Intelligent pillow sleep monitoring and individualized improvement device based on digital twinning

Abstract

The application provides an intelligent pillow sleep monitoring and individual improvement device based on digital twinning, relates to the field of medical health information technology, and quickly judges the reflection condition through hardwired logic and makes emergency intervention, simultaneously generates an individual intervention strategy through the establishment of a dynamic digital twin and executes the strategy, collects implicit physiological feedback data of a user after each intervention, calculates an intervention consensus score, executes differentiated parameter updating based on the intervention consensus score, and performs structure self-evolution when new feature data appears, removes and creates internal representation components, through the dual mechanisms of parameter updating and structure self-evolution, ensures that the digital twin can accurately depict the dynamic sleep portrait of the user at all times, thereby providing continuous and effective individual services that become more considerate, and fundamentally solves the technical problem of static model adaptability attenuation, improves the adaptability to the user, and guarantees good intervention effect.

Description

Technical Field

[0001] This invention relates to the field of medical and health information technology, specifically to a smart pillow sleep monitoring and personalized improvement device based on digital twins. Background Technology

[0002] With the rapid development of health IoT technology, smart bedding, especially smart pillows, has become an important carrier for sleep health monitoring and intervention. Existing technologies, such as the "Smart Pillow and Sleep Monitoring Method" disclosed in CN112704365B, monitor sleep quality through sensors and provide sleep aids when insomnia is detected; the "Smart Wake-up Pillow Based on Respiratory Rate Monitoring" disclosed in CN107669055A monitors respiratory rate through sensors and directly wakes the sleeper through vibration when the respiratory rate is abnormal; and the "Smart Pillow, Smart Pillow-Based Sleep Monitoring Method, System, and Storage Medium" disclosed in CN117643454A detects various parameters of the user's sleep state and sleeping posture through modules to generate a detection report. Existing smart pillows typically integrate multiple sensors to monitor physiological and behavioral data such as heart rate, respiration, snoring, and sleeping posture, and based on preset simple logic, such as raising the pillow when snoring is detected, performing a single intervention.

[0003] The following technical problems exist in the existing technology:

[0004] Problem 1: The decision-making models inside existing smart pillows are often static and fixed. The intervention effect is generally based on fixed logic, such as directly waking up the user when an abnormal breathing rate is detected. Although the intervention model has dynamic data monitoring and data optimization, it cannot optimize the structure of the model's own components. It cannot adjust and optimize itself according to the user's unique sleep habits and long-term data changes, resulting in decreased adaptability and gradual decline in intervention effect after long-term use.

[0005] The second problem is that existing intervention strategy generation mechanisms for sleep monitoring are mostly focused on solving single, immediate sleep problems such as stopping snoring and preventing breathing interruptions. They lack a comprehensive consideration of multiple objectives such as intervention effectiveness, sleep continuity, and long-term user preferences. Furthermore, they lack consideration for users' psychological feelings and privacy boundaries during intervention, resulting in stiff and abrupt intervention methods. In some cases, they may even cause new problems while solving one problem, such as waking users up during intervention, which reduces user acceptance and stickiness, and lowers the user experience. Summary of the Invention

[0006] To achieve the above objectives, the present invention provides the following technical solution: a smart pillow sleep monitoring and personalized improvement device based on digital twins, comprising a processor and a memory, wherein the execution method of the device includes:

[0007] The system acquires user physiological and behavioral data collected imperceptibly from the sensor array inside the smart pillow, and determines whether a conditioned reflex is triggered based on hard-wired logic. If a conditioned reflex is triggered, emergency intervention is performed; otherwise, a dynamic digital twin is constructed based on the user's physiological and behavioral data.

[0008] Based on the dynamic digital twin, the expected intervention effects of N candidate intervention strategies are simulated. Based on the expected intervention effects, multi-objective evaluation is performed to generate evaluation results. Based on the evaluation results, privacy intrusion index is set to fuse strategies and generate personalized intervention strategies.

[0009] Personalized intervention strategies are converted into control commands and sent to the smart pillow, which then works in conjunction with external environmental devices to execute the control commands.

[0010] Acquire implicit physiological feedback data from users after the execution of emergency intervention and control commands, calculate intervention consensus scores, and update parameters and perform structural self-evolution of the dynamic digital twin based on the intervention consensus scores.

[0011] Furthermore, the sensor array includes a micro-vibration sensor, a pressure sensor, a microphone, and a reflective photoelectric sensor; the user's physiological and behavioral data includes heart rate data, respiratory rate data, sleep posture data, body movement data, environmental data, snoring intensity, and blood oxygen saturation data.

[0012] Furthermore, the hardwired logic includes a first judgment condition and a second judgment condition;

[0013] The system acquires real-time respiratory rate data. If an interruption in the respiratory rate signal is detected and the interruption time exceeds a first preset threshold, the first judgment condition is triggered to perform emergency intervention.

[0014] The system acquires real-time snoring intensity. If the snoring intensity exceeds a second preset threshold and blood oxygen saturation continues to decline, a second judgment condition is triggered, and emergency intervention is initiated.

[0015] Furthermore, the step of simulating the expected intervention effects of N candidate intervention strategies based on a dynamic digital twin, and generating evaluation results through multi-objective assessment based on the expected intervention effects, includes:

[0016] N candidate intervention strategies are simulated based on user physiological and behavioral data. Each candidate intervention strategy is input into a dynamic digital twin to generate the expected intervention effect. The expected intervention effect includes the trajectory of changes in user physiological and behavioral data within a specified future time window.

[0017] The expected intervention effect is evaluated by multiple objectives. The comprehensive utility score of each candidate intervention strategy is calculated, and the evaluation result is generated based on the comprehensive utility score. The multi-objective evaluation includes effectiveness evaluation, continuity evaluation, and user preference evaluation.

[0018] Furthermore, the step of setting privacy intrusion indicators based on the evaluation results to perform strategy fusion and generate personalized intervention strategies includes:

[0019] Based on the evaluation results, a privacy intrusion index is set for each candidate intervention strategy. The privacy intrusion index is used to quantify the degree of intrusion of the candidate intervention strategy on the user.

[0020] Candidate intervention strategies are screened based on comprehensive utility scores, and those with comprehensive utility scores greater than or equal to a preset utility threshold are selected to form a candidate strategy set.

[0021] The candidate intervention strategy with the lowest privacy intrusion index in the candidate strategy set is selected as the basic strategy, and the basic strategy is judged based on the effectiveness evaluation.

[0022] If the effectiveness assessment of the basic strategy is greater than or equal to the effectiveness threshold, the basic strategy will be directly used as the personalized intervention strategy.

[0023] If the effectiveness assessment of the basic strategy is less than the effectiveness threshold, the candidate intervention strategy with the highest effectiveness assessment is selected from the candidate strategy set as the supplementary strategy. The basic strategy and the supplementary strategy are then merged according to the temporal logic to generate a distributed fusion instruction as a personalized intervention strategy.

[0024] Furthermore, the personalized intervention strategy is converted into control commands and sent to the smart pillow. The smart pillow, in conjunction with external environmental devices, executes the control commands, including:

[0025] Personalized intervention strategies are acquired and parsed to generate control instructions, which include a first type of instruction and a second type of instruction.

