bed friction, or bodily movement signals, the system reduces the credibility of the sound event as a snoring event, avoiding misjudging ordinary bodily movements or bed noise as snoring or abnormal breathing.

[0102] Fifth, the principle of degrading due to missing or conflicting signals. If key input signals are missing, there are significant conflicts between multiple signals, or multiple source signals are insufficient to support fine-grained risk classification, the system enters a conservative judgment mode. In conservative judgment mode, the system can reduce the intervention level, extend the observation window, increase the monitoring frequency, or only allow minimal, low-speed, and reversible attitude fine-tuning.

[0103] In conservative judgment mode, if the respiratory risk does not reach the high-risk condition, the system will not perform progressive mechanical adjustments, but will prioritize maintaining the current bed posture and continuing monitoring. If multiple signals continuously indicate a high respiratory risk, the system can perform fine-tuning under low-disturbance constraints, or trigger gentle reminders, user terminal prompts, or nursing assistance prompts. Through the above decision rules, the system can reduce false triggers in the case of inconsistent or low-confidence multiple signals, while retaining the necessary low-disturbance intervention capability in the case of a sustained high-risk state.

[0104] The above-mentioned adjudication rules are exemplary rules. The system can also combine, adjust or extend the conflict adjudication methods based on sensor configuration, signal quality, individual user baseline and historical feedback results.

[0105] Intervention suitability assessment

[0106] Intervention suitability (A) characterizes whether bed posture adjustment is appropriate in the current state, and the limits of permissible adjustment intensity. Intervention suitability is not determined solely by respiratory risk level, but rather by a combination of sleep state (S), respiratory risk level (R), postural state (P), signal confidence (Q), and body movement state (M).

[0107] In one implementation, the appropriateness of intervention can be divided into six categories, from A0 to A5, as shown in Table 4.

[0108] Table 4: Definition of Intervention Appropriateness

[0109] The system can first establish a basic intervention suitability based on sleep state (S) and respiratory risk level (R), and then refine it based on postural state (P), signal reliability (Q), and body movement state (M). The basic mapping relationship is shown in Table 5.

[0110] Table 5: Mapping Relationship between Sleep Status and Respiratory Risk Level and Appropriateness of Basic Intervention

[0111] The matrix described above is constructed based on the following control principles: avoiding accidental triggering of automatic actions in awake or suspected awake states; relatively high tolerance for low-disturbance regulation during light sleep; prioritizing sleep continuity during deep sleep, allowing cautious intervention only in cases of high respiratory risk; employing conservative regulation strategies during REM sleep; and reducing the level of automatic regulation when sleep staging is inaccurate or when suspected awakening occurs to minimize the risk of accidental intervention. This matrix is ​​used to determine the initial intervention suitability and can be adjusted in practice based on the user's individual baseline, sensor configuration, bed structure, and historical intervention effects.

[0112] Posture, credibility, and body movement correction rules

[0113] Postural status (P) is used to correct the suitability of the basic intervention and influence the selection of subsequent posture adjustment subjects. If the user is in a supine position and is snoring continuously, the system can increase the intervention priority without violating the safety constraints of the sleep stage; if the user is already in a side-lying or semi-side-lying position, the system prioritizes fine-tuning of the head and neck, occiput, or back, rather than prioritizing lateral tilting movements; if the user is in a prone position, abnormal position, or uncertain position recognition, the system reduces the intervention level or suspends automatic adjustment.

[0114] The signal confidence level (Q) is used to determine whether to employ fine-grained judgment or downgraded control. When the confidence level of EEG staging is low, the confidence level of snoring recognition is low, the body position recognition is uncertain, the blood oxygen signal quality is insufficient, or there are conflicts between multiple signal sources, the system can reduce the intervention level, extend the observation window, or enter a conservative control mode. If the EEG signal quality is insufficient, the system does not directly adopt fine-grained staging results such as light sleep, deep sleep, or REM sleep, but downgrades to coarse-grained states such as stable sleep, unstable sleep, or suspected awakening; if the confidence level of the snoring signal is low, the system does not directly trigger posture adjustment based on the sound event.

[0115] The body movement status M is used to limit continuous movements and avoid cumulative disturbances. When the user's body movement is stable or only slight, the system can execute a low-disturbance strategy according to the appropriateness of intervention; when the user moves frequently or is suspected of turning over, the system suspends automatic posture adjustment; when the bed movement results in increased body movement, sudden pressure changes, or an increased risk of arousal, the system reduces the amplitude and speed of subsequent adjustments, suspends continuous adjustments, or performs a rollback.

[0116] Through the above-mentioned fusion judgment method, the system transforms sleep state, respiratory risk, body position, signal credibility, and body movement state into executable intervention boundaries, enabling the subsequent posture adjustment decision module to clearly determine whether to adjust, when to adjust, the upper limit of adjustment intensity, and whether to suspend, revert, or prompt.

[0117] Attitude Adjustment Decision Module

[0118] The posture adjustment decision module generates a low-disturbance posture adjustment strategy based on the respiratory risk level, intervention suitability, postural status, signal reliability, and body movement status output by the sleep-breathing fusion judgment module, combined with the user's historical intervention effects. The posture adjustment strategy includes the adjustment object, adjustment amplitude, adjustment speed, adjustment sequence, action interval, observation window, cumulative adjustment limit, whether continuous adjustment is allowed, and whether rollback is required.

[0119] First, the posture adjustment decision module determines the action boundaries based on the intervention suitability. When the intervention suitability indicates that the current state is not suitable for mechanical movement, the system does not perform bed adjustments, but only records the state, increases the monitoring frequency, or waits for the state to stabilize. When the intervention suitability allows for fine-tuning or low-disturbance adjustment, the system generates candidate actions only within the limited amplitude, speed, and cumulative adjustment limits. When the intervention suitability indicates a high respiratory risk but requires cautious intervention, the system adopts a small-amplitude, low-speed, phased posture adjustment method, and generates the observation window and retreat conditions simultaneously as part of the strategy. The correspondence between intervention suitability and posture adjustment strategies can be shown in Table 6.

[0120] Table 6: Correspondence between intervention suitability and posture adjustment strategies

[0121] Secondly, the posture adjustment decision module selects candidate adjustment targets based on the user's position. If the user is supine and snoring or has an abnormal breathing rhythm, the system may prioritize slightly raising the head of the bed, or making minor adjustments to head, neck, or occiput support. If the breathing condition does not improve after these adjustments and the user's sleep is stable, further back-related adjustments or lateral tilt assistance may be considered. If the user is on their side but still snoring or has an abnormal breathing rhythm, the system prioritizes minor adjustments to head, neck, occiput, or shoulder and back support, and does not prioritize lateral tilting. If the user is in a semi-lateral position, the system prioritizes maintaining the current position and only makes minor support corrections. If the user is prone, in an abnormal position, or in an uncertain position, the system reduces the automatic adjustment level and may pause mechanical actions or trigger prompts if necessary. The correspondence between the position and the priority adjustment targets can be shown in Table 7.

[0122] Table 7: Postural Status and Priority Adjustment Targets

[0123] The posture adjustment decision module generates specific adjustment parameters based on candidate adjustment objects. These parameters may include single adjustment amplitude, adjustment speed, action duration, action interval, observation window, cumulative adjustment upper limit, and retreat amplitude. These parameters can be determined based on intervention suitability, respiratory risk level, sleep state, body position, bed actuation capability, mattress characteristics, user body type, and historical response results. They can also use factory preset ranges as initial values ​​and be individually calibrated based on user feedback. Exemplary value ranges for low-disturbance posture adjustment parameters are shown in Table 8.

