Mattress negative oxygen ion concentration control system and method based on sensor data analysis

Abstract

The invention discloses a mattress negative oxygen ion concentration control system and method based on sensor data analysis, and particularly relates to the technical field of mattress negative oxygen ion control, a multi-modal feature extraction module collects and analyzes a user physiological signal in real time, and generates a high-order feature vector representing the state of the user physiological signal; the personalized strategy generation module generates a regulation and control strategy of target concentration, action area and duration by using a deep reinforcement learning algorithm based on the features, the user portrait and historical data; the intelligent ion field construction module controls working parameters of the mattress partition negative ion generator according to a strategy, and constructs and maintains a gradient concentration field; the closed-loop optimization module monitors the physiological response of the user to evaluate the effect; according to the method, closed-loop self-adaptive regulation and control from physiological perception, intelligent decision-making to accurate execution are realized, the problems that a regulation and control mode in the prior art is static and lacks individuation and system closed loop are solved, and the accuracy, effectiveness and user experience of sleep health intervention are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of mattress negative ion control technology, and more specifically, to a mattress negative ion concentration control system and method based on sensor data analysis. Background Technology

[0002] With the popularization of healthy sleep concepts and the development of environmental control technologies, negative oxygen ions are receiving increasing attention due to their potential role in improving air quality, promoting relaxation, and enhancing sleep quality. Existing technologies include products that integrate negative oxygen ion generators into mattresses, attempting to create a beneficial sleep microenvironment by continuously or periodically releasing negative oxygen ions. However, most of these products use fixed programs or simple manual modes for control, with preset and uniform ion release concentration, time, and spatial distribution. They cannot dynamically adjust based on the user's real-time physiological state, sleep stage, or individual differences. This "one-size-fits-all" approach struggles to match the actual needs of different users or the same user at different sleep stages, resulting in limited control effects, low personalization, and even potential discomfort due to inappropriate concentrations. Therefore, it fails to achieve truly user-centered, precise health intervention.

[0003] On the other hand, the smart mattress industry has made significant progress in sleep monitoring and data analysis by integrating various physiological sensors (such as heart rate, respiration, and body movement monitoring). Existing solutions can accurately identify sleep stages, statistically analyze sleep parameters, and provide users with sleep reports. However, most of these systems remain at the "monitoring" and "reporting" level, forming an open loop—that is, the system can sense the state but lacks the ability to execute real-time, proactive intervention based on the perceived results. How to deeply integrate advanced physiological state sensing technology with dynamic environmental control technology to form a closed loop of "perception-analysis-decision-execution," enabling the mattress not only to understand the user's sleep but also to proactively create an optimized sleep environment, is a key bottleneck and challenge in current technological development.

[0004] In summary, existing technologies have significant shortcomings: First, the negative ion release control strategy is static and fixed, lacking the ability to dynamically adjust and personalize based on real-time physiological feedback from users. Second, the monitoring function and environmental regulation function of smart mattresses are disconnected, failing to form a closed-loop optimization system. Third, there is a lack of a systematic solution that can comprehensively consider multimodal physiological signals, user historical data, and environmental context, and use intelligent algorithms to generate and execute a refined spatial concentration field. Therefore, there is an urgent need for an integrated system and method that can sense the user's physiological state in real time, make intelligent decisions, and accurately execute negative ion concentration regulation to achieve truly personalized, adaptive, and efficient sleep health environment management. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a mattress negative oxygen ion concentration control system and method based on sensor data analysis.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] The mattress negative ion concentration control system based on sensor data analysis includes the following modules:

[0008] The multimodal feature extraction module is used to collect and preprocess raw physiological signals from multiple physiological sensors integrated in the mattress, and extract high-order feature vectors representing the user's real-time physiological state through a deep convolutional neural network.

[0009] The personalized strategy generation module is used to receive the high-order feature vector and, in combination with the user's historical data and profile, use a deep reinforcement learning algorithm to generate personalized negative oxygen ion regulation strategy instructions. The regulation strategy instructions include at least the target concentration, the area of ​​action, and the regulation duration.

[0010] The intelligent ion field construction module is used to control the working parameters of the negative oxygen ion generators in multiple independent zones on the mattress surface according to the control strategy instructions, so as to construct and maintain a negative oxygen ion concentration field that meets the gradient requirements on the mattress surface.

[0011] The closed-loop optimization module is used to monitor the user's physiological response to negative oxygen ion regulation, evaluate the regulation effect and calculate the deviation from the expected target, and optimize the algorithm model parameters of the personalized strategy generation module online based on the deviation.