[0026] The first type of instruction is used to connect with the smart pillow and drive the smart pillow to execute personalized intervention strategies;

[0027] The second type of instruction is based on the connection between the smart pillow and external environmental devices, and selects and drives the external environmental devices to execute personalized intervention strategies based on the smart pillow.

[0028] Furthermore, the implicit physiological feedback data of the user includes changes in heart rate data, respiratory rate data, and frequency changes in body movement data within a preset time window after the execution of emergency intervention and control commands.

[0029] Furthermore, the acquisition of implicit physiological feedback data from users after the execution of emergency intervention and control commands, and the calculation of intervention consensus scores, include:

[0030] The implicit physiological feedback data of users is compared with the expected intervention effect to calculate the intervention consistency.

[0031] The sleep stages are analyzed based on the user's implicit physiological feedback data to generate sleep stage improvement data.

[0032] Intervention consensus score is calculated based on intervention fit and improvement in sleep stages.

[0033] Furthermore, the parameter update and structural self-evolution of the dynamic digital twin based on the intervention consensus score includes:

[0034] If the intervention consensus score is greater than or equal to the active learning threshold, the parameters of the dynamic digital twin are updated using the first learning rate.

[0035] If the intervention consensus score is less than the active learning threshold but greater than or equal to the conservative learning threshold, the second learning rate is used to update the parameters of the dynamic digital twin.

[0036] If the intervention consensus score is less than the conservative learning threshold, the third learning rate is used to update the parameters of the dynamic digital twin, and the first learning rate > the second learning rate > the third learning rate.

[0037] Furthermore, the parameter update and structural self-evolution of the dynamic digital twin based on the intervention consensus score includes:

[0038] Set a structural evolution threshold and an observation time window. If the average of several pre-consensus scores remains below the structural evolution threshold within the observation time window, then initiate the structural self-evolution of the dynamic digital twin.

[0039] The dynamic digital twin contains M representation components. The contribution of each representation component to the expected intervention effect is calculated, and a contribution threshold is set. Representation components with a contribution lower than the contribution threshold are removed.

[0040] Based on the analysis of user physiological and behavioral data, feature data that does not exist in the dynamic digital twin is extracted, new representation components are created based on the feature data, and the connection relationship with other representation components is initialized.

[0041] This invention provides a smart pillow sleep monitoring and personalized improvement device based on digital twins. It has the following beneficial effects:

[0042] 1. This invention establishes a four-linked method of reflex, cognition, execution, and learning for sleep monitoring. It employs a synergistic approach of rapid conditioned reflex and long-term optimization. Hard-wired logic enables rapid judgment of reflex conditions and prompts emergency intervention. Simultaneously, a dynamic digital twin is established to generate personalized intervention strategies for execution. After each intervention, implicit physiological feedback data from the user is collected to calculate an intervention consensus score. This score accurately reflects the user's acceptance of the intervention. Differentiated parameter updates are performed based on the intervention consensus score, and structural self-evolution occurs when new feature data emerges. Internal representation components are removed and recreated. Through this dual mechanism of parameter updates and structural self-evolution, the digital twin ensures that it can consistently and accurately depict the user's dynamically changing sleep profile, thereby providing continuously effective and increasingly personalized services. This fundamentally solves the technical challenge of static model adaptability decay, improves user compatibility, and guarantees excellent intervention results.

[0043] 2. This invention adopts a different approach from the traditional arbitrary control method that relies on detection and execution. It extrapolates the expected intervention effect from multiple candidate intervention strategies and then evaluates the effectiveness, sleep continuity, and user preferences of these strategies based on the expected intervention effect. A comprehensive utility score is calculated to generate the evaluation result. Simultaneously, a privacy invasiveness index is set for each candidate intervention strategy to quantify its degree of intrusion. Strategies with low invasiveness and high effectiveness are prioritized for personalized intervention. Only when the expected intervention effect is insufficient will low-invasiveness strategies be fused with high-effectiveness strategies in a time sequence to generate distributed fusion instructions for execution. This ensures that the intervention measures address physiological problems while respecting the user's sleep psychological space and physical boundaries to the greatest extent possible, avoiding abrupt awakenings or discomfort caused by harsh intervention. The final control instructions are the optimal solution derived after thorough simulation in a virtual environment and multi-dimensional consideration, not a unique solution. This approach provides minimally invasive and efficient intervention for abnormal sleep while fully protecting the user's sleep quality and experience, achieving a unity of intelligence and humanization, and improving user comfort and trust. Attached Figure Description

[0044] Figure 1 This is a flowchart illustrating the steps of the smart pillow sleep monitoring and personalized improvement device based on digital twins according to the present invention.

[0045] Figure 2 This is a flowchart of the smart pillow sleep monitoring and personalized improvement device based on digital twins according to the present invention. Detailed Implementation

[0046] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Example 1:

[0047] like Figures 1 to 2 As shown, the smart pillow sleep monitoring and personalized improvement device based on digital twins includes a processor and a memory. The execution method of the device includes:

[0048] S100: Acquires user physiological and behavioral data non-intrusibly collected by a sensor array inside the smart pillow. Based on hard-wired logic, it determines whether a conditioned reflex is triggered. If a conditioned reflex is triggered, emergency intervention is executed; otherwise, a dynamic digital twin is constructed based on the user's physiological and behavioral data. The sensor array includes micro-vibration sensors, pressure sensors, a microphone, and reflective photoelectric sensors. The sensor array achieves non-intrusive data acquisition through an embedded design. The sensor array is integrated inside the smart pillow, without direct contact with the user, and the acquisition process requires no active user participation, thus avoiding sleep disturbance. All sensors are connected via flexible circuitry and integrated into the pillow's filling. The outer cover uses breathable and sound-permeable fabric, ensuring user comfort while achieving continuous and non-intrusive collection of user physiological and behavioral data. User physiological and behavioral data includes heart rate data, respiratory rate data, sleep posture data, body movement data, environmental data, snoring intensity, and blood oxygen saturation data.

[0049] S101. When acquiring user physiological and behavioral data, data is collected through a sensor array, including:

[0050] First, the data collected by each sensor is synchronized in time, and all data streams are aligned using a unified timestamp to ensure data consistency;

[0051] Secondly, data preprocessing is performed, including bandpass filtering of data collected by micro-vibration sensors to separate heart rate and respiratory components, and calculation of heart rate and respiratory rate data; smoothing and outlier removal of data collected by pressure sensors to identify sleep posture and body movement data; noise reduction and feature extraction of data collected by microphones to extract environmental data and snoring intensity; and baseline correction and pulse wave extraction of data collected by reflective photoelectric sensors to derive blood oxygen saturation data.

[0052] Finally, the extracted data is integrated into a structured dataset, with all physiological and behavioral indicators at each time point. Inconsistent data are eliminated through weighted averaging or rule-based verification to generate unified user physiological and behavioral data for subsequent hard-wired logic judgments and the construction of dynamic digital twins.