[0124] Table 8: Exemplary value ranges for low-disturbance attitude adjustment parameters

[0125] The parameters described above are exemplary ranges and are not intended to limit the scope of protection of this invention. In practical applications, the system can be adjusted based on the bed structure, drive mechanism capability, mattress firmness, user weight, user body type, sleep state, and historical intervention feedback. For example, users with greater weight or softer mattresses may require larger changes in support to produce effective posture changes; users who are sensitive to bed movements or easily disturbed during deep sleep should have their single adjustment amplitude and speed reduced, and their observation window extended.

[0126] The observation window is a crucial component of the posture adjustment strategy, connecting the execution of an action with post-adjustment feedback evaluation. The system simultaneously determines the corresponding observation window length when generating each candidate adjustment action. The observation window is used to evaluate the short-term impact of the adjustment on snoring, respiratory rhythm, body movement status, EEG arousal risk, and optional blood oxygenation changes after the action is completed. For actions with a smaller impact range, such as minor adjustments to the occiput or head and neck, a shorter observation window can be set; for actions with a larger impact range, such as combined head and back adjustments, mattress zone support, or lateral tilt assistance, a longer observation window can be set. If the observation window is too short, it may lead to misjudgment before respiratory improvement is observed; if the observation window is too long, it may delay necessary subsequent interventions. Therefore, the observation window can be dynamically adjusted based on the subject of adjustment, respiratory risk level, sleep state, and historical feedback results.

[0127] To further reduce the impact of bed movements on sleep continuity, the posture adjustment decision module can establish a pre-emptive mechanism for predicting the risk of arousal through pre-emptive actions. This mechanism estimates the likelihood of candidate pre-emptive actions causing arousal or increased body movement based on sleep state, body position, historical action type, historical adjustment amplitude, historical adjustment speed, past arousal responses, changes in body movement, and individual user sensitivity. When selecting from multiple candidate strategies, the system prioritizes actions with lower expected disturbances and a higher probability of improving breathing. This prediction mechanism can be implemented using rule scoring, historical statistics, moving averages, linear models, decision trees, Bayesian models, reinforcement learning models, or other machine learning models.

[0128] In one implementation, the system can calculate a low-perturbation priority for each candidate adjustment action. The low-perturbation priority can comprehensively consider the expected improvement in breathing, the expected risk of arousal, the amplitude of the action, the speed of the action, and the historical response effect. If an action repeatedly reduces snoring or improves breathing rhythm in similar sleep states and postures, and does not cause an EEG arousal trend or an increase in body movement, the system increases the priority of that action; if an action repeatedly causes an increase in body movement, suspected micro-arousal, or user discomfort, the system decreases the priority of that action, reduces its adjustment amplitude, or restricts its use scenarios.

[0129] The posture adjustment strategy employs a progressive execution approach. Progressive execution means that the system does not perform large-amplitude adjustments all at once, but instead first performs the smallest candidate action that meets the current intervention appropriateness, and evaluates respiratory improvement and sleep disturbances within the corresponding observation window. If the feedback results show improved respiratory status and no EEG arousal trend or increased body movement, the current posture is maintained; if the respiratory status does not improve but the sleep status is stable, the next level of low-disturbance adjustment can be entered within permissible limits; if the feedback results show increased risk of EEG arousal, suspected micro-arousal, increased body movement, or abnormal pressure distribution, the subsequent adjustment amplitude is reduced, the adjustment speed is reduced, continuous adjustment is paused, or a regression is implemented.

[0130] Through the above decision-making process, the posture adjustment decision module can transform the respiratory risk level and intervention suitability output by the sleep-breathing fusion judgment module into a low-disturbance posture adjustment strategy with a clear action target, adjustment parameters, observation window and feedback conditions, so that bed adjustment can form a closed-loop control between respiratory risk relief, sleep continuity protection and individualized adaptation.

[0131] Bed execution module

[0132] The bed execution module is used to execute the low-disturbance posture adjustment strategy generated by the posture adjustment decision module, including one or more of the following: headboard lifting mechanism, back support mechanism, lumbar support mechanism, leg support mechanism, pillow adjustment mechanism, mattress zone support mechanism, and lateral tilt assist mechanism. Each posture adjustment mechanism can act on the headboard area, back area, lumbar area, leg area, pillow area, mattress zone support area, and left and right lateral tilt areas, respectively.

[0133] The headboard lifting mechanism, back support mechanism, lumbar support mechanism, and leg support mechanism can be implemented using electric push rods, linear actuators, lead screw lifting mechanisms, linkage mechanisms, or motor-driven hinge structures to adjust the lifting angle and support posture of the corresponding bed panels. The pillow adjustment mechanism can be implemented using a micro-lifting structure, airbag support unit, or flexible support structure to adjust the height and angle of the user's head and neck. The mattress zone support mechanism can be implemented using airbag arrays, zone lifting plates, flexible support plates, or combinations thereof to adjust the local height, firmness, and pressure distribution of the head and neck area, shoulder and back area, waist and hip area, and leg area. The side tilting assist mechanism can be implemented using left and right zone airbags, left and right zone lifting plates, or bed frame side tilting drive mechanism to guide the user from a supine position to a semi-lateral or slightly lateral position in a low-speed, small-amplitude manner when the user snores or has abnormal breathing rhythm while lying on their back.

[0134] Figure 4A This is a schematic diagram of the layout of the seven-zone support and corresponding actuators of the smart bed according to an embodiment of the present invention. The diagram shows the pillow support zone 1, the shoulder and back support zone 2, the waist support zone 3, the hip and buttock support zone 4, the thigh support zone 5, the calf support zone 6, the foot support zone 7, and their corresponding support actuators 8 to 14. Figure 4B This is a schematic diagram of the left and right tilt adjustment mechanism of the smart bed according to an embodiment of the present invention, showing the adjustment method and tilt angle range of the left and right tilt. Figure 4C This diagram illustrates the main posture adjustment methods and overall mechanical adjustment range of the smart bed according to an embodiment of the present invention. It exemplarily shows the linkage control of headboard lifting, back support, lumbar support, leg lifting, pillow height adjustment, left and right tilting, mattress zone support adjustment, and sit-up mode.

[0135] Feedback and Evaluation Module

[0136] The feedback evaluation module is used to continue collecting data such as EEG, snoring, respiratory rhythm, body position, body movement, and optional blood oxygen after the bed posture adjustment, and to determine whether the posture adjustment improved the breathing state, whether it caused sleep disturbance, and whether the change can be attributed to the posture adjustment action just performed.

[0137] This module does not simply compare changes in a single indicator before and after adjustment. Instead, it employs an evaluation process of "baseline window before adjustment—observation window after adjustment—signal reliability verification—attribution judgment of changes." The baseline window before adjustment records snoring, respiratory rhythm, body movement, EEG arousal risk, and optional blood oxygenation status for a period of time before adjustment. The observation window after adjustment records changes in the same indicators for a period of time after adjustment. By comparing the changes in indicators within the two time windows, the system makes a preliminary judgment on whether the adjustment action resulted in improved breathing or sleep disturbance.

[0138] In one embodiment, the baseline window before adjustment can be 30–120 seconds before adjustment, and the observation window after adjustment can be 30–180 seconds after adjustment. For fine-tuning of the occiput, head and neck, or headboard, a shorter observation window can be used; for combined backrest adjustment, lateral tilt assistance, or mattress zone support adjustment, a longer observation window can be used. The above time ranges can be adjusted according to the adjustment target, respiratory risk level, user sleep state, bed structure, and historical feedback results, and are not intended to limit the scope of protection of this invention.

[0139] The feedback evaluation module can extract the feedback evaluation indicators shown in Table 9.

[0140] Table 9: Adjusted Feedback Evaluation Indicators

[0141] To reduce misattribution caused by natural sleep cycle transitions, spontaneous body movements, or environmental noise, the feedback evaluation module allows setting feedback attribution conditions. The system records a change as a valid response related to the current adjustment only if the improvement in respiratory-related indicators occurs within the post-adjustment observation window, is temporally adjacent to the adjustment action, and is not accompanied by significant spontaneous turning over, natural sleep stage transitions, low sensor confidence levels, or environmental noise interference.

