Intelligent wearable system integrating multi-modal physiological monitoring and active intervention

Abstract

The invention relates to the technical field of intelligent wearable equipment, and discloses an intelligent wearable system integrating multi-modal physiological monitoring and active intervention, which comprises a ring end acquisition module, a mobile end cache module and a cloud intelligent analysis module, and the cloud intelligent analysis module comprises a multi-modal fusion analysis unit, an active intervention decision unit and a user digital twinborn model construction unit, and is used for calculating a sleep score, a readiness score, a pressure index, a blood oxygen trend and a menstrual period prediction result according to the multi-source physiological characteristics, and executing an active intervention decision based on a health state. The sensing module is integrated in the ring body, the light-weight design is suitable for long-term wearing, all-weather data acquisition is guaranteed, compared with a bracelet or a watch, the ring is comfortable and hidden to wear, the continuous use willingness of a user is enhanced, meanwhile, the ring is suitable for high-voltage office or sports scenes, and non-inductive health monitoring is achieved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent wearable devices, in particular to an intelligent wearable system integrating multi-modal physiological monitoring and active intervention. BACKGROUND

[0002] With the acceleration of people's life pace and the increase of work pressure, mental health and physical recovery problems are increasingly prominent. In recent years, intelligent wearable devices have gradually become an important tool for individual health management, among which devices in the form of bracelets, watches, chest bands, etc. are mainly used. The existing intelligent wearable technology is mainly used for basic functions such as step counting, heart rate monitoring, sleep tracking, etc., but there are still the following deficiencies in mental health and sleep recovery: the existing devices usually only monitor a single physiological indicator such as heart rate or blood oxygen, lack multi-modal data fusion analysis capability, and it is difficult to comprehensively evaluate the overall health status, stress level and recovery ability of the user. Most of the existing devices rely on motion or heart rate data for sleep analysis, which cannot accurately identify sleep stages and depth; the psychological stress evaluation is usually realized through simple questionnaires or low-frequency HRV measurement, which lacks high-precision real-time monitoring capability, resulting in large deviation of sleep and stress evaluation results. SUMMARY

[0003] In view of the deficiencies of the prior art, the present application provides an intelligent wearable system integrating multi-modal physiological monitoring and active intervention, which solves the above technical problems.

[0004] To achieve the above purpose, the present application provides the following technical scheme: an intelligent wearable system integrating multi-modal physiological monitoring and active intervention, the system comprising a ring end acquisition module, a mobile end cache module and a cloud end intelligent analysis module; The ring end acquisition module is arranged inside the ring body and comprises a heart rate sensor, a blood oxygen sensor, a galvanic skin response sensor, a body temperature sensor and a three-axis acceleration sensor, which are used for synchronously acquiring multi-modal physiological signals of the user; The ring end acquisition module is internally provided with a low-power signal conditioning circuit and a local data cache unit, which are used for filtering, amplifying and short-time feature extraction of the acquired signals; The mobile end cache module communicates with the ring end acquisition module through low-power Bluetooth, and is used for receiving and encrypting the physiological signals uploaded by the ring end acquisition module, and uploading the data to the cloud end intelligent analysis module; The cloud end intelligent analysis module comprises a multi-modal fusion analysis unit, an active intervention decision unit and a user digital twin model construction unit, which are used for calculating sleep score, readiness score, stress index, blood oxygen trend and menstrual period prediction results according to multi-source physiological characteristics, and executing active intervention decisions based on the health status.

[0005] Preferably, the ring end collection module adopts a flexible annular electrode and an arc-shaped photoelectric lens structure to make the heart rate and blood oxygen signal collection area conform to the user's finger pulp, thereby improving signal stability and noise resistance. The shell of the collection module is composed of a ceramic composite material and a medical-grade silicone lining to improve wearing comfort and prevent metal allergy.

