Mental stress assessment and intelligent dredging method and system

Abstract

The invention discloses a mental stress assessment and intelligent dredging method and system, and belongs to the technical field of digital medical treatment and health informatics. Multi-modal physiological signals are continuously collected through the wearable device, medical diagnosis level pressure state evaluation is carried out based on the personalized physiological baseline, and personalized pressure indexes and levels are generated; secondly, when it is diagnosed that the pressure level exceeds the standard, the system serves as an intelligent decision support system, the environment and schedule information after privacy protection processing is fused, and an optimal grooming action is dynamically selected from a predefined intervention action library and executed; finally, the system serves as a continuous learning system, the intervention efficiency is evaluated in real time according to feedback data of the pressure index after execution, the decision model is updated, and collaborative self-optimization of the diagnosis strategy and the intervention strategy is achieved. According to the method, medical diagnosis, personalized treatment decision and adaptive learning are integrated, and the systematicness, accuracy and intelligent level of mental stress related health problem management are improved.

Description

Technical Field

[0001] This invention belongs to the field of digital healthcare and health informatics technology, specifically relating to a method and system for assessing and intelligently managing mental stress. Background Technology

[0002] Against the backdrop of rapid development in digital health technologies, building intelligent health management systems that integrate information and communication technologies (ICT) has become a core direction for smart healthcare. Mental stress, as a key precursor to anxiety, depression, and even psychosomatic illnesses, has significant clinical and preventative value in its early identification and proactive intervention for maintaining individual mental health. Therefore, developing an integrated health information technology system capable of achieving high-precision, real-time physiological state diagnosis and simultaneously driving personalized intervention responses not only aligns with the cutting-edge trends in precision medicine and health management but also holds significant importance for improving public health effectiveness.

[0003] However, existing technological solutions are mostly simple combinations of isolated physiological monitoring devices and static intervention content libraries, failing to build a truly closed-loop, data-driven medical and health information system with continuous learning capabilities.

[0004] Existing technologies intermittently collect single or limited-dimensional physiological indicators (such as heart rate variability, HRV) from users using dedicated analyzers and compare the collected values ​​with fixed thresholds preset based on population data to determine stress levels. The intervention mechanism involves pushing a uniform adjustment request to the user at the start of the next preset cycle. While this approach achieves automated detection, it has fundamental limitations from the perspective of healthcare information systems: its assessment criteria are non-personalized, and the fixed population thresholds cannot establish an effective individual health baseline model, resulting in low accuracy and clinical reference value in state diagnosis and a high risk of misjudgment; its intervention process is decoupled from real-time diagnostic results, with a significant delay between stress detection and intervention action, and the intervention action is completely independent of the user's real-time physiological feedback. This is an open-loop model where diagnosis and treatment decisions are disconnected, unable to adaptively adjust based on intervention effects, and does not conform to the basic principles of closed-loop health management.

[0005] Alternatively, existing technologies utilize remote servers for comprehensive analysis and risk warning. While this approach incorporates multi-source data input, its core risk assessment and intervention decision-making loop heavily relies on human intervention. Diagnostic results must be sent to counselors or clinicians for manual review and intervention decisions. This results in the system failing to form an autonomous, real-time diagnostic-decision information processing closed loop, leading to low response efficiency, poor scalability, and difficulty in achieving continuous and proactive health monitoring.

[0006] Examining the technical requirements of modern digital healthcare for continuous monitoring, real-time diagnosis, and personalized treatment decision support, the aforementioned existing technical solutions reveal the following deep-seated, structural flaws in system architecture and information processing, which limit their effectiveness and universality in clinical or high-standard health management scenarios:

[0007] Existing technologies treat state diagnosis and intervention execution as two sequential but information-decoupled stages. The triggering conditions, type selection, and intensity setting of intervention actions are not directly and dynamically driven by continuous, real-time physiological state diagnosis results. This leads to health interventions often lagging behind the actual increase in stress, and intervention strategies cannot be adjusted or switched online based on the user's immediate physiological feedback during execution, resulting in poor timeliness and rigid strategies in medical interventions, failing to form an effective diagnosis-treatment loop.