[0012] Specifically, the multimodal feature extraction module is used for:

[0013] It receives multi-channel physiological signals from mattress sensors, including electrocardiogram signals, electromyography signals, skin conductance response signals, and respiratory waveform signals;

[0014] The raw physiological signals are preprocessed to remove interference, correct drift, and normalize.

[0015] The preprocessed signal is input into a five-layer deep convolutional neural network to extract time-frequency domain features;

[0016] The extracted features are weighted by a channel attention mechanism to output a high-order feature vector that represents the user's real-time physiological and sleep state.

[0017] Specifically, the personalized strategy generation module is used for:

[0018] Receive high-order feature vectors and retrieve user historical data and profiles;

[0019] Match current features with historical patterns to identify physiological state trends;

[0020] Based on deep reinforcement learning algorithms and combined with a reward function that dynamically adjusts weights, personalized regulation strategy instructions containing the target negative oxygen ion concentration, the area of ​​action, and the duration of regulation are calculated and generated.

[0021] Specifically, the intelligent ion field construction module is used for:

[0022] According to the control strategy instructions, the mattress is divided into multiple independent control zones;

[0023] Control the independent negative oxygen ion generators in each area and adjust the ion generation rate through pulse width modulation signals;

[0024] Based on real-time ambient temperature and humidity data, the pulse frequency and duty cycle are adaptively adjusted, and an interleaved switching control strategy is adopted to form and maintain a stable negative oxygen ion concentration field on the mattress surface that meets the gradient requirements.

[0025] Specifically, the closed-loop optimization module is used for:

[0026] Real-time monitoring of users' physiological response signals during the process of receiving negative oxygen ion regulation;

[0027] The relaxation improvement rate, sleep quality index, and physiological coordination score are calculated based on the physiological response signals to evaluate the actual regulation effect.

[0028] Compare and analyze the actual control effects with the expected effects to identify control deviations;

[0029] Based on the bias, the model parameters in the personalized strategy generation module are updated using an online learning algorithm.

[0030] Specifically, the multi-channel physiological signals are acquired at different sampling rates using the following sensors:

[0031] A bioelectric amplifier that acquires electrocardiogram and electromyography signals at a sampling rate of 1000 Hz, a digital capacitive sensor that acquires skin conductance response signals at a sampling rate of 100 Hz, and a piezoelectric thin film sensor array that monitors respiratory waveforms at a sampling rate of 50 Hz.

[0032] Specifically, the weighting coefficients of the reward function are:

[0033] The weight coefficients are periodically updated by dynamically adjusting a meta-learning module based on subjective sleep scores and environmental context data provided by users at regular intervals. This is achieved by minimizing the prediction score error and introducing an environmental prior regularization term.

[0034] Specifically, the intelligent ion field construction module also includes:

[0035] Feedback closed-loop calibration uses a network of ion concentration sensors distributed across various regions to monitor the concentration distribution in real time. Based on the deviation between the monitored data and the target concentration, the operating parameters of the negative oxygen ion generators in each region are dynamically adjusted to achieve accurate tracking of the concentration field.

[0036] The method for controlling the concentration of negative oxygen ions in a mattress based on sensor data analysis includes the following steps:

[0037] S1: Physiological signals are collected from the mattress multimodal sensor and high-order feature vectors are extracted through a deep network;

[0038] S2: Based on the feature vector, a personalized negative oxygen ion regulation strategy is generated using a deep reinforcement learning model;

[0039] S3: According to the aforementioned control strategy, the negative oxygen ion generator is controlled in the mattress zones to construct a target concentration field;

[0040] S4: Monitor the user's physiological response, evaluate the effect of regulation, and optimize the deep reinforcement learning model based on the bias.

[0041] The technical effects and advantages of this invention are as follows:

[0042] By constructing a closed-loop control system encompassing multimodal perception, intelligent decision-making, and precise execution, a fundamental shift from a static, uniform control model to a dynamic, personalized one has been achieved. Deep neural networks are used to fuse and analyze multidimensional physiological signals from the user, including electrocardiogram (ECG), electromyography (EMG), electrodermal activity (EDA), and respiration, accurately identifying their real-time state and sleep stage. Then, through deep reinforcement learning algorithms, combined with user profiles and historical data, a personalized control strategy is generated, including target concentration, target area, and duration of action. Finally, a zoned, controllable negative ion generation matrix and adaptive pulse modulation technology are used to precisely construct and maintain a gradient ion concentration field on the mattress surface that meets the required specifications. This complete closed loop enables environmental interventions to truly understand and adapt to user needs, greatly improving the targeting and effectiveness of health interventions.