[0006] Preferably, the multi-modal fusion analysis unit includes a sleep identification sub-module, a stress identification sub-module, and a readiness calculation sub-module. The sleep identification sub-module identifies the user's sleep stage based on heart rate variability (HRV), skin temperature fluctuations, and motion state characteristics using a convolutional neural network-long short-term memory network (CNN-LSTM) model. The stress identification sub-module classifies stress states using 1000Hz high-frequency sampling data based on skin galvanic response signal (GSR) and heart rate variability fusion features. The readiness calculation sub-module generates a user's readiness score based on recovery time, metabolic load, and sleep depth.

[0007] Preferably, the sleep identification sub-module in the multi-modal fusion analysis unit includes the following processing steps: (1) Extract feature vectors from the heart rate variability (HRV), skin temperature fluctuations (TEMP), and three-axis acceleration (ACC) signals collected by the ring end:

[0008] Where: is the standard deviation of heart rate at time t; is the square root of the mean difference between adjacent heartbeats; is the low-to-high frequency power ratio; is the skin temperature gradient; is the acceleration sum; (2) Combine the features of the continuous N time windows into an input matrix:

[0009] (3) After extracting local features through the convolution layer and extracting time series features through the LSTM layer, output the probability of each sleep stage:

[0010] Where: is the LSTM hidden layer output; is the weight matrix; is the bias vector; (4) Calculate the sleep score based on the stage classification result:

[0011] wherein: is the deep sleep proportion; is the wakefulness proportion; is the sleep continuity coefficient; is the experience weight.

[0012] Preferably, the stress identification sub-module in the multi-modal fusion analysis unit has the following processing steps: (1) Collecting the electrodermal response signal at a sampling rate of 1000 Hz and calculating its derivative:

[0013] wherein: represents the electrodermal response original signal value at time ; represents the discrete time derivative at ; represents the sampling time interval; represents the time of the previous sampling point; (2) Calculating the frequency domain ratio of heart rate variability:

[0014] wherein: represents the low-frequency component power in the HRV spectrum (frequency band: ); represents the high-frequency component power in the HRV spectrum (frequency band: ); is the low-to-high frequency power ratio, used to reflect the balance of the autonomic nervous system; (3) Building a fusion feature vector within a given time window:

[0015] wherein: represents the mean value of within the time window; represents the variance of within the time window; represents the deviation of the body surface temperature from the individual baseline temperature (i.e. , unit: ); all "windows" are short-time analysis windows set by the system (e.g. seconds, which can be defined in the specific embodiments of the specification).

[0016] (4) Inputting to the full connection layer of the dual-channel CNN (double-channel convolutional neural network) for probability mapping:

[0017] wherein: represents the output stress probability value, with a value range of ; represents the Sigmoid function, defined as ; represents the weight vector or weight matrix of the fully connected layer (determined by training); represents the bias term (scalar or vector, depending on the network structure).

[0018] (5) Calculate the stress index based on :

[0019] and when the following condition is met:

[0020] , it is determined as a high stress state, wherein: StressIndex is the quantified stress index (range 0-100); is the high stress threshold set for the system; if it is necessary to determine the classification, it can be mapped to the categories of relaxation / mild tension / high stress according to or the interval of StressIndex.

[0021] Preferably, the readiness calculation sub-module in the multi-modal fusion analysis unit, the calculation steps of which include: (1) Construct the readiness feature vector:

[0022] wherein: is the deep sleep proportion in the reference sleep cycle (deep sleep duration / total sleep duration, with a value range of ); is the sleep continuity coefficient, which can be generated by the combination of the number of awakenings and the length of the longest continuous sleep segment (numerical range 0-1); is the stress index defined in claim 12 (0-100); is the post-exercise heart rate recovery time or heart rate recovery measure (e.g. the heart rate drop value 1 minute after the end of exercise, unit: beats / minute); is the resting heart rate (unit: beats / minute); is the exercise or metabolic load score in the past 24 hours (a dimensionless value obtained by weighting the number of steps, exercise intensity and training time).

[0023] (2) Standardize to obtain the standardized feature Then input the regression neural network Calculate the raw readiness score:

[0024] Where: represents the regression neural network mapping function, whose structure is: An input layer Several hidden layers (e.g. three layers, each layer containing several neurons and using ReLU activation) An output layer; the network determines the weight parameters from the training data; represents the raw score of the instantaneous readiness output by the regression network (not smoothed, usually a scaled value, which can be scaled and mapped to 100 interval).