[0008] Current technologies generally rely on static thresholds based on population statistics or simple comparisons of users' own historical means. This fails to fundamentally establish an accurate baseline model of physiological response for individuals, nor can it accommodate the long-term, slow drift of users' physiological baselines with age, lifestyle habits, and health status. As a result, the accuracy of health status diagnosis is limited, potentially leading to persistent false positives (for users with high baselines) or false negatives (for users with low responsiveness), thus weakening the reliability and usability of the system as a health management tool.

[0009] Existing intervention strategy libraries are typically predefined and static. The system lacks the ability to learn from historical clinical interaction data, and cannot memorize and optimize the crucial diagnostic-intervention mapping knowledge—specifically, which intervention action is most effective for the current user in which context. Each intervention decision is almost isolated, failing to form continuously evolving personalized health management strategies based on individual user feedback data, thus limiting the system's long-term effectiveness and adaptability as an intelligent medical assistance tool.

[0010] While some technologies mention environmental information, they mostly remain at the level of simple recording or single-dimensional triggering, lacking a real-time context-aware framework that integrates multi-source information (location, sound, schedule) with clear priorities and strict privacy protection mechanisms. This leads to the system potentially triggering inappropriate interventions in unsuitable situations (such as while driving or during clinical diagnosis and treatment), or failing to utilize contextual information to select the safest and most acceptable health intervention method, thus reducing its safety and context-appropriateness as a medical assistance system. Summary of the Invention

[0011] The purpose of this invention is to provide a method and system for assessing and intelligently managing mental stress, which solves the technical problems of delayed intervention, rigid strategies, and lack of continuous self-learning ability caused by the decoupling of assessment and intervention logic in existing stress management technologies.

[0012] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows:

[0013] A method for assessing and intelligently managing mental stress includes the following steps:

[0014] S1, Signal Acquisition and Personalized Modeling: Multimodal physiological signals of users are continuously acquired through wearable devices. After preprocessing and feature extraction, the signals are normalized based on the user's personal physiological baseline and then weighted and fused to generate a real-time personalized stress index.

[0015] S2, Real-time Status Determination and Triggering: The personalized stress index is compared with a set of personalized dynamic thresholds and mapped to discrete stress levels; when the stress level reaches the preset intervention trigger level, the intervention decision process is initiated.

[0016] S3, Context Awareness and Fusion: Acquires environmental and schedule information in parallel, processes it for privacy protection, and then fuses it according to preset priority rules to generate the current context label;

[0017] S4, Intelligent Decision Making and Action Selection: Based on the current context label, the predefined intervention action library is filtered for context; combined with user preferences in the dynamic user profile and historical intervention effectiveness data, effectiveness filtering and ranking are performed to select the optimal intervention action;

[0018] S5, Closed-loop execution and feedback learning: Execute the optimal intervention action and monitor the changes in the personalized stress index in real time after execution; evaluate the intervention effect based on the stress change rate, adjust the intervention strategy in real time according to the evaluation results, and update the historical intervention effectiveness data to achieve self-optimization.

[0019] Furthermore, in step S1, generating a real-time personalized stress index specifically includes:

[0020] Extract physiological features from multimodal physiological signals, including at least heart rate variability, skin conductance, electromyographic activity, respiratory rate, and range of motion.

[0021] Use formula Personalized Z-score normalization was performed on each physiological characteristic, among which, Let be the real-time value of the i-th feature. and The individual baseline mean and standard deviation are calculated based on data from when users self-labeled themselves as relaxed during the calibration period; Indicates the first A physiological characteristic at any time Standardized eigenvalues;

[0022] The normalized features are then summed using a weighted average. The personalized stress index is obtained. ,in, The total number of physiological characteristics; Indicates the first Weight coefficients for each feature; weight coefficients Based on preset rules, the rules can be dynamically adjusted according to user feedback during actual use.

[0023] Furthermore, in step S2, the personalized dynamic threshold is determined in the following way:

[0024] During the calibration period, a series of stress index sample values ​​were calculated using fixed weighting coefficients and individual baseline parameters;

[0025] The sample values ​​are sorted in ascending order, and their 70th, 85th, 92nd, and 98th percentiles are calculated and assigned to four threshold values ​​respectively. ,satisfy ;

[0026] By comparing personalized stress indices The pressure level will be determined by the threshold value. The value is determined to be a discrete value between 0 and 4, where Intervention is triggered at the appropriate time.