[0053] Among them, heart rate data represents the frequency of heartbeats and is used to monitor cardiovascular status and sleep quality; respiratory rate data represents the respiratory rate and is used to assess respiratory health and sleep stability. It is based on the acquisition of vibration signals caused by heartbeats and respiratory movements by micro-vibration sensors, and extracts heart rate data and respiratory rate data from the vibration signals. The micro-vibration sensors are installed in the core area of ​​the smart pillow, usually located in the lower middle part of the pillow, close to the user's chest and back. They collect data by detecting weak vibrations caused by heartbeats and respiratory movements. The installation is wrapped with soft encapsulation material to reduce external noise interference.

[0054] Sleep posture data represents the user's sleeping position, such as supine or lateral, and is used to analyze sleep comfort and posture-related risks. Body movement data represents the frequency of body movement and is used to detect sleep interruptions and awakening events. Pressure distribution data of the user's head and neck is acquired based on a pressure sensor array. Based on the time series changes of the pressure distribution data, sleep posture data and body movement data are identified. The pressure sensor array is distributed in a grid on the surface layer of the smart pillow, covering the head and neck support area. It collects data by monitoring changes in pressure distribution. During installation, it is embedded in the elastic material of the smart pillow core to ensure even pressure distribution. At the same time, to improve the accuracy of the detection of the user's sleep posture data, a pressure sensor array can be further set up inside the mattress or the smart pillow can be directly connected to the smart mattress to monitor the specific position and posture of the user's limbs, such as the specific position and pressure of the arms and legs when the user is lying on their side, so as to achieve more accurate analysis of sleep posture data and body movement data.

[0055] Environmental data represents the ambient noise level, used to assess sleep environment disturbances; snoring intensity represents the loudness of snoring, used to identify the risk of snoring and sleep apnea. Environmental sound data is acquired based on microphones, and the average sound pressure level is calculated based on the environmental sound data to extract environmental data. The snoring intensity is obtained by extracting the amplitude of the characteristic frequencies of snoring. The microphone is installed on the side or inside of the pillow, facing the user's head, and directional sound pickup technology is used to collect environmental sounds and snoring. External interference is reduced by sound insulation materials.

[0056] Blood oxygen saturation data represents the oxygen content in the blood and is used to monitor hypoxic events. It is obtained by emitting light signals through a reflective photoelectric sensor and calculating the blood oxygen value from the reflected signals based on the principle of photoplethysmography. The reflective photoelectric sensor is installed on the side of the pillow near the neck, usually in the edge area of ​​the smart pillow. It collects data by emitting red or infrared light into the neck skin and receiving the reflected signals. Its installation position is optimized to ensure that the light signal can effectively penetrate the skin.

[0057] Data acquired from sensor arrays is integrated to generate user physiological and behavioral data. This data constructs a comprehensive view of the user's sleep state, providing data input for the generation of a dynamic digital twin.

[0058] In this embodiment, the installation position of the sensor array is only an exemplary conventional position description. In actual use, it can be set according to the actual usage situation. The installation position in this embodiment is only one possible way.

[0059] S102. Perform hard-wired logic judgment on the acquired user physiological and behavioral data to determine whether a conditional reflex is triggered. The hard-wired logic includes a first judgment condition and a second judgment condition.

[0060] The system acquires real-time respiratory rate data. If an interruption in the respiratory rate signal is detected and the interruption time exceeds the first preset threshold, it is determined to be a suspected sleep apnea event, triggering the first judgment condition and performing emergency intervention. Otherwise, the conditional reflex is not triggered, and a dynamic digital twin is constructed based on the user's physiological and behavioral data.

[0061] The system acquires real-time snoring intensity. If the snoring intensity exceeds the second preset threshold and the blood oxygen saturation data continues to decline, it is identified as a high-risk snoring event, triggering the second judgment condition and performing emergency intervention. Otherwise, the conditional reflex is not triggered, and a dynamic digital twin is constructed based on the user's physiological and behavioral data.

[0062] When a conditioned reflex is triggered, any ongoing decision-making process is immediately interrupted, and emergency intervention is executed directly. Emergency interventions are retrieved from a pre-set set of emergency intervention instructions. An emergency intervention is a set of pre-set, immediate operational instructions used to quickly alleviate emergency situations, prevent health deterioration, and ensure user safety when a conditioned reflex is triggered. For the first trigger condition, emergency interventions include adjusting the smart pillow height to open the airway, triggering a slight vibration of the smart pillow to wake the user, or activating external environmental devices such as increasing indoor oxygen concentration. For the second trigger condition, emergency interventions include changing the shape of the smart pillow to reduce neck pressure, playing soothing sounds to suppress snoring, or adjusting environmental devices such as lowering the room temperature. Specific emergency intervention content is selected from the pre-set set of intervention instructions to ensure rapid and targeted response.

[0063] When a conditioned reflex is triggered and emergency intervention is initiated, any ongoing processes, such as dynamic digital twin construction or strategy simulation, are immediately interrupted. The emergency intervention is executed directly, and other steps are paused and enter a waiting state. After the emergency intervention is completed, the normal process is resumed, but the user's status is reassessed first. The user's physiological and behavioral data are updated based on the sensor array, and the dynamic digital twin construction parameters are adjusted based on the updated user physiological and behavioral data to ensure that the emergency intervention is integrated with feedback.

[0064] A dynamic digital twin is a virtual model of a user, simulating the user's physiological and behavioral states, including real-time parameters such as heart rate, respiration, sleep posture, body movement, and environmental response, as well as historical sleep patterns and intervention response data. Based on real-time user physiological and behavioral data, it predicts responses to candidate intervention strategies, thereby generating personalized intervention strategies. First, the dynamic digital twin is initialized based on historical user physiological and behavioral data, and parameter relationships are established using statistical models or machine learning methods. Then, current user physiological and behavioral data is input in real-time, and the parameters of the dynamic digital twin are updated through data assimilation technology, enabling the dynamic digital twin to dynamically reflect changes in user physiological and behavioral data. Finally, a simulation engine is used to deduce the expected intervention effects under different candidate intervention strategies, completing the continuous optimization and validation of the dynamic digital twin.

[0065] Hardwired logic is a predefined, rule-based judgment logic that achieves rapid response based on threshold detection of simple conditions. It does not rely on complex calculations or model reasoning, and immediately identifies high-risk events in real-time monitoring and makes rapid judgments, such as sleep apnea or snoring, triggering emergency intervention to ensure user safety and avoid health risks caused by decision-making delays. Hardwired logic is used to handle emergencies involving user sleep safety, offering low latency and high reliability, and taking precedence over other adaptive processes. Hardwired logic is set based on safety specifications and is not limited to the judgment of the two reflection conditions mentioned above. For example, if it detects that the user's body movement data suddenly shows rapid shaking during sleep, and the duration and frequency of the shaking exceed the set threshold, it determines that the user may be experiencing convulsions and executes emergency intervention, such as connecting the smart pillow with smart speakers or other external environmental devices to perform alarm processing.