[0142] In one implementation, the system may employ the following attribution rules: First, the time proximity rule. Improvement in respiratory status should occur within the observation window after the adjustment is completed. If improvement occurs before the adjustment or in a period of time far from the adjustment action, it is not directly attributed to this adjustment.

[0143] Second, the multi-source consistency rule. The system prioritizes at least two types of respiratory-related indicators showing improvement in the same direction, such as a reduction in snoring events accompanied by a stable respiratory rhythm, or a reduction in abnormal respiratory intervals accompanied by a relief of the downward trend in blood oxygenation. If only a single indicator improves while other signals remain unchanged or conflict with each other, the reliability of the feedback is reduced.

[0144] Third, the disturbance exclusion rule. If a large spontaneous body movement, turning over, sudden change in the center of pressure, or natural change in body position occurs immediately after adjustment, the system reduces the attribution weight of this feedback to the current bed movement, so as to avoid misjudging the reduction of snoring caused by spontaneous turning over as the effectiveness of bed adjustment.

[0145] Fourth, signal reliability rules. If the quality of EEG, snoring, breathing, body position, or blood oxygenation signals before and after adjustment is lower than a preset threshold, or if there is a significant conflict between multiple source signals, the feedback result is marked as low reliability and is not used or has a reduced weight for individualized learning updates.

[0146] Fifth, the sleep stage stability rule. If a natural transition in sleep stage occurs before and after the adjustment, such as a shift from deep sleep to light sleep or a tendency to transition from REM sleep to wakefulness, the system should combine EEG stability and body movement status to determine whether this change is likely caused by a natural change in the sleep cycle. If the natural transition factor cannot be ruled out, the attribution confidence of this feedback should be reduced.

[0147] In one implementation, the system can generate feedback confidence for each adjustment feedback. Feedback confidence can be determined based on temporal proximity, multi-source consistency, signal quality, body movement exclusion results, and sleep stage stability. Feedback confidence can be categorized as high, medium, and low confidence. High-confidence feedback can be used for maintenance, progression, rollback, and individualized learning updates; medium-confidence feedback can be used for short-term policy adjustments, but its weight in individualized learning is reduced; low-confidence feedback is not used for individualized learning updates, or is only saved as anomaly records.

[0148] The feedback evaluation module can categorize each adjustment result into the following types.

[0149] Table 10: Feedback Results After Adjustment and Subsequent Control Actions

[0150] For machine learning or personalized learning methods, the feedback evaluation module can use sleep state, respiratory risk level, postural state, signal confidence, body movement state, regulation object, regulation amplitude, regulation speed, observation window, changes in indicators before and after regulation, and feedback confidence as training or update features. The system can use moving average, exponentially weighted average, linear regression, decision tree, Bayesian model, reinforcement learning model, or other online learning methods to update personalized parameters. To reduce the influence of confounding variables, the system can reduce the weight of low-confidence feedback, exclude samples with obvious spontaneous body movement or natural transitions between sleep stages, and prioritize the use of high-confidence feedback samples to update the user's minimum effective regulation amplitude, preferred action type, and easily aroused action type.

[0151] Through the aforementioned feedback evaluation and attribution judgment mechanism, the system can distinguish between "improvements caused by posture adjustment" and "accidental changes caused by natural sleep changes, spontaneous turning over, or signal noise," thereby reducing the impact of misattribution on subsequent individualized learning modules and improving the reliability of long-term strategy updates.

[0152] Personalized learning module

[0153] The individualized learning module records users' long-term sleep data and intervention results to form control parameters tailored to individual users. This module can record the frequency of snoring at different sleep stages, the risk of snoring in different body positions, the effect of different bed angles on improving snoring and breathing rhythm, the impact of different adjustment amplitudes on the risk of awakening, the user's sensitivity to bed movements, the minimum effective adjustment amplitude, and the adjustment method most likely to cause awakening.

[0154] The individualized learning module can record the state variables and feedback results corresponding to each intervention event. The state variables may include sleep state (S), respiratory risk level (R), postural state (P), signal confidence level (Q), body movement state (M), intervention suitability (A), intervention target, intervention amplitude, intervention speed, action duration, observation window, changes in snoring before and after intervention, changes in respiratory rhythm, changes in body movement, changes in EEG arousal risk, and optional changes in blood oxygenation. The feedback results may include effective and low-perturbation, ineffective but low-perturbation, effective but high-perturbation, ineffective and high-perturbation, and low-confidence feedback.

[0155] In one implementation, the individualized learning module can use each nightly sleep session as the basic update cycle, summarizing and updating the data for that night after the user's sleep session ends; alternatively, it can perform periodic updates after accumulating data over several consecutive nights, such as every 3 nights, every 7 nights, or every week. For events with high risk or clear feedback results, the system can also partially update relevant parameters that night, without immediately and significantly changing the global control strategy, to avoid policy fluctuations caused by accidental events.

[0156] The individualized learning module can be implemented using moving average, exponentially weighted average, linear regression, logistic regression, decision tree, Bayesian optimization, reinforcement learning, or a combination thereof. The above learning algorithms are used to update control parameters and do not limit the scope of this invention. In a simpler implementation, the system can update the user's minimum effective adjustment range using rule updates or moving averages; in an enhanced implementation, the system can use linear regression, logistic regression, or decision tree models to predict the probability of respiratory improvement and the probability of arousal risk corresponding to different adjustment actions; in a further implementation, the system can use Bayesian optimization or reinforcement learning methods to progressively find more suitable adjustment objects, adjustment ranges, and adjustment sequences for the user within safety constraints.

[0157] In one implementation, the individualized learning module can update the following control parameters: Table 11: Updateable Control Parameters of the Personalized Learning Module

[0158] When updating historical data, the individualized learning module can assign different weights based on signal confidence and feedback confidence. High-confidence feedback can be used to update individualized parameters; medium-confidence feedback can participate in the update with reduced weight; low-confidence feedback, feedback accompanied by obvious spontaneous turning over, feedback during natural transitions in sleep stages, or feedback with obvious conflict between multiple sources can be excluded from learning updates or simply saved as anomaly records. In this way, the system can reduce the impact of misattribution or low-quality signals on the individualized strategy.

[0159] In one implementation of the rule update, if an adjustment action is repeatedly determined to be "effective and low disturbance" under the same or similar conditions, the system can increase the priority of the action and use its corresponding minimum effective amplitude as the starting adjustment amplitude under subsequent similar conditions; if an adjustment action is repeatedly determined to be "ineffective" or "high disturbance", the system can reduce the priority of the action, reduce its amplitude, reduce its speed, or restrict the use of the action under the same conditions.

[0160] Through the aforementioned individualized learning mechanism, the system can gradually transition from fixed rules to user-oriented adaptive control. For users who snore easily during light sleep in a supine position and have good tolerance for headboard elevation, the system can gradually prioritize slight headboard elevation. For users who are sensitive to headboard elevation but not to slight lateral tilting, the system can increase the priority of lateral tilt assistance. For users who are easily awakened by movements during deep sleep, the system can reduce the amplitude and speed of single adjustments and extend the observation window during deep sleep. Therefore, the system can improve breathing during long-term use while reducing the impact of bed movements on sleep continuity.

[0161] Safety protection and fault degradation module

[0162] The safety protection and fault degradation module monitors the health status of sensors, data reliability, actuator operating status, and communication status during system operation. In case of anomalies, it executes safety protection strategies such as pausing, reverting, degrading operation, or issuing warnings. This module serves as a horizontal safety layer within the intelligent bed control system, working collaboratively with the sleep stage recognition module, snoring and breathing event sensing module, body position and pressure recognition module, sleep-breathing fusion judgment module, posture adjustment decision module, bed execution module, and feedback evaluation module.