[0027] Furthermore, in step S3, the generation of the context label depends on the fusion of at least two information sources, and is performed in the following priority order:

[0028] Highest priority: Standardized event category tags parsed from user calendar events;

[0029] Second highest priority: Predefined location labels determined by geofencing technology based on user location information;

[0030] Priority Supplement: Semantic tags identified from environmental sounds using a local audio event detection model.

[0031] Furthermore, the user location information and calendar event category information undergo differential privacy perturbation processing before fusion, wherein the location information addition satisfies a Laplace distribution. Noise, scale parameters The accuracy of the geographical range of the required protected location is set; the schedule event categories are set by probability. Perform random flipping, that is, in each processing cycle, with probability... Replace the true category with a value randomly selected from the other possible categories.

[0032] Furthermore, in step S4, the performance filtering and ranking specifically includes:

[0033] Query the action-context efficacy table maintained in the dynamic user profile. This table records the historical average efficacy value of each intervention action for users in different contexts. ;

[0034] The historical average performance value The calculation method is as follows: ,in The stress index at the start of the intervention. The value taken when the pressure index reaches a steady state after the intervention ends. This steady state is determined by predefined numerical conditions, specifically: within a continuous period... Within each sampling period, the pressure index The absolute values ​​of all changes are less than the preset stability threshold. ,in , ; The duration of the intervention.

[0035] From the candidate actions filtered by the context, actions with historical average effectiveness values ​​lower than a preset effectiveness threshold are removed, and from the remaining actions, the action with the highest historical average effectiveness value in the current context is selected as the optimal intervention action.

[0036] Furthermore, in step S5, the assessment of intervention effectiveness and strategy update based on the rate of change of pressure includes:

[0037] Preset time window after performing the selected intervention action Internally, the time of intervention start is used as Calculate the average rate of decrease of the pressure index. ;

[0038] in, For intervention implementation Stress index after a period of time;

[0039] like ,in If the minimum effective rate of descent is a preset threshold, the intervention is deemed effective, and the formula is used. Update the historical average efficacy value of the corresponding item in the action-situation efficacy table, where Forgetting factor, This is the efficacy value calculated for this intervention; This represents the updated historical average performance value; This represents the historical average performance value before the update;

[0040] like If the intervention fails to achieve the expected results, the action switching or intensity enhancement strategy should be implemented immediately.

[0041] Furthermore, the method also includes a multi-timescale system adaptive learning step:

[0042] Short-term learning: After each intervention, update the action-situation efficacy scale;

[0043] Chung-Shi Learning: Regularly recalculates individual physiological baseline parameters using newly tagged relaxation state data from users. and After the baseline is updated, perform at least one of the following operations on the action-context performance table:

[0044] (a) All intervention events recorded under the old baseline conditions are marked as historical data, and a weighted mixed update formula is used when calculating the new historical average efficacy value:

[0045] ;

[0046] in This represents the average performance value under the old baseline. This represents the average performance value under the new baseline. The weighting coefficients for historical data satisfy the following conditions: ;

[0047] Or (b) clear the performance table and restart recording and learning based on the new baseline;

[0048] Long-term learning: Regularly recalculate the stress level classification threshold based on the user's newly generated relaxation state stress index data, generate updated suggestions, and take effect after user confirmation.

[0049] Furthermore, the actions in the predefined intervention action library include at least one of the following types: binaural beat or guided breathing training played through bone conduction audio, deep pressure stimulation simulated through a haptic feedback device, color temperature and illuminance adjusted through an ambient lighting adjustment device, and interactive cognitive reconstruction or mindfulness meditation guidance presented on a mobile terminal application interface.

[0050] In addition, this invention also discloses a mental stress assessment and intelligent counseling system for implementing the mental stress assessment and intelligent counseling method described above, comprising:

[0051] The signal acquisition module, deployed in wearable devices, is used to continuously acquire multimodal physiological signals;

[0052] The edge computing unit, connected to the signal acquisition module, is used for preprocessing physiological signals, feature extraction, and personalized stress index calculation.

[0053] The central collaborative controller, connected to the edge computing unit, is used to execute stress level determination, context fusion, intervention decision-making, and learning update logic;

[0054] A context-aware interface, connected to the central collaborative controller, is used to acquire environmental and schedule information; an execution terminal, connected to the central collaborative controller, is used to execute intervention actions.