[0066] The first preset threshold is the time threshold for interruption of the respiratory rate signal, used to determine suspected sleep apnea events. It is generally set to 10 to 20 seconds, based on the clinical diagnostic criteria for sleep apnea. For example, an airflow interruption of ≥10 seconds is defined as sleep apnea. Emergency intervention is triggered when the respiratory interruption exceeds the first preset threshold. The second preset threshold is the sound pressure threshold for snoring intensity, used to determine high-risk snoring events in conjunction with a decrease in blood oxygen saturation. It is generally set to 50 to 70 decibels, with a sustained decrease in blood oxygen saturation exceeding 5%. It is set based on common loudness levels and blood oxygen decreases in snoring studies. Emergency intervention is triggered when the snoring intensity exceeds the second preset threshold and the blood oxygen saturation data decreases by more than 5% simultaneously. The first and second preset thresholds, as well as other preset thresholds of their hard-wired logic, are dynamically adjusted based on historical user data or clinical recommendations, but the initial values ​​are based on medical guidelines.

[0067] S200: Simulate the expected intervention effects of N candidate intervention strategies based on a dynamic digital twin, where N is an integer; perform multi-objective evaluation based on the expected intervention effects to generate evaluation results; set privacy intrusion index based on the evaluation results to perform strategy fusion and generate personalized intervention strategies.

[0068] S201. Simulate the expected intervention effects of N candidate intervention strategies based on a dynamic digital twin, and generate evaluation results based on the expected intervention effects through multi-objective assessment, including:

[0069] First, N candidate intervention strategies are simulated based on user physiological and behavioral data. Each candidate intervention strategy is then input into a dynamic digital twin to generate the expected intervention effect, which includes the trajectory of changes in user physiological and behavioral data within a specified future time window.

[0070] Then, a multi-objective evaluation of the expected intervention effect is conducted, calculating the comprehensive utility score of each candidate intervention strategy, and generating evaluation results based on the comprehensive utility score. The multi-objective evaluation includes effectiveness evaluation, continuity evaluation, and user preference evaluation.

[0071] The candidate intervention strategies are a set of predefined or dynamically generated sleep improvement measures stored in a unified predefined strategy library. These strategies optimize user sleep quality by adjusting the smart pillow or external environment. Candidate intervention strategies include physical interventions such as adjusting pillow height or firmness, sound interventions such as playing white noise or soothing music, environmental interventions such as adjusting room temperature or light, and combined interventions such as combining multiple measures, including simultaneous sound and light intervention. The most suitable candidate intervention strategy is selected by simulating the expected intervention effect of the candidate intervention strategies in a dynamic digital twin. The generation of N candidate intervention strategies is mainly based on historically monitored user physiological and behavioral data, real-time user physiological and behavioral data, the predefined strategy library, and user preference profiles. First, strategies related to real-time user physiological and behavioral data are selected from the predefined strategy library; for example, an anti-snoring strategy is selected based on the user's snoring history. Then, strategy parameters are adjusted according to real-time user physiological and behavioral data; for example, vigorous intervention is avoided during deep sleep. Finally, N candidate intervention strategies are generated to ensure diversity, covering different intervention types and intensities, for subsequent simulation in the dynamic digital twin.

[0072] The expected intervention effects include the trajectory of changes in the user's physiological and behavioral data within a specified future time window, such as heart rate stability, respiratory rate regularity, sleep posture adjustment, reduced body movement frequency, environmental adaptability, and improved blood oxygen saturation. By inputting N candidate intervention strategies into a dynamic digital twin, the user's response to the candidate intervention strategies is simulated and extrapolated. The extrapolation considers the user's current physiological state, historical response patterns, and changes in environmental parameters. The extrapolation includes:

[0073] First, the user's physiological and behavioral data obtained at the start of the simulation are set as the initial state parameters of the dynamic digital twin. By querying the user profile, historical interaction and feedback records similar to the current initial state parameters are obtained, and the historical interaction and feedback records are loaded into the dynamic digital twin as prior knowledge for the simulation.

[0074] Secondly, the current environmental data is used as the fixed boundary conditions for the extrapolation process, the candidate intervention strategies are transformed into specific quantifiable physical actions, and these physical actions and their time characteristics are input into the dynamic digital twin.

[0075] Then, the dynamic digital twin first calculates the direct impact of physical actions on the user's body. It then uses representation components to extrapolate the direct impact and initial state parameters to obtain the extrapolation results. The dynamic digital twin has M representation components, where M is an integer. It uses historical interaction and feedback records to weight and correct the extrapolation results. Throughout the extrapolation process, environmental data acts as a modulation factor. After weighting and correction, it finally generates a complete trajectory map of changes in the user's physiological and behavioral data within a specified time window in the future (such as 30 minutes in the next sleep cycle).

[0076] Finally, check whether there are logical contradictions between the predicted change trajectories of various data in the user's physiological and behavioral data in the complete change trajectory diagram. If there are contradictions, adjust the parameters of the representation component and conduct a new round of iterative deduction until an expected intervention effect without logical contradictions and with internal consistency is output. The expected intervention effect is used as time series data, including the predicted value of each type of data in the user's physiological and behavioral data, for multi-objective evaluation.

[0077] The multi-objective assessment includes effectiveness assessment, continuity assessment, and user preference assessment. Effectiveness assessment measures the efficacy of candidate intervention strategies in improving sleep by comparing the expected intervention effect with baseline data to calculate the percentage improvement of physiological indicators, such as the reduction rate of sleep apnea events or the increase in blood oxygen saturation. Continuity assessment ensures that the sleep process is not interrupted when candidate intervention strategies are implemented by analyzing body movement data and sleep stage transition frequency in the expected intervention effect to assess the number of awakenings and the degree of sleep fragmentation caused by the intervention. User preference assessment reflects the degree of matching between candidate intervention strategies and user habits by calculating preference compliance based on users' historical selection data, such as the past acceptance rate of sound interventions, and using weighted scoring to match the type of candidate intervention strategy with user profiles.

[0078] The overall utility score is a quantitative score that integrates the evaluation results generated by multi-objective assessments. It is used to rank candidate intervention strategies, including effectiveness score, continuity score, and preference score. The overall utility score provides a unified comparison benchmark for the selection of candidate intervention strategies. First, each evaluation dimension is standardized and scored, converting the expected intervention effect into a value within the range of [0,1]. For example, the effectiveness score is based on a linear mapping of the percentage improvement. Then, weights are set for each dimension, such as effectiveness weight 0.5, continuity weight 0.3, and preference weight 0.2, adjusted based on user priority. Finally, the overall utility score for each candidate intervention strategy is calculated using a weighted summation formula: Overall Utility Score = Effectiveness Score × w1 + Continuity Score × w2 + Preference Score × w3, where w1, w2, and w3 are weights. The overall utility score is then used for subsequent screening.

[0079] S202. Based on the evaluation results, set privacy intrusion indicators to perform strategy fusion and generate personalized intervention strategies, including:

[0080] First, based on the evaluation results, a privacy intrusion index is set for each candidate intervention strategy. The privacy intrusion index is used to quantify the degree of intrusion of the candidate intervention strategy on the user.