[0163] In one implementation, the safety protection and fault degradation module can monitor the health status of the EEG headband, electrode contact status, microphone, mattress pressure sensor, bed micro-vibration sensor, body position sensor, blood oxygen sensor, and communication link. The sensor health status can be determined based on whether the signal is missing, exceeds a reasonable range, remains constant, exhibits abrupt changes, drifts, has a signal-to-noise ratio below a preset threshold, has decreased recognition confidence, and whether there is significant conflict between multiple source signals.

[0164] When the EEG headband falls off, electrode contact remains abnormal, EEG signal quality falls below a preset threshold, or reliable EEG characteristics cannot be obtained within multiple consecutive time windows, the system will no longer use the results of fine sleep stages as a constraint for posture regulation control, but will instead downgrade to coarse-grained states such as stable sleep, unstable sleep, or suspected arousal. If EEG signals are completely unavailable, the system can pause active posture regulation based on fine sleep stages, retaining only conservative judgments based on snoring, respiratory rhythm, body movement, and body position signals; when respiratory risk does not reach high-risk conditions, the system will prioritize maintaining the current posture and increase the monitoring frequency.

[0165] When mattress pressure sensors, body position pressure arrays, or body motion sensors exhibit reading drift, constant fluctuations, abrupt changes, local sensor point failures, or insufficient reliability in body position recognition, the system reduces the role of body position in the adjustment strategy, refrains from performing tilting movements that rely on direction determination, and marks the body position as uncertain. If the pressure or body position signal remains abnormal, the system may pause automatic posture adjustment or only allow minimal adjustments to the headboard or pillow area that are independent of body position direction.

[0166] When the microphone, bone conduction sensor, or bedside acoustic sensor is affected by environmental noise, the voice of the bedmate, bed friction noise, or communication abnormalities, causing the snoring recognition confidence level to fall below a preset threshold, the system will not directly trigger posture adjustment based on this sound event. In this case, the system can extend the observation window and make a conservative judgment by combining respiratory rhythm, body movement, body position, and optional blood oxygenation signals; if multiple signals do not jointly point to a higher respiratory risk, the system will maintain the current posture and continue monitoring.

[0167] The bed actuator module can provide feedback to the safety protection and fault degradation module on the position, angle, speed, drive current, motor temperature, running time, airbag pressure, valve status, deviation between target and actual positions, and whether there is jamming, overcurrent, overheating, or abnormal air pressure of each adjustment mechanism. If actuator jamming, abnormal increase in drive current, motor temperature exceeding a preset threshold, abnormal airbag pressure, position feedback deviation from target position continuously exceeding a preset range, or adjustment action timeout is detected, the system stops the current action and, according to the safety strategy, maintains the current position, returns to the nearest safe posture, or restores to the preset safe posture.

[0168] In one implementation, the safety protection and fault degradation module can employ a tiered degradation strategy. Level 1 degradation maintains the current posture and increases monitoring frequency; Level 2 degradation allows only minimal, low-speed, reversible posture fine-tuning; Level 3 degradation suspends all automatic mechanical adjustments, only recording status and providing risk alerts; Level 4 degradation triggers user terminal prompts, gentle reminders, or nursing assistance prompts. The degradation level can be determined based on the type of signal anomaly, duration of the anomaly, respiratory risk level, actuator status, and the user's current sleep state.

[0169] When the system enters degraded operation mode, the individualized learning module does not use low-reliability data or data from the period of failure to update the user's long-term control parameters, or only records anomalous samples with low weight. After the sensor signals recover, the actuator status is normal, and the reliability requirements are met for multiple consecutive time windows, the system can exit degraded mode and resume the normal fusion judgment and attitude adjustment strategy.

[0170] Through the aforementioned safety protection and fault degradation mechanisms, the system can avoid erroneous intervention caused by low-reliability signals or actuator failures in cases such as EEG headband detachment, abnormal sensor data, pressure sensor drift, acoustic signal contamination, communication abnormalities, actuator jamming, overheating, and overcurrent, thereby improving the safety and reliability of the smart bed during long-term use.

[0171] method

[0172] The method of the present invention can be implemented according to the following steps.

[0173] Step S1: Collect EEG signals during the user's sleep process, and perform signal quality evaluation, preprocessing, artifact identification, and feature extraction on the EEG signals. Based on the above features, the system identifies the user's current sleep stage, sleep stability, awakening risk, and sleep stage confidence. The sleep stage may include wakefulness, light sleep, deep sleep, and REM sleep, and may further include stable sleep, unstable sleep, and states of suspected awakening or suspected micro-awakening.

[0174] Step S2: Collect respiratory-related signals, position-related signals, and body movement-related signals during the user's sleep process. The respiratory-related signals may include snoring signals, respiratory rhythm signals, bed pressure signals, bed micro-vibration signals, body movement signals, and optional blood oxygenation signals; the position-related signals may include supine, lateral, prone, semi-lateral, and pressure distribution in different areas of the body.

[0175] Step S3: Determine the user's current respiratory risk level based on snoring intensity, snoring duration, changes in respiratory rhythm, abnormal breathing intervals, body position, body movement, and optional blood oxygenation changes.

[0176] Step S4: Determine the appropriateness of intervention based on sleep state, respiratory risk level, body position, signal reliability, and body movement status. The appropriateness of intervention is used to characterize whether bed posture adjustment is suitable in the current state, and the permissible adjustment intensity boundary.

[0177] Step S5: Generate a low-disturbance posture adjustment strategy based on respiratory risk level, intervention suitability, body position, body movement status, and historical intervention effects. The low-disturbance posture adjustment strategy may include the adjustment object, adjustment amplitude, adjustment speed, adjustment sequence, action interval, observation window, cumulative adjustment limit, and rollback conditions.

[0178] Step S6: Control at least one posture adjustment unit of the smart bed to execute the low-disturbance posture adjustment strategy. During execution, the single adjustment amplitude, adjustment speed, action duration, action interval, and cumulative adjustment limit can be limited, and small-amplitude, low-speed, phased, and reversible action modes are preferred.

[0179] Step S7: After the adjustment is completed, based on the baseline window before adjustment and the observation window after adjustment, continue to collect EEG signals, snoring signals, respiratory rhythm signals, body position signals, body movement signals and optional blood oxygen signals, and combine the signal reliability and body movement exclusion rules to determine whether the adjustment improves the respiratory state, whether it causes sleep disturbances, and whether the change can be attributed to this posture adjustment.

[0180] Step S8: Update the subsequent adjustment strategy based on the feedback evaluation results. If the adjustment is effective and does not cause sleep disturbance, maintain the current posture; if the adjustment is ineffective and does not cause sleep disturbance, proceed to the next level of adjustment; if the adjustment causes a tendency to wake up or an increase in body movement, reduce the amplitude of subsequent adjustments or pause the adjustment; if multiple posture adjustments fail to meet the preset improvement conditions and the respiratory risk continues to increase, trigger a gentle reminder, user terminal notification, or nursing assistance prompt. During the execution of the above steps, the system continuously monitors the health status of sensors, data reliability, actuator operating status, and communication status, and performs pause, rollback, degraded operation, or prompts in abnormal situations.

[0181] like Figure 5 As shown, the process includes collecting EEG data, identifying sleep stages, collecting snoring, respiratory status and body position signals, determining risk level, assessing intervention appropriateness, generating low-disturbance strategies, implementing adjustments, providing feedback and evaluation, and updating individualized parameters.