[0043] This invention possesses continuous self-optimization and long-term adaptability, ensuring continuous improvement in regulatory effects and user experience. Through a closed-loop optimization module, it monitors user physiological responses in real time, quantitatively assesses regulatory effects, and uses deviation data to update decision model parameters online, driving the system to continuously learn and evolve. Its core meta-learning mechanism dynamically adjusts the weights of optimization objectives based on user subjective feedback, seasonal changes, and environmental quality, enabling the system to adapt not only to individual differences but also to long-term changes in the user's lifestyle and environment. This not only enhances the system's robustness and intelligence but also ensures that it can continuously provide users with a safe, comfortable, and efficient healthy sleep environment during long-term use. Attached Figure Description

[0044] Figure 1 This is a system block diagram of the present invention. Detailed Implementation

[0045] 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.

[0046] like Figure 1 As shown, the mattress negative ion concentration control system module based on sensor data analysis is as follows:

[0047] The multimodal feature extraction module is responsible for collecting raw physiological signals (such as ECG, EMG, electrodermal activity, and respiratory waves) from various sensors integrated into the mattress, and performing preprocessing and deep feature extraction. Through a five-layer deep convolutional neural network containing convolutional layers, pooling layers, and an attention mechanism, the multi-channel temporal signals are transformed into a 256-dimensional high-order feature vector to comprehensively and deeply represent the user's real-time physiological and sleep states. The steps include:

[0048] The device receives multi-channel physiological signals from mattress sensors, including electrocardiogram (ECG) signals, electromyography (EMG) signals, electrodermal response (GSR) signals, and respiratory waveform signals. Specifically, a bioelectric amplifier (such as TI's ADS1299) is used to acquire ECG and EMG signals at a sampling rate of 1000 Hz; a digital capacitive sensor (such as NXP's MPR121) is used to acquire GSR signals at a sampling rate of 100 Hz; and a high-sensitivity piezoelectric thin-film sensor array is embedded under the mattress fabric layer to monitor the waveform of changes in intrathoracic pressure caused by respiration at a sampling rate of 50 Hz.

[0049] The original signal undergoes preprocessing, including power frequency interference removal, baseline drift correction, and signal normalization. Power frequency interference removal utilizes a 50Hz notch filter; baseline drift correction employs a combination of first-order differential and moving average filtering; and signal normalization uses a sliding window-based Z-score normalization method. ,in for, For the original signal data points, and These represent the mean and standard deviation of the data within the current sliding window, respectively.

[0050] The preprocessed signal is input into a designed deep convolutional neural network, which contains three convolutional layers and two pooling layers to extract the time-frequency domain features of the signal. The preprocessed multi-channel signal is concatenated into a two-dimensional feature map (dimension: time step × number of channels), and then input into a five-layer deep convolutional neural network. The specific structure of this network is as follows: the first layer is a convolutional layer using 64 one-dimensional convolutional kernels with a width of 5 and the ReLU activation function; followed by a max-pooling layer with a width of 2; the second convolutional layer uses 128 convolutional kernels with a width of 3, followed by another max-pooling layer with a width of 2; the third convolutional layer uses 256 convolutional kernels with a width of 3. Finally, the output is flattened and reduced in dimensionality through a fully connected layer, outputting a 256-dimensional high-order feature vector.

[0051] The extracted features are weighted using an attention mechanism to highlight key features related to negative oxygen ion regulation. Specifically, a channel attention module (Squeeze-and-ExcitationBlock) is used: first, global average pooling (Squeeze) is performed on the feature map obtained from convolution to obtain a global descriptor for each channel; then, the weight relationship of each channel is learned through a bottleneck structure (Excitation) composed of two fully connected layers, and a weight coefficient between 0 and 1 is generated using the Sigmoid function; finally, the weight coefficient is multiplied with the original feature map channel by channel (Reweight) to achieve adaptive calibration of the feature channels.

[0052] The higher-order feature vectors are output to the next module, and the feature importance scores are recorded for model optimization.