[0025] (3) Exponential smoothing (exponential weighted moving average) is performed on the raw score to reduce the impact of short-term fluctuations:

[0026] Where: ReadyScore is the smoothed readiness score at time ; ReadyScore is the readiness score at the previous time; is the smoothing coefficient, with a value range of , usually taking 0.3-0.7 (the specific value can be set by system parameters or user preferences); the initial condition can be set as ReadyScore or initialized by the system.

[0027] (4) Scale ReadyScore to the final readiness score in the 0-100 interval and classify it according to the threshold:

[0028] Where: and are the minimum and maximum mapping reference values (for linear scaling) determined by training or historical data, and the setting method or adaptive estimation method can be given in the embodiments of the specification; represents truncating to the range of ; the final readiness score can be used to generate behavior recommendations (e.g. if FinalReadiness , recommend high-intensity training; if FinalReadiness , recommend rest and recovery).

[0029] Preferably, the active intervention decision unit dynamically generates a personalized intervention plan according to the cloud analysis results, including but not limited to: respiration rhythm guidance, mindfulness training, pre-sleep relaxation tips, exercise recovery suggestions, and menstrual health reminders; The intervention result is fed back to the user in real time through a mobile terminal application or a voice assistant.

[0030] Preferably, the digital twin model construction unit is used to generate an individual health portrait based on long-term physiological data of the user. The health portrait includes heart rate baseline, sleep cycle, stress fluctuation trend, and physiological cycle regularity, for realizing health state prediction and abnormality early warning.

[0031] Preferably, the cloud intelligent analysis module uses differential privacy and blockchain storage mechanism to encrypt and store the uploaded data; the differential privacy mechanism is used to protect the physiological data of the user from being reversed, and the blockchain storage module is used to guarantee the data authenticity and traceability.

[0032] Compared with the prior art, the present application provides an intelligent wearable system integrating multi-modal physiological monitoring and active intervention, which has the following beneficial effects: The present application integrates the sensing module in the ring body, which is lightweight and suitable for long-term wear, ensuring all-weather data acquisition. Compared with a bracelet or a watch, it is comfortable and concealed to wear, enhances the user's continuous use willingness, and is suitable for high-pressure office or sports scenes, realizing unobtrusive health monitoring.

[0033] The present application uploads the user intervention behavior data to the cloud in real time, which is used to update the individual model and intervention strategy, realizes closed-loop adaptive optimization, and continuously iteratively optimizes the intervention scheme, so as to make the health management strategy personalized and dynamic, and long-term improve the mental health, sleep quality and body recovery ability. BRIEF DESCRIPTION OF DRAWINGS

[0034] Figure 1 The intelligent wearable system block diagram of the present application. DETAILED DESCRIPTION

[0035] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0036] An intelligent wearable system integrating multi-modal physiological monitoring and active intervention, the system comprising a ring end acquisition module, a mobile terminal cache module and a cloud intelligent analysis module. The ring end collection module is arranged inside the ring body and comprises a heart rate sensor, a blood oxygen sensor, a galvanic skin response sensor, a body temperature sensor and a three-axis acceleration sensor, and is used for synchronously collecting multi-modal physiological signals of a user; The ring end collection module is internally provided with a low-power signal conditioning circuit and a local data buffer unit, which are used for filtering, amplifying and short-time feature extraction of the collected signals; The mobile terminal buffer module communicates with the ring end collection module through low-power Bluetooth, receives and encrypts the physiological signals uploaded by the ring end collection module, and uploads the data to the cloud intelligent analysis module; The cloud intelligent analysis module comprises a multi-modal fusion analysis unit, an active intervention decision unit and a user digital twin model construction unit, which are used for calculating sleep scores, readiness scores, stress indexes, blood oxygen trends and menstrual period prediction results according to multi-source physiological characteristics, and performing active intervention decisions based on health states.