[0055] The edge computing unit and the central collaborative controller form an edge-central collaborative processing architecture. The central collaborative controller assigns the highest priority to the pressure level determination task, ensuring that the response time from the pressure level exceeding the threshold to the start decision is less than 100 milliseconds.

[0056] Compared with the prior art, the present invention has the following beneficial effects:

[0057] This invention calculates personalized stress indices and stress levels by collecting multimodal physiological signals in real time, and uses this as the core state feedback quantity to dynamically drive the execution and optimization of intervention actions. This completely changes the existing linear decoupled mode of detection-alarm-human response, realizing a paradigm shift from passive, delayed alarms to active, real-time, and adaptive adjustment, significantly improving the timeliness and accuracy of interventions.

[0058] This invention acquires physiological data of users in a relaxed state through initial calibration, establishes an individual physiological baseline, and calculates the relative deviation using the Z-score normalization method to synthesize a personalized stress index. The threshold on which stress level determination depends is also dynamically generated based on a specific percentile of the individual stress index distribution and supports periodic updates. This fundamentally overcomes the limitations of existing technologies that use fixed group thresholds, enabling the assessment results to accurately reflect the unique physiological response patterns of individuals and their long-term changes, significantly improving the accuracy and reliability of the assessment. An intelligent decision-making mechanism integrating multi-source context awareness and privacy protection is constructed, enhancing the adaptability and safety of interventions. This invention integrates multi-source information such as schedules, locations, and audio through priority rules to generate current context labels, and introduces differential privacy perturbation technology at the information processing front end. During decision-making, the system combines context labels with a historically recorded action-context efficacy table for two-layer filtering (context filtering and efficacy filtering), ensuring that the recommended actions not only conform to environmental constraints and user preferences but are also historically optimal in that context. This makes the intervention strategy highly context-adaptable and user-acceptable, while ensuring the security of user privacy data.

[0059] This invention calculates the average rate of decrease in the stress index in real time after the intervention is performed to quantify the effect, and updates the action-context efficacy table based on the results using an algorithm with a forgetting factor. The system also possesses multi-timescale learning capabilities, including short-term (single update), medium-term (updating baseline and decaying historical efficacy), and long-term (updating threshold) adaptation. This allows the system to continuously learn from the user's actual physiological feedback, constantly optimizing the optimal intervention strategy library for that user in different contexts, thereby achieving a long-term personalized intervention effect that becomes increasingly intelligent with use. Attached Figure Description

[0060] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained from these drawings without creative effort.

[0061] Figure 1 This is an overall flowchart of the method described in this invention.

[0062] Figure 2 This is a flowchart of step S1 of the method described in this invention.

[0063] Figure 3 This is a system architecture diagram of the present invention. Detailed Implementation

[0064] In the following description, only certain exemplary embodiments are briefly described. As those skilled in the art will recognize, the described embodiments can be modified in various ways without departing from the spirit or scope of the embodiments of the invention. Therefore, the drawings and description are considered to be exemplary in nature and not restrictive.

[0065] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0066] Example 1: See Figure 1 and Figure 2 This embodiment discloses a method for assessing and intelligently managing mental stress, specifically a closed-loop stress regulation method based on real-time physiological feedback, establishing an autonomous cycle of perception-decision-execution-learning. Multimodal physiological signals are continuously collected through wearable devices, and a comprehensive, personalized stress index is calculated in real time. and the corresponding pressure level When the pressure level When the preset intervention trigger level (e.g., moderate anxiety) is reached, immediately apply the current situation label. Using a motion-situation efficacy scale that records historical efficacy, the optimal action is selected and executed from a predefined intervention action library; subsequently, stress indices are monitored. Based on the rate of change after implementation, the intensity of intervention or the action can be adjusted in real time, and the performance table can be updated to optimize future decisions, thereby achieving continuous self-optimization of the intervention strategy.

[0067] This embodiment focuses on the closed-loop pressure regulation method and includes the following steps:

[0068] Signal Acquisition and Personalized Modeling: Raw physiological signals are acquired through multiple sensors, preprocessed, and feature-calculated. Personalized stress indices are then established using user-calibrated data. Computational model.