[0182] Gradual low-disturbance intervention strategy

[0183] A progressive, low-disturbance intervention strategy is a key element of this invention in achieving a balance between "respiratory risk mitigation" and "sleep continuity protection." Traditional smart beds or anti-snoring devices often execute fixed-amplitude bed movements immediately upon detecting snoring, focusing primarily on reducing snoring while paying less attention to the impact of the movements on sleep structure, awakening risk, and user comfort. This invention does not treat bed posture adjustment as a one-off action but rather organizes it into an observable, progressive, and reversible low-disturbance control process.

[0184] This strategy transforms bed movements from direct, fixed-amplitude adjustments into a low-disturbance adjustment process constrained by sleep state, respiratory risk, body position, signal reliability, body movement, and historical feedback. When generating posture adjustment strategies, the system considers not only the presence of snoring or breathing abnormalities but also sleep state, respiratory risk level, body position, signal reliability, body movement, and historical feedback. Therefore, bed adjustments no longer default to maximum amplitude or fixed movements but prioritize movement paths that minimize sleep disturbance and have the potential to improve breathing.

[0185] The progressive, low-disturbance intervention strategy also enables this invention to achieve closed-loop optimization. Changes in snoring, breathing rhythm, EEG arousal risk, body movement, and optional blood oxygenation after each adjustment can all serve as the basis for subsequent strategy adjustments. If the adjustment is effective and does not cause disturbance, the system can maintain its current posture; if the adjustment is ineffective or disturbance increases, the system can reduce its intensity, pause the progression, or revert to a previous state. Therefore, this invention avoids the problems of excessive movement, continuous movement, and misintervention that can occur with fixed anti-snoring adjustments.

[0186] Dynamic intervention strategies under sleep stage constraints

[0187] The dynamic intervention strategy under sleep stage constraints is the core control strategy that distinguishes this invention from traditional snoring threshold triggering schemes. Traditional anti-snoring devices typically use snoring intensity, body movement, or a single physiological threshold as the trigger condition for action, easily forming a "regulate upon detecting an abnormality" control mode. However, during actual sleep, users' tolerance to external mechanical movements varies depending on whether they are in light sleep, deep sleep, REM sleep, suspected awakening, or a state of frequent body movement. If the differences in sleep stages are ignored, even if regulation can reduce snoring, it may still trigger micro-awakening, increased body movement, or sleep fragmentation.

[0188] This invention incorporates sleep stages, sleep stability, and awakening risk into the posture adjustment control logic, so that bed movements are no longer determined solely by respiratory risk, but are simultaneously constrained by the timing and intensity of the sleep state. In other words, after recognizing snoring or abnormal breathing, the system does not immediately execute a fixed action, but further determines whether the current sleep state is suitable for intervention, what intensity of intervention is permissible, whether a pause is needed, whether a more conservative movement should be adopted, and whether a regression is necessary after adjustment.

[0189] This strategy elevates the technical focus of the invention from "simply reducing snoring" to "selecting appropriate times for low-disturbance intervention during the sleep cycle." When respiratory risk is low and the user is in a sleep state sensitive to intervention, the system can prioritize protecting sleep continuity; when respiratory risk is high, the system can perform more cautious posture adjustments under sleep stage constraints. Thus, the invention can dynamically determine whether to adjust, when to adjust, to what extent to adjust, and whether to maintain or revert the adjustment based on the user's current sleep state.

[0190] A dynamic intervention strategy under sleep stage constraints is adopted. The system determines the intervention threshold, allowable adjustment intensity, and feedback control mode in different sleep states by combining respiratory risk level and intervention suitability.

[0191] Specifically, the system employs differentiated intervention strategies for different sleep stages, as shown below: For the light sleep stage (N1 / N2), the EEG characteristics are mainly enhanced theta waves and weakened alpha waves (frequency range 4-8Hz). This stage is more easily awakened by external disturbances, and snoring often begins in this stage. However, it has relatively good tolerance to low-amplitude adjustments, making it a "golden window" for early intervention. The appropriate level of intervention is suitable for small-scale interventions. For medium-risk and above cases, small-scale interventions can be triggered, such as raising the head of the bed by 0-3° or adjusting the occiput by 0-10mm, with an adjustment speed of 0.1-0.3° / s or 1-2mm / s. After intervention, the focus should be on assessing changes in snoring intensity. The strategy aims to improve breathing with minimal movement to prevent forced large-scale interventions once deeper sleep has begun.

[0192] During deep sleep (N3), EEG is dominated by delta waves (frequency range 0.5~4Hz), with high sleep stability and a high arousal threshold. However, excessively large amplitude or speed of mechanical movements may still trigger body movements, micro-arousals, or sleep fragmentation. The appropriate level of intervention is default postponement or only minor adjustments. Extremely small, very slow adjustments are permitted only in cases of high risk and persistent abnormalities, such as amplitudes of 0~1°, 0~5mm, or 0~1° of lateral tilt, with speeds ≤0.1° / s or ≤1mm / s. Post-intervention assessment should focus on respiratory rhythm and the number of apneas. The strategy prioritizes protecting the structural integrity of deep sleep and avoiding unnecessary movement interventions.

[0193] For rapid eye movement (REM) sleep, EEG is characterized by low-amplitude mixed-frequency activity, active eye movements, vivid dreams, low muscle tone, sensitivity to changes in body position, and frequent respiratory events. The appropriateness of intervention is cautious (depending on risk). In medium-risk situations, cautious, small-amplitude adjustments can be made; in high-risk situations, cautious intervention is recommended, for example, with amplitudes of 0-2°, 0-8mm, or 0-2° of lateral tilt, and speeds of 0.05-0.2° / s or ≤1mm / s. Post-intervention assessment should focus on EEG arousal and micro-arousal characteristics. The strategy avoids large, structured movements performed in one go; multi-step, gradual interventions should be used when necessary.

[0194] For awake or frequently moving states, EEG is dominated by beta waves (frequency range >13Hz), including awake states or frequent turning over, coughing, getting up, etc. Respiratory signals are easily interfered with, posing a high risk of misinterpretation, and regulation is easily perceived as intervention. The appropriate level of intervention is to suspend automatic intervention, not to execute automatic regulation, and to temporarily suspend automatic intervention for all risk levels (although reminders or suggestions can be issued). After intervention, the focus is on assessing the frequency of body movement / turning over and optional changes in blood oxygenation trends. Strategy explanation: Suspend automatic intervention and reassess the need for intervention after returning to a stable sleep stage.

[0195] Through the aforementioned phased differentiated strategy, the system has formed a complete intervention process: "detecting snoring / breathing events → determining the current sleep stage → assessing the appropriateness of intervention → generating low-disturbance intervention strategies → implementing and evaluating the effects → updating individualized strategies".

[0196] Sensor configuration

[0197] In a specific embodiment of the present invention, the sleep stage recognition module, the snoring and breathing event sensing module, the bed posture execution module, and the feedback evaluation module constitute the core technical solution of the present invention, used to form a low-disturbance posture adjustment closed loop based on sleep state constraints; the sensor configuration can be flexibly selected according to actual needs. Breathing state sensing can be achieved through snoring, respiratory rhythm, bed pressure, bed micro-vibration, body movement, or a combination thereof. To further improve the reliability of snoring, respiratory rhythm, and body position judgment, a microphone, mattress pressure sensor, bed micro-vibration sensor, and body position pressure array can be preferably configured. Furthermore, depending on the product form and usage scenario, a blood oxygen sensor, PPG sensor, millimeter-wave radar, chest and abdominal exercise belt, IMU inertial measurement unit, or mobile terminal application can also be selected to enhance risk grading, individualized learning, or external linkage capabilities. Blood oxygen signal is an optional enhancement signal that can be used to improve the reliability of respiratory risk grading. It should be noted that the respiratory-related signals may include at least one of the following: snoring signals, respiratory rhythm signals, bed pressure signals, body position signals, and body movement signals; in optional embodiments, they may also include blood oxygenation signals, heart rate signals, or pulse wave signals.