[0053] The personalized strategy generation module receives feature vectors and, combined with user historical data and profiles, uses Deep Q-Network (DQN) to generate optimal regulatory instructions. Its core is a dynamically adjusted reward function that automatically optimizes the weights of various health indicators (such as improved heart rate variability and increased relaxation) through a meta-learning mechanism. This achieves truly personalized strategies for negative oxygen ion concentration, action area, and duration that adapt to individual differences and long-term changes. The steps include:

[0054] It receives high-order feature vectors from the feature extraction module and retrieves the user's historical regulation records and physiological response data from the user database.

[0055] The system matches current features with historical patterns to calculate similarity scores and identify trends in current physiological states. It also considers specific user needs, including age, health status preferences, and personalizes basic strategies accordingly. For similarity calculation, the Dynamic Time Warping (DTW) algorithm is used to calculate the similarity distance between the current feature sequence and feature template sequences of typical sleep stages (sleep onset, deep sleep, REM sleep) in the user's historical database; the smaller the distance, the higher the similarity score.

[0056] For personalized adjustments, a user profile vector P is maintained, including age, gender, underlying health markers (such as respiratory sensitivity, anxiety tendency), and the intensity of subjective preference for negative oxygen ions. In the input layer of the DQN network, the state vector... It is concatenated with the user profile vector P and used together as network input, so that the strategy generation process is intrinsically adapted to the user's personality.

[0057] A deep reinforcement learning algorithm, specifically a deep Q-network, is used to calculate the optimal control strategy based on the current state and the long-term reward function. The elements of the deep Q-network (DQN) are defined as follows:

[0058] State ): It is composed of the 256-dimensional physiological feature vector obtained from the previous module at time t, the current environmental temperature and humidity data, and the control actions taken in the previous period;

[0059] Action ): This is a three-dimensional discrete action space, including: concentration level (low: 1000-2000 ions / cm³, medium: 2001-5000 ions / cm³, high: 5001-10000 ions / cm³), action area mode (evenly applied to the whole body, with emphasis on the upper body, with emphasis on the lower body), and control duration (short: 10 minutes, medium: 20 minutes, long: 30 minutes).

[0060] Reward function ): ;in, Let be the immediate reward value obtained at time t (or the t-th decision period). The user profile matching score is a scalar used to quantify the degree to which the current state matches the user's personalized needs. The percentage improvement in heart rate variability (high-frequency power) after regulation; The percentage decrease in skin electrical conductivity level after regulation (negative values ​​indicate relaxation); This is a sleep efficiency score calculated based on the user's sleep stages (deep sleep, light sleep, REM). This is a penalty factor calculated based on the energy consumption of the negative ion generator; , , These are weighting coefficients used to balance the contributions of various indicators.

[0061] Among them, the weighting coefficient , , Instead of being fixed values, they are dynamically adjusted through a meta-learning module to adapt to different users' long-term preferences and seasonal environmental changes. The adaptive weight adjustment process is as follows:

[0062] Data collection phase: Continuously collect two types of key data: The first type is explicit user feedback data, which is collected regularly (e.g., weekly) through the mobile application to collect users' subjective comprehensive rating of their sleep quality over the past week. The first category is data with a score of 1 indicating very poor performance and a score of 5 indicating very good performance; the second category is environmental context data, including seasonal identifiers. (corresponding to spring, summer, autumn, and winter respectively) and the local historical average air quality index (AQI);

[0063] Weight adjustment strategy: The meta-learning module maintains an adjustable weight vector. and each Day (for example) =30 (i.e., monthly) updates once a month; the update objective is to minimize the difference between user subjective ratings and system predicted ratings, while taking into account the impact of environmental factors on the ideal weights;

[0064] Define system prediction score For the past Normalized aggregation of all effective regulatory cycle reward values ​​within a day:

[0065]

[0066] in User subjective ratings predicted based on historical reward data. The Sigmoid function maps the reward value to the (0,1) interval; For the past The number of effective control measures per day; For the first The reward value for this adjustment; multiplying by 4 and adding 1 is to adjust the output scale to match... Consistent [1,5] interval;

[0067] Loss function construction and optimization: The loss function for weight updates consists of a prediction error term and an environment regularization term.

[0068]

[0069] in The loss function value during the weight update process. The optimized weighting coefficients, i.e. , , , Used as seasonal markers (1: Spring, 2: Summer, 3: Autumn, 4: Winter). This represents the local historical average air quality index. The basic weight template vector is determined based on prior knowledge of the environment and medicine. This is the regularization coefficient, used to control the balance between personalized adjustments and environmental priors; For example, a basic weight template based on environmental and medical prior knowledge, such as:

[0070]

[0071] Online update execution: Each At the end of the day, the meta-learning module uses gradient descent to solve the problem. Update the weight coefficients; the specific update formula is as follows:

[0072]

[0073] in The updated weighting coefficients, The weighting coefficients before the update. The learning rate. New weighting coefficients. Will be used for the following The DQN reward calculation is underway. Simultaneously, historical weight sequences are retained, and a manual review mechanism is triggered when a significant weight mutation is detected to ensure the rationality of the adjustment.