[0081] Then, candidate intervention strategies are screened based on the comprehensive utility score, and candidate intervention strategies with a comprehensive utility score greater than or equal to a preset utility threshold are selected to form a candidate strategy set;

[0082] Finally, the candidate intervention strategy with the lowest privacy intrusion index in the candidate strategy set is selected as the basic strategy, and the basic strategy is judged based on the effectiveness evaluation.

[0083] If the effectiveness assessment of the basic strategy is greater than or equal to the effectiveness threshold, the basic strategy will be directly used as the personalized intervention strategy.

[0084] If the effectiveness assessment of the basic strategy is less than the effectiveness threshold, the candidate intervention strategy with the highest effectiveness assessment is selected from the candidate strategy set as the supplementary strategy. The basic strategy and the supplementary strategy are then merged according to the temporal logic to generate a distributed fusion instruction as a personalized intervention strategy.

[0085] When integrating basic and supplementary strategies according to temporal logic, the process ensures coordinated and non-conflicting interventions based on the execution order, dependencies, and changes in the user's sleep stages. First, the intervention timing of basic and supplementary strategies is analyzed; for example, basic strategies are executed in the early stages of sleep to maintain comfort, while supplementary strategies are triggered when sleep problems are detected. Then, triggering conditions are set based on real-time physiological and behavioral data (such as sleep depth), and the instructions for candidate intervention strategies are serialized. For example, the basic strategy of slowly adjusting the pillow is executed first, followed by the triggering sound intervention of the supplementary strategy when breathing abnormalities occur. Finally, distributed fusion instructions are generated, defining the execution time and duration of each instruction to ensure a smooth transition.

[0086] The privacy intrusion index is a quantitative value that represents the degree to which candidate intervention strategies infringe upon user privacy and personal space, the scope of data collection, or the intensity of physical intervention. In selecting candidate intervention strategies, low-intrusion strategies are prioritized to improve user comfort and acceptance. The privacy intrusion index is set based on the strategy type, the scope of data use, and historical user feedback data. Candidate intervention strategies include physical intervention, audio intervention, environmental intervention, and conformity intervention; for example, physical intervention is more intrusive than environmental intervention. First, an intrusion level table is defined to classify candidate intervention strategies into low, medium, and high intrusion. For example, adjusting pillow height is low intrusion (assigned a value of 0.2), and collecting audio data is high intrusion (assigned a value of 0.8). Then, the assigned values ​​are adjusted based on the candidate intervention strategies and user profiles (such as user sensitivity to privacy). Finally, a privacy intrusion index is assigned to each candidate intervention strategy, quantified within the range [0,1], for use in the candidate intervention strategy fusion decision.

[0087] A user profile is a dynamically evolving digital model that comprehensively describes an individual user's characteristics and preferences. It is a data and knowledge base tightly coupled with and co-evolving with a dynamic digital twin, providing a basis for generating assessment results and personalized intervention strategies. The user profile includes static physiological characteristics, dynamic behavioral patterns, explicit and implicit preferences, and historical interaction and feedback records. Among these:

[0088] Static physiological characteristics include the user's relatively stable basic physiological data, including basic information such as age, gender, height, and weight; body characteristics such as neck curvature, shoulder width, and preferred sleeping position (such as the initial left / right side sleeping ratio); and known health conditions such as whether the user has a history of asthma or allergies.

[0089] Dynamic behavioral patterns are user sleep habits and patterns learned through long-term data, including sleep rhythms such as usual bedtime, wake-up time and their differences between weekdays and weekends; typical physiological baselines such as average heart rate and respiratory rate range in a resting state; and behavioral pattern databases such as the correlation strength between specific sleeping positions (such as supine) and snoring events, typical frequency and time distribution of nighttime body movement data, and sensitivity to environmental noise.

[0090] Explicit and implicit preferences refer to users' direct choices and indirect feedback on intervention strategies; including explicit preferences actively set by users through apps and other means, and implicit preferences inferred from implicit physiological feedback data; explicit preferences include a preference for white noise over natural rain sounds, a refusal to allow the pillow to physically adjust its shape during sleep, and setting a favorite room temperature range; implicit preferences include a user's implicit preference for this type of sound if playing white noise at a certain frequency results in a more significant improvement in heart rate variability and a decrease in the frequency of body movement data.

[0091] Historical interaction and feedback records are a collection of changes in users' physiological and behavioral data and effect records of past interventions. They record the type, parameters and execution time of all personalized intervention strategies that have been implemented in the past, as well as data such as the intervention consensus score and sleep stage improvement corresponding to each intervention, forming an experience base of "what strategies are effective or ineffective for users under what circumstances".

[0092] User profiles evolve synchronously with dynamic digital twins. As the dynamic digital twin updates its parameters and undergoes structural self-evolution based on the intervention consensus score, its new understanding of the user is also updated in the user profile, ensuring that the user profile reflects the user's latest physiological state and behavioral habits in real time. For example, user A's profile records: {Age: 35, Weight: 70kg, Known health problems: None, Typical sleep onset time: 23:30, Average heart rate: 65bpm, Probability of snoring while lying on the back: 85%, Preferred intervention: Air conditioning cooling > Pillow height adjustment, Most effective white noise type: Pink noise}. This information is not formed in one day, but is generated through initial design, continuous monitoring, and analysis of long-term data such as "When snoring while lying on the back, adjusting the pillow height is more effective than playing music in reducing snoring intensity without waking the user." Ultimately, the user profile enables the system to provide user A with a truly "tailor-made" sleep improvement plan.

[0093] The preset utility threshold is the minimum allowable value for the overall utility score, used to screen candidate intervention strategies. Based on historical data distribution and user satisfaction statistics, it ensures that the overall effect of candidate intervention strategies meets the target and avoids inefficient strategies from entering the candidate strategy set. The preset utility threshold is generally set to 0.7. The effectiveness threshold is the minimum value for effectiveness evaluation. Based on clinical sleep improvement standards or user baseline performance data, it is generally set to 0.6. The effectiveness threshold ensures that the strategy has a basic improvement effect and provides a basic guarantee for the integration of candidate intervention strategies.

[0094] Distributed fusion instructions are personalized intervention strategies that have been integrated, comprising multiple sub-instructions and distributed execution plans. Specifically, they include control instructions for the smart pillow (such as adjusting airbag pressure), control instructions for external environmental devices (such as adjusting air conditioning temperature), and timing logic instructions (such as delayed execution or conditional triggering). For example, distributed fusion instructions might be "adjust the pillow height to 5cm at time T1" and "play white noise for 10 minutes at time T2." During execution, the execution endpoints are distributed across different devices, and execution is carried out through unified scheduling.

[0095] Personalized intervention strategies are sleep improvement plans customized for users, including specific intervention actions, timing of execution, and parameter settings. For example, a personalized intervention strategy is to slowly adjust the height of the left side of the pillow to alleviate snoring when the user enters light sleep, while lowering the room temperature by 2 degrees Celsius and playing soothing ocean sounds when body movement is detected. Personalized intervention strategies are generated based on the expected intervention effects simulated by a dynamic digital twin, ensuring that they are tailored to the user's unique physiological characteristics and behavioral patterns.