[0037] The ring end collection module adopts a flexible ring-shaped electrode and an arc-shaped photoelectric lens structure, so that the heart rate and blood oxygen signal collection area is attached to the user's finger pulp, improving signal stability and noise resistance; The shell of the collection module is composed of a ceramic composite material and a medical-grade silicone lining, which is used to improve wearing comfort and prevent metal allergy.

[0038] The multi-modal fusion analysis unit comprises a sleep identification sub-module, a stress identification sub-module and a readiness calculation sub-module, the sleep identification sub-module identifies the sleep stage of the user by using a convolutional neural network-long short-term memory network (CNN-LSTM) model based on heart rate variability (HRV), skin temperature fluctuation and motion state characteristics; The stress identification sub-module classifies the stress state by using 1000Hz high-frequency sampling data based on the fusion features of galvanic skin response signals (GSR) and heart rate variability; The readiness calculation sub-module generates the readiness score of the user according to the recovery time, metabolic load and sleep depth.

[0039] The sleep identification sub-module in the multi-modal fusion analysis unit comprises the following processing steps: (1) extracting a feature vector from the heart rate variability (HRV), skin temperature fluctuation (TEMP) and three-axis acceleration (ACC) signals collected by the ring end:

[0040] Wherein: is the standard deviation of heart rate at time t; is the square root of the difference between adjacent heart rates; is the low-to-high frequency power ratio; For skin temperature gradient; For acceleration sum; (2) Form an input matrix by combining the features of N continuous time windows:

[0041] (3) After extracting local features by the convolution layer and extracting time series features by the LSTM layer, output the probability of each sleep stage:

[0042] Wherein: is the output of the LSTM hidden layer; is the weight matrix; is the bias vector; (4) Calculate the sleep score according to the stage classification result:

[0043] Wherein: is the deep sleep proportion; is the wakefulness proportion; is the sleep continuity coefficient; is the experience weight.

[0044] The processing steps of the stress recognition sub-module in the multi-modal fusion analysis unit include: (1) Collect the skin electrical response signal at a sampling rate of 1000 Hz , and calculate its derivative:

[0045] Wherein: represents the skin electrical response original signal value at time ; represents the discrete time derivative of ; represents the sampling time interval; represents the time of the previous sampling point; (2) Calculate the frequency domain ratio of heart rate variability:

[0046] Wherein: represents the low-frequency component power in the HRV spectrum; represents the high-frequency component power in the HRV spectrum; is the low-to-high frequency power ratio, used to reflect the balance of the autonomic nervous system; (3) Construct a fusion feature vector within a given time window:

[0047] Wherein: represents the mean of the stress values within the time window; represents the variance of the stress values within the time window; represents the deviation of the body surface temperature from the individual baseline temperature (i.e. , unit: ); (4) the probability mapping after inputting the into the fully connected layer of the dual-channel CNN (Convolutional Neural Network):

[0048] wherein: represents the output stress probability value, with a value range of ; represents the Sigmoid function, defined as ; represents the weight vector or weight matrix of the fully connected layer (determined by training); represents the bias term (scalar or vector, depending on the network structure); (5) calculate the stress index based on :

[0049] and determine as high pressure state when satisfying:

[0050] , wherein: StressIndex is the quantitative stress index; is the high pressure threshold set by the system.

[0051] The readiness calculation submodule in the multi-modal fusion analysis unit, the calculation steps of which include: (1) construct a readiness feature vector:

[0052] wherein: is the deep sleep proportion in the reference sleep cycle; is the sleep continuity coefficient, which can be generated by combining and normalizing the number of awakenings, the length of the longest continuous sleep segment, etc.; is the stress index defined in claim 12; is the post-exercise heart rate recovery time or heart rate recovery measure; is the resting heart rate; is the exercise or metabolic load score in the past 24 hours; (2) standardize to obtain the standardized features ​​Then input the regression neural network Calculate the original readiness score:

[0053] Wherein: represents a regression neural network mapping function, the structure of which is an input layer to several hidden layers to an output layer; the network determines the weight parameters from the training data; represents the instantaneous readiness original score output by the regression network; (3) Exponential smoothing is performed on the original score to reduce the influence of short-term fluctuations:

[0054] Wherein: ReadyScore is the smoothed readiness score at time t; is the readiness score at the previous time; is a smoothing coefficient, the value range of which is , usually 0.3-0.7; the initial condition can be set to ReadyScore or initialized by the system; (4) Scale ReadyScore to the final readiness score in the range of 0-100 and classify it according to the threshold:

[0055] Wherein: and are the minimum and maximum mapping reference values determined by the training or historical data, and the setting method or adaptive estimation method can be given in the embodiments of the specification; represents that is truncated to the range of .