[0074] Cold Start and Security Assurance: For new users, the initial weight is directly adopted. And fine-tune based on the health questionnaire filled out by the user during registration (if there is a history of respiratory diseases, reduce the amount). Those with anxiety tendencies will increase In addition, after each weight adjustment, the safety of the simulated regulation under the new weights will be verified to ensure that it does not exceed the medical safety boundary.

[0075] Through a multi-objective optimization weight adaptive mechanism, the relative importance of heart rate variability improvement, relaxation enhancement, and sleep efficiency in the optimization objectives can be dynamically adjusted based on the user's long-term subjective feelings, seasonal changes, and air quality conditions. This achieves true long-term personalized adaptation, thereby significantly improving user experience and health benefits while maintaining the stability of the algorithm framework.

[0076] It generates detailed control instructions that include target concentration, duration, and change gradient, and estimates the expected value of the control effect.

[0077] The intelligent ion field construction module translates control commands into a spatial concentration field that can be executed on the mattress surface. It divides the mattress into six independent zones, each equipped with an independent negative ion generator. By introducing an adaptive pulse modulation strategy, it dynamically adjusts the pulse frequency and duty cycle of each zone according to ambient temperature and humidity, and uses interleaved switching control to ultimately form a stable, uniform, and command-compliant gradient negative oxygen ion concentration distribution on the mattress surface. The steps include:

[0078] According to the control strategy instructions, the mattress is divided into multiple independent control zones, each equipped with an independent negative ion generating unit. In one embodiment, a standard double mattress is divided into six independent rectangular zones along its length, corresponding to the left / right head, left / right torso, and left / right legs, respectively. A negative ion generating plate based on a carbon nanotube emitting tip is embedded in the center of each zone. Its driving circuit is controlled by an independent MOSFET and can receive pulse width modulation (PWM) signals to adjust the ion generation rate.

[0079] The system controls the activation timing and intensity of the negative ion generators in each area to create a gradient-distributed negative ion concentration field on the mattress surface. For example, when the strategy command is to promote deep sleep, the control system generates a longitudinal gradient field where the concentration is higher in the head than in the torso, and higher in the torso than in the legs. This is achieved by assigning different PWM duty cycles (e.g., 70%, 50%, 30%) to the ion generators in the head, torso, and leg areas.

[0080] Based on fixed PWM duty cycle control, an adaptive pulse modulation strategy is further introduced to optimize the spatial distribution uniformity and stability of negative oxygen ions and compensate for the influence of environmental factors on ion diffusion efficiency; including the following specific processes:

[0081] Environmental parameter monitoring and ion mobility correction:

[0082] The system monitors the ambient temperature in real time. (Unit: °C) and relative humidity (Unit: %), and the correction coefficient for the mobility of negative oxygen ions in the air was calculated based on an empirical model. :

[0083]

[0084] in: The corrected air negative oxygen ion mobility;

[0085] The ambient temperature is monitored in real time, in °C.

[0086] The relative humidity of the environment is measured in real time, expressed as % (%).

[0087] For reference temperature, 25℃; For reference humidity, 50%;

[0088] For standard conditions ( =25℃, Reference values ​​for ion mobility at 50% (=50%);

[0089] For temperature coefficient, take values ​​of The effect of increasing reaction temperature on promoting ion diffusion;

[0090] The humidity coefficient is denoted by [value]. This reflects that excessive humidity increases air viscosity resistance;

[0091] This correction factor is used to subsequently adjust the pulse parameters to compensate for differences in ion diffusion efficiency caused by environmental changes.

[0092] Pulse frequency adaptive adjustment: basic pulse frequency Set to 50Hz; dynamically adjust the actual operating frequency according to environmental conditions. :

[0093]

[0094] in: The actual operating pulse frequency after adaptive adjustment;

[0095] The base pulse frequency is set to 50Hz;

[0096] This is the current average ion concentration measurement.

[0097] The target concentration value;

[0098] Feedback gain coefficient (0 < <0.2);

[0099] When the environment is conducive to ion diffusion ( When the concentration is too high or the actual concentration is too low, the frequency should be increased appropriately to enhance ion generation; conversely, the frequency should be decreased to avoid excessive accumulation of ions near the emitter.