[0069] Real-time status determination: continuous calculation And determine the real-time pressure level by comparing it with personalized thresholds. .

[0070] Context awareness: Acquires environmental and schedule information in parallel and generates current context labels. .

[0071] Intelligent decision-making: when At the time, combined Using a historical efficacy action-situation efficacy table, the optimal intervention action is selected through filtering and ranking.

[0072] Closed-loop execution and learning: Execute the selected action and monitor it during and after execution. Based on the changes, actions or intensity are adjusted in real time according to the intervention effect, and the efficacy scale is updated to optimize future decisions.

[0073] The following detailed description, in conjunction with the accompanying drawings, details the hardware modules, data processing, and process control involved in each step.

[0074] The overall system architecture of this embodiment includes a multimodal physiological signal acquisition module deployed in a wearable terminal device, a local edge computing unit, a central collaborative controller, an execution terminal, and a closed-loop feedback calibration mechanism. Each component interacts with the other via Bluetooth Low Energy 5.0 protocol or a local bus, allowing the entire system to independently complete the entire process from signal acquisition to signal delivery even without a network connection.

[0075] The multimodal physiological signal acquisition module consists of a heart rate variability sensor, skin conductance electrodes, electromyography (EMG) activity detection unit, respiratory rate monitor, and inertial measurement unit. The heart rate variability sensor uses photoplethysmography (PPG) technology to continuously acquire raw PPG signals at a sampling rate of 200Hz. The skin conductance electrodes are placed on the inside of the user's palm and measure changes in skin conductivity through constant current source excitation. The EMG activity detection unit integrates a bipolar differential amplifier circuit and is attached to the surface of the forearm flexor muscles to capture local muscle activation. The respiratory rate monitor uses a flexible strain gauge surrounding the lower edge of the chest cavity, reflecting the respiratory cycle through deformation voltage output. The inertial measurement unit includes a triaxial accelerometer and a triaxial gyroscope to identify changes in user posture and movement.

[0076] All sensors start and acquire data independently, with each data packet carrying a millisecond-level system timestamp uniformly distributed by the local edge computing unit. During the preprocessing stage, the algorithm uses this timestamp to interpolate and align the signals from each channel to a unified time-series grid, ensuring the temporal correlation of subsequent cross-modal feature calculations. The system does not require physical clock synchronization of the sensor hardware.

[0077] The raw sampling data is transmitted to the local edge computing unit via Bluetooth Low Energy 5.0. This unit has a built-in preprocessing engine that performs targeted denoising and correction operations on each signal. For heart rate variability signals, the preprocessing engine uses a wavelet thresholding denoising algorithm, employing the Daubechies4 wavelet basis function, to perform soft thresholding suppression on high-frequency noise components with a decomposition level of 5. Electrodermal signals are processed using a sliding window median filter with a window length set to 500 milliseconds to effectively eliminate transient spike interference caused by limb micro-movements. Electromyographic signals are first passed through a second-order Butterworth bandpass filter, with the passband range limited to 20Hz to 500Hz, and then input to a zero-phase digital filter to avoid timing shifts caused by phase distortion. Respiratory signals employ an adaptive baseline drift correction algorithm, dynamically adjusting the zero-point offset based on local minimum tracking. The output of the inertial measurement unit is fused using a Kalman filter to generate a smooth attitude angle sequence.

[0078] The preprocessed multimodal signals are sent to the pressure index calculation module. This module calculates the pressure index according to a preset time window. (For example, The following calculations are performed continuously for a specified number of seconds and a sliding interval (e.g., 1 second) to output a personalized stress index. :

[0079] Feature extraction: Within each sliding window, calculate a set of features. Several core physiological characteristics include: time-domain standard deviation of heart rate variability. Skin conductance response slope Mean amplitude of electromyographic activity respiratory rate and body mobility calculated based on acceleration modulus .

[0080] Feature normalization: for each dimension, the newly calculated real-time value Z-score normalization is performed using the user's personal baseline parameters to obtain standardized feature values. :

[0081] ;

[0082] in, Let be the real-time value of the i-th feature. and The individual baseline mean and standard deviation are calculated based on data from when users self-labeled themselves as relaxed during the calibration period; Indicates the first A physiological characteristic at any time The standardized characteristic values ​​are calculated as follows: During the system initialization calibration period, the system prompts the user to select their current subjective state (e.g., relaxed, neutral, stressed) at random or fixed time intervals via a paired mobile terminal application. The system only collects and stores all user data in the self-labeled 'relaxed' or 'resting' state. The data were used to calculate the individual baseline mean and standard deviation, respectively. and The initial value.