[0056] The active intervention decision unit dynamically generates a personalized intervention plan according to the cloud analysis result, including but not limited to: Breathing rhythm guidance, mindfulness training, pre-sleep relaxation prompt, exercise recovery suggestion, and menstrual health reminder; The intervention result is fed back to the user in real time through a mobile terminal application or a voice assistant.

[0057] The active intervention decision unit is used to generate a personalized health intervention strategy according to the multi-dimensional health indicators (including sleep score SleepScore, stress score StressIndex, and readiness score ReadinessScore) output by the cloud fusion analysis unit. The decision-making process includes the following steps: (1) Feature normalization and state determination​ Three core indicators are normalized to interval, and the comprehensive health status vector is calculated:

[0058] wherein: SleepScore represents the normalized score of sleep quality; StressIndex represents the normalized score of stress level; ReadinessScore represents the normalized score of physical readiness.

[0059] User status categories are defined according to the following conditions: When and : high stress / sleep deficiency status is determined; When and : fatigue recovery status is determined; When : good status is determined; The remaining cases are defined as mild stress status.

[0060] (2) Intervention scheme generation According to the status category, the corresponding intervention template is called, and the decision function generates the intervention strategy set:

[0061] wherein, the output contains multiple dimensions of intervention items:

[0062] : breathing rhythm guide parameters (rhythm frequency, duration, mode), controlled by StressIndex; : mindfulness training scheme (audio type, guide duration, intensity), jointly regulated by StressIndex and SleepScore; : pre-sleep relaxation prompt (time, music / breathing rhythm combination), automatically triggered when SleepScore is below the threshold; : exercise recovery suggestion (training intensity adjustment coefficient , dynamically adjusted according to ReadinessScore;

[0063] When , the original training intensity is maintained, and when , the training intensity is halved; Menstrual health reminders (including menstrual cycle prediction, exercise intensity suggestions, and nutritional tips) are automatically triggered based on cloud-based physiological model output.

[0064] The digital twin model building unit is used to generate an individual health profile based on the user's long-term physiological data; The health profile includes baseline heart rate, sleep cycle, stress fluctuation trend, and physiological cycle pattern, which is used to predict health status and provide early warning of abnormalities.

[0065] Data integration and feature extraction 1. The physiological data collected by the system over a long period of time includes: heart rate Heart rate variability Skin conductance Skin temperature Motion acceleration Sleep stage marker (SleepStage) ); Cycle information ( ).

[0066] 2. Perform statistical analysis and time series feature extraction on the collected data:

[0067]

[0068]

[0069]

[0070] in, User's baseline heart rate, in beats per minute; : The number of sampling points within the statistics window; Heart rate standard deviation, used to reflect heart rate fluctuations; Total sleep duration on a certain night Deep sleep duration; Deep sleep percentage, value StressIndex Stress score; Pressure trends are smoothed using an exponential moving average (EMA). Smoothing coefficient .

[0071] (2) Generation of individual health profiles Construct a health profile vector from long-term statistics and trend characteristics:

[0072] Wherein: is the physiological cycle phase (such as menstrual cycle days or physiological stage indicators) for periodic health assessment; It can be updated daily / weekly / monthly to form a dynamic digital twin.

[0073] (3) Health status prediction and abnormal early warning 1. Construct a prediction model :

[0074] Wherein: represents the health status prediction value at future time , such as sleep quality, stress level or heart rate abnormal probability; is the prediction time span, which can be set according to the application (for example, 1 day, 1 week).

[0075] 2. Abnormal early warning conditions:

[0076] Wherein: represents the abnormal trigger; , is the system setting or user personalized threshold value for judging whether the health indicators are too low or too high.