[0198] Typical operating scenarios and control examples

[0199] The following typical operating scenarios illustrate the control logic of this system under different sleep stages, different respiratory risks, and different body positions. Each scenario is only used to explain the system's operation and does not limit the scope of protection of this invention. The sensor combinations, adjustment objects, adjustment amplitudes, adjustment speeds, and feedback indicators in different scenarios can be combined or replaced according to product form and usage environment.

[0200] Scenario 1: Bedside fine-tuning based on EEG sleep stages and snoring detection

[0201] The user wears a sleep EEG headband, and the smart bed collects snoring signals through microphones at the head or side of the bed and identifies the user's position through mattress pressure sensors. When the system detects that the user is in a light sleep stage, and the snoring intensity continues to exceed a preset threshold, and the user is in a supine position, the sleep-breathing fusion judgment module classifies this state as a moderate respiratory risk suitable for low-amplitude intervention.

[0202] The posture adjustment decision module generates a low-speed, micro-elevation strategy for the head of the bed, such as controlling the head or occipital support area to rise at a small angle at a speed below a preset threshold. After adjustment, the system enters an observation window to continue monitoring snoring intensity, respiratory rhythm, and EEG arousal characteristics. If snoring decreases and the EEG does not show an arousal trend, the current posture is maintained; if snoring does not decrease and the EEG remains stable, the next level of fine-tuning is initiated; if the EEG shows an arousal trend or increased body movement, further adjustment is paused or partially reversed.

[0203] Scenario 2: Phased Regulation of High-Risk, Low-Disturbance Breathing During Deep Sleep

[0204] The system identifies the user as being in deep sleep through electroencephalogram (EEG) signals, detects abnormal breathing rhythms through a mattress micro-vibration sensor, and detects continuous snoring through a snoring acquisition unit. In an optional implementation, a wrist-mounted or ring-mounted pulse oximeter further detects a downward trend in blood oxygen levels. The sleep-breathing fusion assessment module classifies this state as having a high respiratory risk, but because the user is in deep sleep, the intervention appropriateness is set to cautious intervention.

[0205] In this state, the system prioritizes low-speed, small-amplitude, and phased adjustments. For example, it first slightly elevates the head and neck support area with a relatively long observation window; if the breathing rhythm recovers or snoring decreases, the current posture is maintained; if there is no improvement, then small combined adjustments are made to the back support area; if the EEG shows a micro-awakening trend, further actions are immediately stopped. This scenario demonstrates the control logic of this invention, which balances respiratory risk mitigation and deep sleep protection in high-risk situations.

[0206] Scenario 3: Side tilt adjustment for supine induced snoring

[0207] The system identifies the user's supine position using a mattress pressure array and detects continuous snoring and abnormal breathing rhythm through microphone and bed micro-vibration signals. The system first performs a slight head-up movement and assesses whether the snoring and breathing rhythm improve within an observation window. If snoring persists after head-up and the EEG signal does not show a clear trend of awakening, the posture adjustment decision module switches to a tilt-assisted strategy.

[0208] Lateral tilt assistance can be achieved through left and right zone support units, shoulder and back local support units, or bed tilting mechanisms. The system slowly changes the support height on one side, allowing the user to transition from a fully supine position to a semi-recumbent or slightly lateral position. This adjustment guides the change in posture through gradual support changes, and uses EEG arousal responses and postural changes to determine whether to continue adjusting or retract. After adjustment, the system continues to monitor snoring, respiratory rhythm, body movement, and EEG stability, and decides to maintain, retract, or further fine-tune based on feedback.

[0209] Scenario 4: Personalized strategy updates based on long-term response records

[0210] The system continuously records the user's sleep data over multiple nights, including sleep stages, snoring intensity, body position, bed angle, mattress support, breathing rhythm, body movement, optional blood oxygen saturation changes, and EEG arousal responses. The individualized learning module uses historical data to identify the effectiveness and degree of interference of different posture adjustment methods for the user.

[0211] For example, the system detects that when a user snores during light sleep while lying on their back, raising the headboard at a preset small angle can reduce the snoring intensity without causing awakening. However, during deep sleep, similar headboard movements are more likely to cause body movement. Therefore, in subsequent controls, the system sets a relatively lenient headboard adjustment threshold during light sleep, while setting a more conservative adjustment range and speed during deep sleep. For another user, if the system finds that side tilt assist is more effective and less likely to cause awakening than headboard raising, then side tilt assist will be used preferentially in subsequent operations.

[0212] Public research basis and prototype evaluation indicators for posture adjustment

[0213] The technical effects of the low-disturbance posture adjustment strategy of this invention can be quantified by existing published sleep medicine and sleep engineering research. Related studies have shown that head-up elevation, upper body tilting, and postural adjustments can, to some extent, reduce snoring duration, decrease apnea-hypopnea events, or improve sleep stability. For example, a study by Souza et al. (2017, Sleep and Breathing) showed that a 7.5° head-up elevation reduced the apnea-hypopnea index (AHI) in obstructive sleep apnea (OSA) patients from 15.7 breaths / hour to 10.7 breaths / hour, and increased the lowest oxygen saturation from 83.5% to 87%, supporting the use of a small-angle head-up elevation as a low-disturbance breathing improvement strategy in this invention. Danoff-Burg et al. (2022, JMIR Formative Research) collected home data from 25 snorers over 8 weeks and more than 1,000 nights. The results showed that a 12° upper body tilt reduced the objective snoring duration by 7%, the number of awakenings by 4%, and the proportion of deep sleep by 5%, providing a basis for using slight bed tilt as a non-invasive anti-snoring method in the home environment. Iannella et al. (2022, Journal of Clinical Medicine) showed that a 30° head-chest elevation reduced the AHI from 23.8±13.3 to 17.7±12.4 in 45 OSA patients, and the proportion of apnea decreased from 55±28.1% to 44±25.8%, supporting the idea that combined head and back adjustments can improve upper airway patency in this invention. Furthermore, the postural therapy study by Srijithesh et al. (2019, Cochrane Systematic Review) showed that, compared to the non-intervention control, postural therapy reduced the average AHI by 7.38 times / hour and lowered the Epworth sleepiness score, supporting the use of lateral tilt assistance or semi-lateral guidance in this invention to address supine-induced snoring. The above research data demonstrate that changes in sleep posture, head-up, and upper body tilt have an experimental basis for improving snoring or respiratory events. Building upon this, this invention further introduces EEG sleep stage recognition, intervention suitability assessment, progressive low-disturbance adjustment, post-adjustment feedback evaluation, and individualized learning mechanisms. This transforms posture adjustment from a fixed-angle movement into a dynamic determination of adjustment timing, amplitude, speed, and regression conditions based on sleep stage, respiratory risk level, postural state, and historical feedback.

[0214] In summary, this invention provides a smart bed low-disturbance posture adjustment scheme based on sleep cycle perception and snoring and breathing state recognition. It uses EEG sleep stages, sleep stability, and awakening risk as timing and intensity constraints for bed posture adjustment. This eliminates reliance on a single snoring threshold or fixed physical sign triggering, instead dynamically determining respiratory risk levels and intervention appropriateness by fusing multi-source signals such as snoring, respiratory rhythm, postural pressure, body movement, and selectable blood oxygenation. Furthermore, the system employs a progressive, reversible low-disturbance adjustment strategy, starting with the smallest candidate action, evaluating breathing improvement and sleep disturbance within an observation window, and deciding whether to maintain, advance, or regress based on feedback results. This approach alleviates the risk of snoring or breathing difficulties while maximizing the protection of sleep continuity and sleep structure integrity.