[0100] Dynamic duty cycle modulation and gradient optimization: base duty cycle for each region (i=1,2,...,6 correspond to six regions) determined by the control strategy command; actual duty cycle Dynamic modulation over time:

[0101]

[0102] in: Let be the actual duty cycle of the i-th control region at time t; The base duty cycle of the i-th control region (determined by the control strategy command); It is a time variable; The modulation depth of region i (0≤ ≤0.25), correlated with regional importance (head region) Smaller values ​​are used to ensure stability, limb area The value can be increased appropriately).

[0103] The modulation period is set to a value between 5 and 15 minutes to avoid resonance with the user's physiological rhythm.

[0104] For the regional phase difference, according to The configuration forms a six-phase staggered modulation, reducing overall current surges.

[0105] Meanwhile, to maintain the stability of the concentration gradient, a gradient preservation constraint is introduced:

[0106]

[0107] in , These represent the actual duty cycles of the i-th and j-th regions, respectively. , These represent the target ion concentrations in the i-th and j-th regions, respectively. , For the area, For the area spacing, For reference distance;

[0108] Coordinated switching control and harmonic suppression: To avoid current harmonics and electromagnetic interference caused by multiple generators switching simultaneously, interleaved switching control is used. Define the control signal. for:

[0109]

[0110] in This represents the final control signal driving the negative ion generator in the i-th region. sign[...] is a sign function that returns +1 or -1 depending on the sign of the expression within the parentheses. =6 represents the total number of regions; this control strategy ensures that:

[0111] The switching times of adjacent areas differ by at least Second;

[0112] The number of generators operating simultaneously at any given time shall not exceed indivual;

[0113] The rate of change of total current is limited to a safe range;

[0114] Feedback closed-loop calibration: combining real-time monitoring data from an ion concentration sensor network. Establish a feedback relationship between concentration distribution and control parameters:

[0115]

[0116] Among the performance indicators Defined as:

[0117]

[0118] in: The index function value is used to evaluate and optimize the performance of concentration field tracking. Let be the ion concentration measured by the sensor in the i-th region at time t. The basic duty cycle of the i-th region rate of change, The performance index J represents the base duty cycle. The partial derivatives, Regional weights (the head region has a higher weight). For smoothing coefficients, The learning rate is adjusted online using gradient descent. This enables precise tracking of the concentration field.

[0119] The ion distribution in each area is monitored in real time through a matrix-type ion concentration sensor network. This network consists of miniature negative oxygen ion detection modules (such as the SDS011 improved module based on the parallel plate capacitor principle) distributed in the above six areas. Each module is networked through an I2C bus and reports the measured ion concentration of its area to the main controller every 10 seconds with an accuracy of ±100 ions / cm³.

[0120] The operating parameters of each generating unit are dynamically adjusted based on monitoring data to ensure the stable maintenance of the target concentration field.

[0121] When a change in the user's body position is detected, the concentration field distribution is automatically adjusted to ensure that key body parts are always in the optimal ion environment.

[0122] The closed-loop optimization module monitors the user's physiological response in real time after regulation, calculates effect indicators such as sleep quality, and compares them with the expected results. Using the generated deviation data, it updates the neural network model parameters in the strategy generation module through an online learning algorithm (such as stochastic gradient descent), thereby driving the entire system to continuously self-optimize and constantly improve the accuracy and personalization of regulation. The execution steps include:

[0123] Real-time monitoring of users' physiological response signals during the negative oxygen ion regulation process, including changes in heart rate variability, skin conductivity, and blood oxygen saturation;

[0124] The evaluation indicators for the regulation effect were calculated, including the relaxation improvement rate, sleep quality index, and physiological coordination score; the calculation methods for each indicator are as follows:

[0125] Relaxation improvement rate = (Average GSR value in the 5 minutes before the start of regulation - Average GSR value in the 5 minutes after the end of regulation) / Average GSR value in the 5 minutes before the start of regulation × 100%;

[0126] Sleep Quality Index (SQI) = 0.4 × (Deep sleep duration / Total time in bed) + 0.3 × (1 - Sleep latency / 60) + 0.3 × (1 - Number of awakenings per night / 5). All parameters are derived from periodic analysis of the aforementioned physiological signals.