[0077] 3. Prediction model Can be used: time series model: LSTM, GRU, etc. Neural network; statistical model: ARIMA, exponential smoothing; hybrid model: weighted prediction combining historical statistical features and deep learning output.

[0078] Through the above steps, a digital twin health portrait based on long-term data can be generated, realizing personalized health status prediction, abnormal early warning and intervention scheme input.

[0079] The cloud intelligent analysis module uses differential privacy and blockchain storage mechanism to encrypt the uploaded data; the differential privacy mechanism is used to protect the user's physiological data from being reversed, and the blockchain storage module is used to guarantee the authenticity and traceability of the data.

[0080] Although embodiments of the present application have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A smart wearable system integrating multimodal physiological monitoring and active intervention, characterized in that, The system includes a ring-end data acquisition module, a mobile terminal caching module, and a cloud-based intelligent analysis module; The ring-end acquisition module is located inside the ring body and includes a heart rate sensor, a blood oxygen sensor, a skin conductance sensor, a body temperature sensor, and a triaxial accelerometer, used to synchronously acquire the user's multimodal physiological signals. The ring-end acquisition module is equipped with a low-power signal conditioning circuit and a local data buffer unit, which are used to filter, amplify and extract short-time features from the acquired signal. The mobile terminal caching module communicates with the ring terminal acquisition module via Bluetooth Low Energy to receive and encrypt the physiological signals uploaded by the ring terminal acquisition module, and upload the data to the cloud intelligent analysis module. The cloud-based intelligent analysis module includes a multimodal fusion analysis unit, an active intervention decision-making unit, and a user digital twin model construction unit. It is used to calculate sleep scores, readiness scores, stress index, blood oxygen trends, and menstrual cycle prediction results based on multi-source physiological characteristics, and to execute active intervention decisions based on health status.

2. The intelligent wearable system integrating multimodal physiological monitoring and active intervention according to claim 1, characterized in that, The ring-end acquisition module adopts a flexible ring electrode and arc photoelectric lens structure, so that the heart rate and blood oxygen signal acquisition area fits the user's fingertip, improving signal stability and anti-noise performance. The outer shell of the acquisition module is composed of ceramic composite material and medical-grade silicone lining, which is used to improve wearing comfort and prevent metal allergies.

3. The intelligent wearable system integrating multimodal physiological monitoring and active intervention according to claim 1, characterized in that, The multimodal fusion analysis unit includes a sleep recognition submodule, a stress recognition submodule, and a readiness calculation submodule. The sleep recognition submodule identifies the user's sleep stages based on heart rate variability (HRV), skin temperature fluctuations, and movement state characteristics, using a convolutional neural network-long short-term memory network (CNN-LSTM) model. The stress recognition submodule classifies stress states based on the fusion features of skin conductance response (GSR) signals and heart rate variability, using 1000Hz high-frequency sampling data. The readiness calculation submodule generates a user's readiness score based on a combination of recovery time, metabolic load, and sleep depth.

4. The intelligent wearable system integrating multimodal physiological monitoring and active intervention according to claim 1, characterized in that, The sleep recognition submodule in the multimodal fusion analysis unit includes the following processing steps: (1) Extract feature vectors from the heart rate variability (HRV), skin temperature fluctuation (TEMP), and triaxial acceleration (ACC) signals collected from the ring end: in: Let t be the standard deviation of heart rate at time t; The root mean square of the time difference between adjacent heartbeats; The ratio of low-frequency to high-frequency power; For skin temperature gradient; The resultant of acceleration; (2) Combine the features of N consecutive time windows into an input matrix: (3) After extracting local features through convolutional layers and time series features through LSTM layers, the probabilities of each sleep stage are output: in: Output of the LSTM hidden layer; This is the weight matrix; It is the bias vector; (4) Calculate the sleep score based on the stage classification results: in: The percentage of deep sleepers; For the awakening ratio; The sleep continuity coefficient; For experience weights.