[0215] This invention further introduces a post-adjustment feedback evaluation and individualized closed-loop learning mechanism. By comparing the pre-adjustment baseline window and the post-adjustment observation window, and combining attribution rules such as temporal proximity, multi-source consistency, disturbance exclusion, and signal credibility, the intervention effect is accurately judged and feedback confidence is generated. Individualized control parameters, including minimum effective adjustment amplitude, action thresholds at different sleep stages, easily aroused action types, and observation window duration, are then updated nightly or periodically. Thus, this invention forms a complete closed-loop control system: "sleep state perception—respiratory risk assessment—intervention suitability determination—low-disturbance posture adjustment—feedback evaluation—individualized strategy update." This system can dynamically adjust intervention strategies based on the user's real-time sleep state and respiratory risk, significantly reducing awakenings and sleep fragmentation caused by bed movements, improving respiratory patency, and enhancing overall sleep quality. It can be widely applied in home sleep, health and wellness spaces, nursing spaces, and smart bedrooms.

[0216] The above description, in conjunction with specific / preferred embodiments, provides a further detailed explanation of the present invention. It should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, various substitutions or modifications can be made to these described embodiments without departing from the inventive concept, and all such substitutions or modifications should be considered within the scope of protection of the present invention.

Claims

1. A method for low-disturbance posture adjustment of an intelligent bed based on sleep cycle perception and snoring and breathing state recognition, characterized in that, Includes the following steps: S1. Collect the EEG signals during the user's sleep process, and perform signal quality evaluation, preprocessing, artifact recognition and feature extraction on the EEG signals. Based on the extracted features, identify the user's current sleep stage, sleep stability or awakening risk, and generate sleep stage confidence. S2. Collect respiratory-related signals, body position-related signals, and body movement-related signals during the user's sleep process; the respiratory-related signals include at least one of snoring signals, respiratory rhythm signals, bed pressure signals, and bed micro-vibration signals; S3. Based on the snoring signal, respiratory rhythm signal, body position, and body movement, determine the user's current respiratory risk level through multi-source signal fusion; S4. Determine the appropriateness of intervention based on sleep state, respiratory risk level, body position, signal reliability, and body movement status; the appropriateness of intervention is used to characterize whether it is suitable to perform bed posture adjustment in the current state and the allowable adjustment intensity boundary; S5. Based on the respiratory risk level, intervention suitability, body position, and historical intervention effects, generate a low-disturbance posture adjustment strategy; the low-disturbance posture adjustment strategy includes the adjustment object, single adjustment amplitude, adjustment speed, action interval, observation window, cumulative adjustment limit, and retreat condition; S6. Control at least one posture adjustment unit of the smart bed to execute the low-disturbance posture adjustment strategy and adopt a progressive execution method: first execute the minimum candidate action that meets the current intervention suitability, evaluate breathing improvement and sleep disturbance within the observation window, and decide to maintain, advance or regress based on the evaluation results; S7. After the adjustment is completed, based on the baseline window before adjustment and the observation window after adjustment, continue to collect EEG signals, snoring signals, respiratory rhythm signals, body position signals and body movement signals. Combine the signal reliability and body movement exclusion rules to determine whether the adjustment improved the respiratory state, whether it caused sleep disturbances, and whether the change can be attributed to the posture adjustment. S8. Update subsequent adjustment strategies based on feedback evaluation results: If the adjustment is effective and does not cause sleep disturbance, maintain the current posture or record the candidate action as an effective response; if the adjustment is ineffective and does not cause sleep disturbance, proceed to the next level of low-disturbance adjustment within the allowable range of intervention suitability; if the adjustment causes an increased risk of arousal or increased body movement, reduce the amplitude of subsequent adjustments, reduce the speed of adjustment, suspend continuous adjustment, or perform a regression; and update individualized control parameters based on historical response data.

2. The method as described in claim 1, characterized in that, In step S1, identifying the user's current sleep stage, sleep stability, or awakening risk includes: performing EEG processing using a layered time window, wherein sleep stage identification uses a basic staging window of a first preset time length and is updated with a first sliding step size; sleep stability and awakening risk assessment use a sliding window of a second preset time length; and suspected awakening, artifacts, and sudden body movement risk identification use a short window of a third preset time length. When the confidence level of the sleep stage is higher than a preset threshold, the fine sleep stage results are used as control constraints. When the confidence level of the sleep stage is lower than a preset threshold or the EEG signal quality score is lower than a preset threshold, the state is downgraded to a coarse-grained state. The coarse-grained state includes stable sleep, unstable sleep, or suspected awakening, and the posture adjustment level is reduced, the single adjustment amplitude is reduced, the adjustment speed is reduced, the observation window is extended, or automatic posture adjustment is temporarily suspended.

3. The method as described in claim 1, characterized in that, In step S3, determining the user's current respiratory risk level through multi-source signal fusion includes: preprocessing and extracting features from snoring signals, respiratory rhythm signals, postural pressure signals, and body movement signals. Extracted features include one or more of the following: snoring intensity, snoring duration, event frequency, dominant frequency range, frequency band energy, Mel frequency cepstral coefficient features, respiratory cycle, respiratory interval statistical variance, respiratory cycle coefficient of variation, respiratory amplitude change, suspected pause-like interval, pulse oximetry saturation decrease magnitude, decrease duration, decrease slope, minimum value, and recovery speed. Each feature is normalized, and a weighted scoring or machine learning model is used to output a comprehensive respiratory risk score. Based on preset thresholds, the respiratory risk level is classified into low risk, medium risk, high risk, and warning risk. In case of signal conflict, a decision is made according to the principles of credibility priority, multi-source consistency, risk conservatism, body movement exclusion, and missing / conflicting downgrade.

4. The method according to claim 1, characterized in that, The signals used in the multi-source signal fusion in step S3 also include blood oxygen signals; the signals collected in step S7 also include blood oxygen signals.

5. The method as described in claim 1, characterized in that, In step S4, determining the appropriateness of intervention includes: first, establishing a basic mapping relationship of intervention appropriateness based on sleep state and respiratory risk level, and then correcting it based on body position, signal confidence, and body movement status; wherein, the appropriateness of intervention includes at least six levels: suspending automatic intervention, no motion monitoring, allowing only fine-tuning, suitable for low-disturbance adjustment, cautious phased adjustment, and gentle reminders or terminal prompts; when the confidence level of EEG staging is low, the confidence level of snoring recognition is low, the body position recognition is uncertain, or there is conflict between multiple sources of signals, the intervention level is reduced or the observation window is extended; when the user moves frequently or is suspected of turning over, automatic posture adjustment is suspended; when body movement increases or the risk of awakening increases after adjustment, the subsequent adjustment amplitude is reduced, the adjustment speed is reduced, or a rollback is performed.

6. The method as described in claim 1, characterized in that, In step S5, the generation of low-disturbance posture adjustment strategies further includes: selecting candidate adjustment objects based on body position; when the user is in a supine position, priority is given to slight head-up, head and neck or occiput support fine-tuning; when the user is in a side-lying position, priority is given to head and neck, occiput or shoulder and back support fine-tuning; the length of the observation window is dynamically adjusted according to the adjustment object, respiratory risk level, sleep state and historical feedback results; a shorter observation window is used for occiput or head and neck fine-tuning, and a longer observation window is used for back joint adjustment, side tilt assistance or mattress zone support adjustment; the method also includes establishing an adjustment action-awakening risk prediction mechanism, estimating the probability of candidate adjustment actions causing awakening or increased body movement based on sleep state, body position, historical action type, historical adjustment amplitude, historical adjustment speed, past awakening response, body movement changes and user individual sensitivity, and prioritizing actions with lower expected disturbance and potential for respiratory improvement among multiple candidate strategies.