[0096] S300: The personalized intervention strategy is converted into control commands and sent to the smart pillow. The smart pillow, in conjunction with external environmental devices, executes the control commands, including:

[0097] The system acquires and parses personalized intervention strategies to generate control commands. These strategies are broken down into P atomic operations; for example, "adjusting pillow height" is decomposed into motor control signals, and "reducing ambient brightness" into light control signals. Each atomic operation is mapped to the control protocols of the smart pillow and external environment devices. Command parameters (such as pillow height value and execution time) are defined using JSON format. Finally, the feasibility of the control commands and device status are verified. After ensuring no conflicts, the final control commands are generated and sent to the smart pillow and external environment devices. These control commands include both first-type and second-type commands.

[0098] The first type of instruction is used to connect with the smart pillow and drive the smart pillow to execute personalized intervention strategies; the second type of control instruction is an instruction for the smart pillow, including adjusting the pillow's physical parameters (such as height, firmness, and shape) and built-in functions (such as vibration reminders) to directly improve the user's sleeping posture and comfort, such as the instruction "drive the airbag to inflate to 50%" to raise the head and reduce snoring;

[0099] The second type of instruction is based on the connection between the smart pillow and external environmental devices. The smart pillow selects and drives the external environmental devices to execute personalized intervention strategies. The second type of control instruction is directed at the external environmental devices, including adjusting environmental factors such as temperature, humidity, light, and sound to create the best sleep environment. For example, the instruction "set the air conditioner temperature to 22℃" is executed in conjunction with the smart pillow. The two work together to ensure comprehensive sleep intervention.

[0100] External environmental devices include smart air conditioners, lighting systems, audio systems, humidifiers, etc. These devices help regulate the environment and aid sleep, such as lowering the room temperature to promote deep sleep. The smart pillow pairs with these devices via wireless communication protocols (such as Wi-Fi or Bluetooth) and sends control commands using standard APIs. For example, if the smart pillow detects that the user is too hot, it sends a temperature adjustment command to the air conditioner via Wi-Fi to achieve automatic adjustment.

[0101] S400: Acquire implicit physiological feedback data of users after the execution of emergency intervention and control commands, calculate intervention consensus score, and update parameters and perform structural self-evolution of dynamic digital twin based on intervention consensus score. The implicit physiological feedback data of users is indirect data reflecting changes in the user's physiological state that is collected imperceptibly by a sensor array within a preset time window after the execution of emergency intervention and control commands. The implicit physiological feedback data of users includes the frequency changes of heart rate data, respiratory rate data, and body movement data. It does not require active reporting by the user and is used to calculate intervention consensus score, thereby verifying the effectiveness of intervention strategy and driving the optimization of digital twin.

[0102] S401. Obtain implicit physiological feedback data from users after the execution of emergency intervention and control instructions, and calculate the intervention consensus score, including:

[0103] First, the implicit physiological feedback data of users is compared with the expected intervention effect to calculate the intervention consistency. The intervention consistency is an indicator that quantifies the degree of matching between the actual changes in users' physiological and behavioral data and the expected intervention effect predicted by the dynamic digital twin. It is used to evaluate the prediction accuracy of the dynamic digital twin and provide a basis for updating the dynamic digital twin. The predicted change trajectory of physiological parameters is extracted from the expected intervention effect and time-aligned with the actual change trajectory in the implicit physiological feedback data of users. The absolute error between the predicted value and the actual value at each time point is calculated, and the average error rate is obtained. Finally, the result is normalized to a score in the range of [0,1] using the formula: Intervention consistency = 1 - Average error rate. The higher the score, the more accurate the prediction.

[0104] Then, sleep stages are analyzed based on users' implicit physiological feedback data to generate sleep stage improvement status. Sleep stage improvement status includes an increase in the proportion of deep sleep, a decrease in the proportion of light sleep, a decrease in the number of awakenings, and an improvement in sleep efficiency, where sleep efficiency = total sleep time / time in bed, used to comprehensively evaluate the overall impact of personalized intervention strategies on sleep quality. First, sleep stages are identified using a rule model based on users' implicit physiological feedback data, including wakefulness, light sleep, and deep sleep. Then, the distribution of sleep stages within the same duration before and after the intervention is compared, and the change in each feature data of users' implicit physiological feedback data is calculated. Finally, an improvement status report is generated as input for calculating the intervention consensus score.

[0105] Finally, an intervention consensus score is calculated based on the intervention consistency and sleep stage improvement. This score is a comprehensive score reflecting the consistency between the actual and expected intervention effects of the personalized intervention strategy and the degree of improvement in sleep quality. It includes the intervention consistency score and the sleep stage improvement score, guiding the parameter updates and structural evolution of the dynamic digital twin. First, the intervention consistency and sleep stage improvement are standardized to scores in the range [0,1]. Then, weights are set, such as 0.6 for intervention consistency and 0.4 for improvement. These weights are adjusted based on the optimization goal, which is pre-set and prioritizes either improving the user's sleep or the accuracy of the dynamic digital twin's expected intervention effect. Finally, a weighted summation formula is used: Intervention Consensus Score = Intervention Consensus Score × z1 + Sleep Stage Improvement Score × z2, where z1 and z2 are weights. The total output score is the intervention consensus score, used for judging the parameter updates and structural evolution of the dynamic digital twin.

[0106] S402. Based on the intervention consensus score, perform parameter updates and structural self-evolution on the dynamic digital twin. Parameter updates include:

[0107] If the intervention consensus score is greater than or equal to the active learning threshold, the parameters of the dynamic digital twin are updated using the first learning rate.

[0108] If the intervention consensus score is less than the active learning threshold but greater than or equal to the conservative learning threshold, the second learning rate is used to update the parameters of the dynamic digital twin.

[0109] If the intervention consensus score is less than the conservative learning threshold, the third learning rate is used to update the parameters of the dynamic digital twin, and the first learning rate > the second learning rate > the third learning rate.

[0110] Among them, the active learning threshold is the high threshold of the intervention consensus score, and the conservative learning threshold is the low threshold. The intervention consensus score is divided into three intervals by the active learning threshold and the conservative learning threshold, which correspond to different update intensities of the dynamic digital twin: above the active learning threshold, the intervention effect is good and can be learned quickly; below the conservative learning threshold, the effect is poor and needs to be updated cautiously. The active learning threshold and the conservative learning threshold are set according to the statistical distribution of historical intervention data and the model stability requirements. Generally, the active learning threshold is set to 0.7-0.8 and the conservative learning threshold is set to 0.5-0.6 to ensure that the learning rate adjustment is both responsive and avoids overfitting.