5. The intelligent wearable system integrating multimodal physiological monitoring and active intervention according to claim 1, characterized in that, The pressure identification submodule in the multimodal fusion analysis unit includes the following processing steps: (1) Acquire skin conductance response signals at a sampling rate of 1000 Hz And calculate its derivative: in: Indicates at time The original signal value of the skin conductance response; express The discrete-time derivative; Indicates the sampling time interval; Indicates the time of the previous sampling point; (2) Calculate the frequency domain ratio of heart rate variability: in: This represents the power of the low-frequency component in the HRV spectrum; This represents the power of the high-frequency components in the HRV spectrum; The low-to-high frequency power ratio is used to reflect autonomic nervous system balance. (3) Construct a fused feature vector within a given time window: in: Indicates within this time window The mean; Indicates within this time window The variance; This indicates the deviation of body surface temperature from an individual's baseline temperature (i.e., ,unit: ); (4) Probability mapping is performed on the fully connected layer following the input to the dual-channel CNN (two-channel convolutional neural network): in: This represents the output stress probability value, with a range of values ​​of [value missing]. ; The Sigmoid function is defined as follows: ; Represents the weight vector or weight matrix of the fully connected layer (determined by training); Represents the bias term (scalar or vector, depending on the network structure); (5) Based on Calculate the stress index: And when the following conditions are met: The system is determined to be in a high-pressure state, where: StressIndex is a quantified pressure index; The high voltage threshold set for the system.

6. The intelligent wearable system integrating multimodal physiological monitoring and active intervention according to claim 1, characterized in that, The readiness calculation submodule in the multimodal fusion analysis unit includes the following calculation steps: (1) Constructing the readiness feature vector: in: The percentage of deep sleep within the reference sleep cycle; The sleep continuity coefficient can be generated by normalizing a combination of factors such as the number of awakenings and the length of the longest continuous sleep segment. The pressure index as defined in claim 12; This refers to the heart rate recovery time or heart rate recovery measure after exercise. This is the resting heart rate; Scoring of exercise or metabolic load over the past 24 hours; (2) To Standardization processes are performed to obtain standardized features. Then input into the regression neural network Calculate the raw readiness score: in: This represents the mapping function of a regression neural network, whose structure is an input layer to several hidden layers to an output layer; the weight parameters of this network are determined by the training data. This represents the raw score of instantaneous readiness output by the regression network; (3) Apply exponential sliding smoothing to the original scores to reduce the impact of short-term fluctuations: Among them: ReadyScore For a moment The smoothed readiness score; The score represents the level of readiness at the previous moment; This is a smoothing coefficient, with a range of values. Typically, it is set to 0.3-0.7; the initial condition can be set to ReadyScore. Or it may be initialized by the system; (4) Set ReadyScore The final readiness scores are scaled to the 0-100 range and categorized by threshold: in: and The minimum and maximum mapping baseline values, respectively, are determined from training or historical data. The setting method or adaptive estimation method can be given in the embodiments of the specification. Indicates will Cut off to Within the range.

7. The intelligent wearable system integrating multimodal physiological monitoring and active intervention as described in claim 1, characterized in that, The proactive intervention decision-making unit dynamically generates personalized intervention plans based on cloud-based analysis results, including but not limited to: The intervention includes guided breathing rhythms, mindfulness training, relaxation tips before sleep, exercise recovery suggestions, and menstrual health reminders; the results of the intervention are fed back to the user in real time via a mobile application or voice assistant.

8. The intelligent wearable system integrating multimodal physiological monitoring and active intervention as described in claim 1, characterized in that, The digital twin model building unit is used to generate an individual health profile based on the user's long-term physiological data; The health profile includes baseline heart rate, sleep cycle, stress fluctuation trend, and physiological cycle pattern, which is used to predict health status and provide early warning of abnormalities.

9. The intelligent wearable system integrating multimodal physiological monitoring and active intervention as described in claim 1, characterized in that, The cloud-based intelligent analysis module uses differential privacy and blockchain evidence storage mechanisms to encrypt and store uploaded data; the differential privacy mechanism is used to protect user physiological data from being reverse-engineered, and the blockchain evidence storage module is used to ensure the authenticity and traceability of the data.

Images