7. The method as described in claim 1, characterized in that, In step S7, determining whether the adjustment improves the respiratory state, causes sleep disturbances, and whether the change can be attributed to the posture adjustment includes: setting a baseline window before adjustment and an observation window after adjustment; comparing changes in snoring intensity, snoring duration, respiratory cycle, respiratory interval statistical variance, EEG arousal risk, and body movement intensity within the two windows; and generating feedback confidence by combining time proximity rules, multi-source consistency rules, disturbance exclusion rules, signal credibility rules, and sleep stage stability rules; when the improvement of respiratory-related indicators occurs within the observation window after adjustment, is adjacent to the time of the adjustment action, is not accompanied by obvious spontaneous turning over or natural transition of sleep stages, and the signal quality meets preset conditions, the change is attributed to the posture adjustment and marked as high-confidence feedback; otherwise, it is marked as medium-confidence or low-confidence feedback.

8. The method as described in claim 7, characterized in that, The indicators being compared also include changes in blood oxygen levels.

9. The method as described in claim 1, characterized in that, In step S8, updating the individualized control parameters includes: using each nightly sleep session as the basic update cycle, or periodically updating based on data from several consecutive nights; recording the sleep state, respiratory risk level, body position, signal confidence, body movement state, intervention suitability, adjustment target, adjustment amplitude, adjustment speed, observation window, and changes in each indicator before and after adjustment, as well as feedback confidence, corresponding to each intervention event; updating at least one of the following control parameters using moving average, exponentially weighted average, linear regression, logistic regression, decision tree, Bayesian optimization, or reinforcement learning methods: minimum effective adjustment amplitude, head-of-bed elevation angle threshold, occipital or head-neck adjustment parameters, lateral tilt assistance priority, easily aroused movement type, movement threshold under different sleep stages, and observation window duration; wherein, samples with low-confidence feedback, accompanied by obvious spontaneous body movement, or natural transitions in sleep stages are not included in the learning update or have their weights reduced.

10. The method as described in claim 1, characterized in that, The method also includes safety protection and fault degradation steps: monitoring sensor health status, data reliability, actuator operating status, and communication status; when the EEG headband falls off, electrode contact remains abnormal, or the EEG signal quality is lower than a preset threshold, the method is downgraded to a coarse-grained state or the active posture adjustment based on fine sleep stages is suspended, the coarse-grained state includes stable sleep, unstable sleep, or suspected awakening; when pressure or body position signals are abnormal, the role of body position in the adjustment strategy is reduced, and tilting movements that rely on direction judgment are not performed; when the actuator detects jamming, overcurrent, overheating, or position deviation exceeding a preset range, the current action is stopped and the current position is maintained, the nearest safe posture is returned, or the preset safe posture is restored according to the safety strategy. A tiered degradation strategy is adopted, including maintaining the current posture and increasing the monitoring frequency, allowing only the minimum amplitude low-speed reversible fine adjustment, suspending all automatic mechanical adjustments, and triggering user terminal prompts or nursing assistance prompts.

11. A smart bed low-disturbance posture adjustment system based on sleep cycle sensing and snoring and breathing state recognition, used to implement the smart bed low-disturbance posture adjustment method based on sleep cycle sensing and snoring and breathing state recognition as described in any one of claims 1 to 10, characterized in that, include: The sleep stage recognition module is used to collect or receive the user's electroencephalogram (EEG) signals during sleep, and process the EEG signals to output sleep state, sleep stability or awakening risk, and sleep stage confidence. The snoring and breathing event sensing module is used to collect snoring signals, breathing rhythm signals, bed pressure signals, and bed micro-vibration signals, and output respiratory risk characteristics. The body position and pressure recognition module is used to determine the user's body position and the pressure distribution in different areas of the body; The sleep-breathing fusion judgment module is connected to the sleep stage recognition module, the snoring and breathing event perception module, and the body position and pressure recognition module, respectively, and is used to determine the respiratory risk level and intervention appropriateness based on sleep state, respiratory risk characteristics, body position, signal reliability and body movement state. The posture adjustment decision module, connected to the sleep-breathing fusion judgment module, is used to generate a low-disturbance posture adjustment strategy based on respiratory risk level, intervention suitability, body position and historical intervention effects. The strategy includes adjustment object, single adjustment amplitude, adjustment speed, action interval, observation window, cumulative adjustment limit and backtracking condition. The bed execution module, connected to the attitude adjustment decision module, includes at least one attitude adjustment unit for executing the low-disturbance attitude adjustment strategy and adopting a progressive execution method. The feedback evaluation module is connected to the sleep stage recognition module, the snoring and breathing event perception module, the body position and pressure recognition module and the bed execution module, respectively. It is used to evaluate the improvement of breathing status, sleep disturbance and change attribution based on the baseline window before adjustment and the observation window after adjustment after adjustment, and generate feedback confidence. The individualized learning module, connected to the feedback evaluation module and the attitude adjustment decision module, is used to update the individualized control parameters based on historical response data.

12. The intelligent bed low-disturbance posture adjustment system based on sleep cycle perception and snoring and breathing state recognition as described in claim 11, characterized in that, The snoring and breathing event sensing module is also used to collect blood oxygen signals.

13. The intelligent bed low-disturbance posture adjustment system based on sleep cycle perception and snoring and breathing state recognition as described in claim 11, characterized in that, The system also includes a safety protection and fault degradation module, which is connected to the sleep stage recognition module, snoring and breathing event perception module, body position and pressure recognition module, sleep-breathing fusion judgment module, posture adjustment decision module and bed execution module, respectively. It is used to monitor the health status of sensors, data reliability, actuator operation status and communication status, and to perform pause, rollback, degradation operation or prompt operation when an abnormality is detected.

14. A smart bed low-disturbance posture adjustment device based on sleep cycle sensing and snoring and breathing state recognition, used to implement the smart bed low-disturbance posture adjustment method based on sleep cycle sensing and snoring and breathing state recognition as described in any one of claims 1 to 10, characterized in that, include: Bed frame or mattress body; At least one posture adjustment unit is disposed on the bed frame or mattress body for adjusting at least one area of ​​the headboard, back, waist, legs, pillow, left and right tilt mechanism or mattress zone support area; The breathing state sensing unit is used to collect snoring signals, breathing rhythm signals, bed pressure signals, or bed micro-vibration signals. EEG signal acquisition or receiving unit, used to acquire EEG signals during the user's sleep process; The controller is connected to the posture adjustment unit, the breathing state sensing unit, and the EEG signal acquisition or receiving unit, respectively, and is configured to perform the following operations: The user's current sleep stage, sleep stability, or risk of awakening is identified based on the electroencephalogram (EEG) signals. The respiratory risk level is identified based on the signals collected by the respiratory state sensing unit. The appropriateness of intervention is determined based on sleep status, respiratory risk level, body position, signal reliability, and body movement status. A low-disturbance posture adjustment strategy is generated based on respiratory risk level, intervention suitability, body position, and historical intervention effects. The strategy includes adjustment target, single adjustment amplitude, adjustment speed, observation window, and retreat conditions. The attitude adjustment unit is controlled to execute the strategy in a progressive manner, and after adjustment, signals are collected for feedback evaluation based on the baseline window before adjustment and the observation window after adjustment. The subsequent adjustment strategy and individualized control parameters will be updated based on the feedback and evaluation results.

15. The intelligent bed low-disturbance posture adjustment device based on sleep cycle perception and snoring and breathing state recognition as described in claim 14, characterized in that, The attitude adjustment unit includes at least one of the following: an electric actuator, a linear actuator, a lead screw lifting mechanism, a linkage mechanism, an airbag array, a partitioned lifting plate, or a flexible support plate; the EEG signal acquisition or receiving unit is integrated into a flexible fabric headband, a flexible silicone headband, an elastic webbing, a periauricular wearing structure, or a occipital contact structure, and includes flexible dry electrodes, semi-dry electrodes, fabric electrodes, flexible conductive rubber electrodes, gel electrodes, or low-pressure contact electrodes; the device also includes a safety protection unit connected to the controller, used to monitor the actuator status and sensor health status, and to perform degraded operation in case of abnormality.

Images