[0127] Physiological coordination score = Pearson correlation coefficient (heart rate sequence, respiratory sequence) averaged over the control period. The closer the value is to 1, the better the cardiopulmonary coordination.

[0128] Compare and analyze the actual control effects with the expected effects to identify control deviations and pinpoint the causes.

[0129] Based on the deviation analysis results, an online learning algorithm is used to adjust the parameters of the control model in real time, specifically employing stochastic gradient descent to update the neural network weights; the optimization process is carried out on a daily basis: all state-action-reward-new state data sets collected each night are processed. , , , The data is stored in an experience replay buffer. Every morning, an offline optimization thread is started, randomly sampling a mini-batch of data from the buffer, calculating the temporal difference error (TDError) between the Q-value prediction of the current DQN network (online network) and the target Q-value (calculated through the target network), and minimizing this error using stochastic gradient descent to update the weights of the online network. The weights of the target network are synchronized from the online network every fixed period (e.g., every 100 training iterations).

[0130] Establish a user feedback mechanism to collect users' subjective feelings and ratings through mobile applications, and use them as part of the optimization goals;

[0131] Regularly generate reports on the effects of regulation and provide users with personalized health improvement suggestions.

[0132] The steps of the mattress negative oxygen ion concentration control method based on sensor data analysis are as follows:

[0133] Step 1: Signal Acquisition and Feature Extraction. This step involves real-time acquisition of raw physiological signals from the mattress's built-in multimodal sensors (ECG, EMG, SCEA, and respiration). The signals undergo filtering, correction, and standardization preprocessing before being input into a five-layer deep convolutional neural network integrating an attention mechanism to extract a 256-dimensional high-order feature vector. This feature vector deeply characterizes the user's real-time physiological state and sleep stage, providing crucial information for subsequent decision-making.

[0134] Step Two: Personalized Strategy Generation. The feature vector is received and combined with user historical data, personalized profiles (age, health status, etc.), and environmental data, then input into a Deep Reinforcement Learning (DQN) model. A meta-learning mechanism dynamically optimizes the weights of the reward function, enabling the DQN network to calculate the optimal negative oxygen ion regulation strategy, including the target concentration value, the effective region pattern, and the regulation duration.

[0135] Step 3: Intelligent ion field execution, converting the generated regulation strategy instructions into specific control signals. The system divides the mattress into six independent zones. Through an adaptive pulse modulation strategy, it dynamically adjusts the pulse frequency, duty cycle, and switching sequence of the negative ion generators in each zone according to the real-time ambient temperature and humidity, thereby constructing and maintaining a stable, uniform, and gradient-compliant negative oxygen ion concentration field on the mattress surface.

[0136] Step Four: Effect Evaluation and Deviation Analysis. During the implementation of the treatment, real-time monitoring of physiological response signals such as heart rate variability and skin conductance is conducted. Based on these signals, quantitative indicators such as relaxation improvement rate, sleep quality index, and physiological coordination score are calculated. The actual treatment effect is compared with the expected target to identify and locate treatment deviations.

[0137] Step 5: Closed-loop feedback and model update. Based on the deviation analysis results, the parameters of the DQN network in the strategy generation module are updated using an online learning algorithm. The model is optimized through experience replay and gradient descent, and user subjective feedback is incorporated into the optimization objective, thus forming a complete perception-decision-execution-optimization closed loop, driving the system to achieve continuous adaptation and performance improvement.

[0138] The above formulas are all dimensionless calculations. Dimensionless calculations can be performed using various methods such as standardization, which will not be elaborated here. The formulas are derived from software simulations based on a large amount of collected data, and the preset parameters in the formulas can be set by those skilled in the art according to the actual situation.

[0139] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, ATA hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. The semiconductor medium can be a solid-state ATA hard disk.

[0140] It should be understood that in the various embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0141] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0142] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0143] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0144] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0145] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable ATA hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0146] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A mattress negative oxygen ion concentration control system based on sensor data analysis, characterized in that, Includes the following modules: The multimodal feature extraction module is used to collect and preprocess raw physiological signals from multiple physiological sensors integrated in the mattress, and extract high-order feature vectors representing the user's real-time physiological state through a deep convolutional neural network. The personalized strategy generation module is used to receive the high-order feature vector and, in combination with the user's historical data and profile, use a deep reinforcement learning algorithm to generate personalized negative oxygen ion regulation strategy instructions. The regulation strategy instructions include at least the target concentration, the area of ​​action, and the regulation duration. The intelligent ion field construction module is used to control the working parameters of the negative oxygen ion generators in multiple independent zones on the mattress surface according to the control strategy instructions, so as to construct and maintain a negative oxygen ion concentration field that meets the gradient requirements on the mattress surface. The closed-loop optimization module is used to monitor the user's physiological response to negative oxygen ion regulation, evaluate the regulation effect and calculate the deviation from the expected target, and optimize the algorithm model parameters of the personalized strategy generation module online based on the deviation.