[0111] The first learning rate is a relatively high update rate, used when the consensus score is above the active learning threshold, indicating a highly successful intervention with a high degree of consistency between prediction and feedback. Active learning can then be implemented to strengthen the dynamic digital twin's rapid absorption of effective experience. The second learning rate is a medium rate, used when the consensus score is between the two thresholds, indicating a moderate effect with partially accurate predictions. Robust learning is needed to fine-tune the dynamic digital twin and balance learning. The third learning rate is a relatively low rate, used when the consensus score is below the conservative learning threshold, indicating inaccurate predictions. Conservative learning should be implemented to avoid introducing noise and prevent... To prevent the degradation of dynamic digital twins, the adaptive capability of dynamic digital twins is optimized through differentiated update intensity. A high learning rate accelerates optimization when the effect is clear, a medium learning rate maintains stability when the effect is uncertain, and a low learning rate prevents the dynamic digital twin from deviating when the effect is poor, thereby achieving safe and efficient adaptive updates. The first, second, and third learning rates are all set based on the convergence of the dynamic digital twin and historical experimental data. Generally, the first learning rate is 0.1-0.2, the second learning rate is 0.05-0.1, and the third learning rate is 0.01-0.05.

[0112] S403. Based on the intervention consensus score, update the parameters and perform structural self-evolution on the dynamic digital twin. The structural self-evolution includes:

[0113] First, set a structural evolution threshold and an observation time window. If the average of several pre-consensus scores remains below the structural evolution threshold within the observation time window, then initiate the structural self-evolution of the dynamic digital twin.

[0114] Then, the contribution of each representation component in the dynamic digital twin to the expected intervention effect is calculated, and a contribution threshold is set to remove representation components whose contribution is lower than the contribution threshold.

[0115] Finally, based on the analysis of user physiological and behavioral data, feature data that does not exist in the dynamic digital twin is extracted, new representation components are created based on the feature data, and the connection relationship with other representation components is initialized.

[0116] The structural evolution threshold is the lower limit of the average intervention consensus score. It is used to determine whether the structure of the dynamic digital twin needs to be reconstructed and to trigger the structural self-evolution process. When the intervention consensus score is consistently lower than the structural evolution threshold, it indicates that the current structure of the dynamic digital twin is insufficient to accurately simulate the user and needs to be optimized. The structural evolution threshold is set based on the risk of model performance degradation and long-term effect data, and is generally set to 0.4-0.6 to ensure timely response to structural problems. The observation time window is the time range for evaluating the average score to avoid single fluctuations from triggering evolution and to ensure that structural changes are based on long-term trends. It is set based on the stability of the sleep cycle and the frequency of changes in user behavior patterns, and is generally set to 5-10 days to balance response speed and reliability.

[0117] Within the observation window, the average intervention consensus score is obtained by summing the daily intervention consensus scores and dividing by the number of days. If the average intervention consensus score is lower than the structural evolution threshold, it indicates that the current structure of the dynamic digital twin cannot accurately predict and adapt to users in the long term. The dynamic digital twin has missing key features or redundant representation components. Initiating structural self-evolution is to fundamentally optimize the architecture of the dynamic digital twin by removing low-contribution representation components and adding new representation components to improve prediction accuracy and personalization capabilities.

[0118] Representational components are functional units within a dynamic digital twin, used to simulate specific physiological or behavioral characteristics of a user. These include parametric models (such as heart rate response models), feature extractors (such as sleep stage classifiers), and relational mappers (such as intervention effect association networks). M representational components collectively construct a virtual representation of the user, supporting the extrapolation and prediction of intervention effects. When calculating contribution, firstly, the output value of each representational component is recorded during the dynamic digital twin extrapolation process. Then, correlation analysis or variance contribution ratio method is used to calculate the correlation strength between the representational component output and the final expected intervention effect. Finally, the correlation strength is normalized to a contribution score of [0,1], where a higher score indicates a greater influence of the representational component on the prediction result.

[0119] The contribution threshold is the minimum allowable value (e.g., 0.1) for the contribution of a representation component. It is used to filter out low-contribution or redundant representation components for removal. It is set according to the historical performance distribution of representation components and the simplification requirements of the dynamic digital twin, and is generally set to 0.1-0.2 to ensure that core representation components are retained while optimizing the structure. Low-contribution representation components are prone to introducing noise or computational redundancy, reducing model efficiency and accuracy. By removing low-contribution representation components, the structure of the dynamic digital twin can be simplified, the risk of overfitting can be reduced, and resources can be focused on key features, thereby improving the overall simulation performance.

[0120] When extracting feature data not present in a dynamic digital twin, the process begins by analyzing user physiological and behavioral data to identify patterns not modeled in the existing dynamic digital twin. Next, cluster analysis or principal component analysis (PCA) is used to extract feature data. Based on this feature data, the functionality of the representation component is defined. Then, the internal parameters of the representation component are initialized, existing representation components related to the new component are identified, input-output links are established, and the connection validity is verified using test data. This completes the creation of the new representation component. For example, identifying a new unmodeled pattern in the dynamic digital twin as a correlation between body movement and blood oxygenation, constructing a body movement-blood oxygenation correlation model, and establishing an input-output link between this model and the existing blood oxygen saturation representation component. The output of the new representation component is used as the input to the blood oxygen saturation representation component, and the connection validity is verified, thus completing the creation of the new body movement-blood oxygenation representation component.

[0121] In this embodiment, a four-linked approach of reflex, cognition, execution, and learning is established. Sleep monitoring employs a synergistic approach of rapid conditioned reflex and long-term optimization. Hard-wired logic is used to quickly determine reflex conditions and make emergency interventions. Simultaneously, a personalized intervention strategy is generated and executed by establishing a dynamic digital twin. After each intervention, implicit physiological feedback data of the user is collected to calculate the intervention consensus score. The intervention consensus score truly reflects the user's body's acceptance of the intervention. Differentiated parameter updates are performed based on the intervention consensus score, and structural self-evolution is performed when new feature data appears. Internal representation components are removed and created. Through the dual mechanism of parameter updates and structural self-evolution, it is ensured that the digital twin can always accurately depict the user's dynamically changing sleep profile, thereby providing continuously effective and increasingly personalized services. This fundamentally solves the technical problem of static model adaptability decay, improves adaptability to users, and ensures good intervention effects.

[0122] This approach departs from the traditional arbitrary control method of execution after detection. It extrapolates the expected intervention effects from multiple candidate intervention strategies and then conducts multi-objective evaluations, including effectiveness assessment, sleep continuity assessment, and user preference assessment, based on these expected effects. A comprehensive utility score is calculated to generate the evaluation results. Simultaneously, a privacy invasiveness index is set for each candidate intervention strategy to quantify its degree of intrusion. Strategies with low invasiveness and high effectiveness are prioritized for personalized intervention. Only when the expected intervention effect is insufficient will low-invasiveness strategies be fused with high-effectiveness strategies in a time sequence to generate distributed fusion commands for execution. This ensures that intervention measures address physiological problems while maximally respecting the user's sleep psychological space and physical boundaries, avoiding abrupt awakenings or discomfort caused by harsh interventions. The final generated control commands are the optimal solution derived after thorough simulation in a virtual environment and multi-dimensional considerations, not a unique solution. This approach provides minimally invasive and efficient intervention for abnormal sleep while fully protecting user sleep quality and experience, achieving a unity of intelligence and humanization, and improving user comfort and trust. Example 2:

[0123] This application also provides an electronic device. The electronic device may include one or more processors and one or more memories. The memories store computer-readable code, which, when executed by the one or more processors, can perform the digital twin-based smart pillow sleep monitoring and personalized improvement device as described above.