2. The mattress negative oxygen ion concentration control system based on sensor data analysis according to claim 1, characterized in that, The multimodal feature extraction module is specifically used for: It receives multi-channel physiological signals from mattress sensors, including electrocardiogram signals, electromyography signals, skin conductance response signals, and respiratory waveform signals; The raw physiological signals are preprocessed to remove interference, correct drift, and normalize. The preprocessed signal is input into a five-layer deep convolutional neural network to extract time-frequency domain features; The extracted features are weighted by a channel attention mechanism to output a high-order feature vector that represents the user's real-time physiological and sleep state.

3. The mattress negative oxygen ion concentration control system based on sensor data analysis according to claim 1, characterized in that, The personalized strategy generation module is specifically used for: Receive high-order feature vectors and retrieve user historical data and profiles; Match current features with historical patterns to identify physiological state trends; Based on deep reinforcement learning algorithms and combined with a reward function that dynamically adjusts weights, personalized regulation strategy instructions containing the target negative oxygen ion concentration, the area of ​​action, and the duration of regulation are calculated and generated.

4. The mattress negative oxygen ion concentration control system based on sensor data analysis according to claim 1, characterized in that, The intelligent ion field construction module is specifically used for: According to the control strategy instructions, the mattress is divided into multiple independent control zones; Control the independent negative oxygen ion generators in each area and adjust the ion generation rate through pulse width modulation signals; Based on real-time ambient temperature and humidity data, the pulse frequency and duty cycle are adaptively adjusted, and an interleaved switching control strategy is adopted to form and maintain a stable negative oxygen ion concentration field on the mattress surface that meets the gradient requirements.

5. The mattress negative oxygen ion concentration control system based on sensor data analysis according to claim 1, characterized in that, The closed-loop optimization module is specifically used for: Real-time monitoring of users' physiological response signals during the process of receiving negative oxygen ion regulation; The relaxation improvement rate, sleep quality index, and physiological coordination score are calculated based on the physiological response signals to evaluate the actual regulation effect. Compare and analyze the actual control effects with the expected effects to identify control deviations; Based on the bias, the model parameters in the personalized strategy generation module are updated using an online learning algorithm.

6. The mattress negative oxygen ion concentration control system based on sensor data analysis according to claim 2, characterized in that, The multi-channel physiological signals were acquired at different sampling rates using the following sensors: A bioelectric amplifier that acquires electrocardiogram and electromyography signals at a sampling rate of 1000 Hz, a digital capacitive sensor that acquires skin conductance response signals at a sampling rate of 100 Hz, and a piezoelectric thin film sensor array that monitors respiratory waveforms at a sampling rate of 50 Hz.

7. The mattress negative oxygen ion concentration control system based on sensor data analysis according to claim 3, characterized in that, The weighting coefficients of the reward function: The weight coefficients are periodically updated by dynamically adjusting a meta-learning module based on subjective sleep scores and environmental context data provided by users at regular intervals. This is achieved by minimizing the prediction score error and introducing an environmental prior regularization term.

8. The mattress negative oxygen ion concentration control system based on sensor data analysis according to claim 4, characterized in that, The intelligent ion field construction module also includes: Feedback closed-loop calibration uses a network of ion concentration sensors distributed across various regions to monitor the concentration distribution in real time. Based on the deviation between the monitored data and the target concentration, the operating parameters of the negative oxygen ion generators in each region are dynamically adjusted to achieve accurate tracking of the concentration field.

9. A method for controlling the concentration of negative oxygen ions in a mattress based on sensor data analysis, as described in any one of claims 1-8, characterized in that, Includes the following steps: S1: Physiological signals are collected from the mattress multimodal sensor and high-order feature vectors are extracted through a deep network; S2: Based on the feature vector, a personalized negative oxygen ion regulation strategy is generated using a deep reinforcement learning model; S3: According to the aforementioned control strategy, the negative oxygen ion generator is controlled in the mattress zones to construct a target concentration field; S4: Monitor the user's physiological response, evaluate the effect of regulation, and optimize the deep reinforcement learning model based on the bias.

Images