[0124] The method described in this application can also be implemented using the architecture of the electronic device shown in this application. The electronic device may include a bus, one or more CPUs, ROM, RAM, a communication port connected to a network, input / output, a hard disk, etc. The storage device in the electronic device, such as a ROM or hard disk, may store the digital twin-based smart pillow sleep monitoring and personalized improvement device provided in this application. Furthermore, the electronic device may also include a user interface. Of course, the architecture shown in this application is merely exemplary; when implementing different devices, one or more components of the electronic device shown in this application may be omitted according to actual needs.

[0125] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising a reference structure" does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.

[0126] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A smart pillow sleep monitoring and personalized improvement device based on digital twins, comprising a processor and a memory, characterized in that, The execution method of the device includes: The system acquires user physiological and behavioral data collected imperceptibly from the sensor array inside the smart pillow, and determines whether a conditioned reflex is triggered based on hard-wired logic. If a conditioned reflex is triggered, emergency intervention is performed; otherwise, a dynamic digital twin is constructed based on the user's physiological and behavioral data. The expected intervention effects of N candidate intervention strategies are simulated using a dynamic digital twin. A privacy intrusion index is set for each candidate intervention strategy. A multi-objective evaluation is performed based on the expected intervention effects. The multi-objective evaluation includes effectiveness evaluation, continuity evaluation, and user preference evaluation. The comprehensive utility score of each candidate intervention strategy is calculated. An evaluation result is generated based on the comprehensive utility score. Candidate intervention strategies are screened based on the comprehensive utility score. Candidate intervention strategies with a comprehensive utility score greater than or equal to a preset utility threshold are selected to form a candidate strategy set. The candidate intervention strategy with the lowest privacy intrusion index in the candidate strategy set is selected as the basic strategy. The basic strategy is judged based on the effectiveness evaluation. If the effectiveness evaluation of the basic strategy is greater than or equal to the effectiveness threshold, the basic strategy is directly used as a personalized intervention strategy. If the effectiveness evaluation of the basic strategy is less than the effectiveness threshold, the candidate intervention strategy with the highest effectiveness evaluation is selected from the candidate strategy set as a supplementary strategy. The basic strategy and the supplementary strategy are merged according to temporal logic to generate a distributed fusion instruction as a personalized intervention strategy. Personalized intervention strategies are converted into control commands and sent to the smart pillow, which then works in conjunction with external environmental devices to execute the control commands. Acquire implicit physiological feedback data from users after the execution of emergency intervention and control commands, calculate intervention consensus scores, and update parameters and perform structural self-evolution of the dynamic digital twin based on the intervention consensus scores.

2. The smart pillow sleep monitoring and personalized improvement device based on digital twins according to claim 1, characterized in that, The sensor array includes a micro-vibration sensor, a pressure sensor, a microphone, and a reflective photoelectric sensor; the user's physiological and behavioral data includes heart rate data, respiratory rate data, sleep posture data, body movement data, environmental data, snoring intensity, and blood oxygen saturation data.

3. The smart pillow sleep monitoring and personalized improvement device based on digital twin as described in claim 2, characterized in that, The hardwired logic includes a first judgment condition and a second judgment condition; The system acquires real-time respiratory rate data. If an interruption in the respiratory rate signal is detected and the interruption time exceeds a first preset threshold, the first judgment condition is triggered to perform emergency intervention. The system acquires real-time snoring intensity. If the snoring intensity exceeds a second preset threshold and blood oxygen saturation continues to decline, a second judgment condition is triggered, and emergency intervention is initiated.

4. The smart pillow sleep monitoring and personalized improvement device based on digital twin as described in claim 1, characterized in that, The method simulates the expected intervention effects of N candidate intervention strategies based on dynamic digital twins, sets a privacy intrusion index for each candidate intervention strategy, and performs multi-objective evaluation based on the expected intervention effects, including: N candidate intervention strategies are simulated based on user physiological and behavioral data. Each candidate intervention strategy is input into a dynamic digital twin to generate the expected intervention effect. The expected intervention effect includes the trajectory of changes in user physiological and behavioral data within a specified future time window. The expected intervention effect is evaluated by multiple objectives. The comprehensive utility score of each candidate intervention strategy is calculated, and the evaluation result is generated based on the comprehensive utility score. The multi-objective evaluation includes effectiveness evaluation, continuity evaluation, and user preference evaluation.

5. The smart pillow sleep monitoring and personalized improvement device based on digital twin as described in claim 1, characterized in that, The process of converting personalized intervention strategies into control commands and sending them to the smart pillow, with the smart pillow working in conjunction with external environmental devices to execute the control commands, includes: Personalized intervention strategies are acquired and parsed to generate control instructions, which include a first type of instruction and a second type of instruction. The first type of instruction is used to connect with the smart pillow and drive the smart pillow to execute personalized intervention strategies; The second type of instruction is based on the connection between the smart pillow and external environmental devices, and selects and drives the external environmental devices to execute personalized intervention strategies based on the smart pillow.

6. The smart pillow sleep monitoring and personalized improvement device based on digital twin according to claim 1, characterized in that, The implicit physiological feedback data from the user includes changes in heart rate, respiratory rate, and frequency of body movement data within a preset time window after the execution of emergency intervention and control commands.

7. The smart pillow sleep monitoring and personalized improvement device based on digital twin according to claim 1, characterized in that, The process of acquiring implicit physiological feedback data from users after the execution of emergency intervention and control commands, and calculating the intervention consensus score, includes: The implicit physiological feedback data of users is compared with the expected intervention effect to calculate the intervention consistency. The sleep stages are analyzed based on the user's implicit physiological feedback data to generate sleep stage improvement data. Intervention consensus score is calculated based on intervention fit and improvement in sleep stages.

8. The smart pillow sleep monitoring and personalized improvement device based on digital twin according to claim 7, characterized in that, The parameter update and structural self-evolution of the dynamic digital twin based on the intervention consensus score includes: If the intervention consensus score is greater than or equal to the active learning threshold, the parameters of the dynamic digital twin are updated using the first learning rate. If the intervention consensus score is less than the active learning threshold but greater than or equal to the conservative learning threshold, the second learning rate is used to update the parameters of the dynamic digital twin. If the intervention consensus score is less than the conservative learning threshold, the third learning rate is used to update the parameters of the dynamic digital twin, and the first learning rate > the second learning rate > the third learning rate.

9. The smart pillow sleep monitoring and personalized improvement device based on digital twin as described in claim 8, characterized in that, The parameter update and structural self-evolution of the dynamic digital twin based on the intervention consensus score, wherein the structural self-evolution includes: Set a structural evolution threshold and an observation time window. If the average of several pre-consensus scores remains below the structural evolution threshold within the observation time window, then initiate the structural self-evolution of the dynamic digital twin. The dynamic digital twin contains M representation components. The contribution of each representation component to the expected intervention effect is calculated, and a contribution threshold is set. Representation components with a contribution lower than the contribution threshold are removed. Based on the analysis of user physiological and behavioral data, feature data that does not exist in the dynamic digital twin is extracted, new representation components are created based on the feature data, and the connection relationship with other representation components is initialized.

Images