[0083] Index synthesis: The normalized features are weighted and summed to obtain a real-time personalized stress index. : The personalized stress index It is a dimensionless scalar. Due to its standardized eigenvalues... After Z-score normalization based on individual baselines, This indicates that the user's physiological state is at their personal relaxation baseline level; This indicates that physiological arousal is higher than baseline; the higher the value, the stronger the stress response. This indicates a more relaxed weighting than the baseline. Set to equal values ​​during system initialization, i.e. ,in The total number of features.

[0084] As an optional implementation, the system can use standardized feature values ​​that users mark as effective interventions in subsequent use. The range of change is dynamically adjusted, and the weighting coefficients are fine-tuned using the following formula: Subsequently, for all Normalization is performed to ensure ,in The learning rate;

[0085] The updated version Each feature weight coefficient;

[0086] : The first one before the update Each feature weight coefficient;

[0087] : The first day before the intervention began Normalized values ​​of each feature;

[0088] : The day after the intervention ended The normalized values ​​of each feature.

[0089] This optimization method makes the stress index model more closely match the individual physiological response characteristics of users.

[0090] Pressure level determination is performed by a pressure level determination module deployed in a local edge computing unit or a central collaborative controller. The pressure level determination module receives real-time data from the pressure index calculation module. The value is compared with a set of four personalized dynamic thresholds. satisfy Compare and output discrete real-time pressure levels. :

[0091] (No pressure), when ;

[0092] (Mild tension), when ;

[0093] (Moderate anxiety), when ;

[0094] (High stress), when ;

[0095] (Crisis Critical) When .

[0096] Initial threshold to Based on the calibration period, fixed weighting coefficients are used. and the individual baseline mean calculated based on data from when users self-labeled themselves as relaxed during the calibration period. and standard deviation All the calculated pressure indices The sample set of values ​​is automatically set. Specifically, the sample set is sorted in ascending order of values, and the 70th, 85th, 92nd, and 98th percentiles are calculated using linear interpolation. The calculation results are then assigned to [the appropriate values]. .

[0097] The central collaborative controller continuously receives personalized stress indices from the feature calculation and rating determination module. and current pressure level And simultaneously obtain the current context label output by the environment context-aware interface. .

[0098] The environmental context awareness interface consists of a Global Navigation Satellite System (GNSS) receiver, an ambient light intensity sensor, a microphone array, and a calendar event parser. The GNSS receiver, combined with geofencing technology, determines whether the user's current location falls within a predefined area type, including workplaces, living spaces, public transportation, and open outdoor areas. The ambient light intensity sensor outputs illuminance values, which are converted into lighting comfort levels via nonlinear mapping. The microphone array uses a four-element linear array layout and employs delayed summation beamforming technology to directionally pick up speech signals within a 60-degree cone area in front of the user, inputting them into a pre-trained, locally deployed lightweight audio event detection model. This model takes the Mel-spectrum graph of an audio segment as input and outputs a probability distribution of predefined semantic labels (such as meeting, argument, alone time), using the label with the highest probability as the recognition result. This model uses model compression techniques (such as knowledge distillation and quantization) to adapt to the computational resource constraints of edge devices. The calendar event parser accesses the user's schedule with operating system permissions, extracts keywords from event titles, and maps them to standardized context categories, such as high-intensity meetings, family gatherings, or commuting times.

[0099] All contextual information undergoes differential privacy perturbation processing before entering the central collaborative controller, with the perturbation amplitude set according to the information sensitivity level. Specifically, the location information addition follows a Laplace distribution. Noise, scale parameters The settings can be customized according to user privacy preferences; the default value is 50 meters. Scheduled event categories are categorized by probability. Perform independent and identically distributed random flipping.

[0100] Context fusion: Processed multi-source context information is input to the context fusion module. The context fusion module generates unique, discrete current context labels based on preset priority rules. .

[0101] The priority rules are set as follows:

[0102] Absolute Priority (Schedule): If the standardized event category label parsed from the user's schedule is not empty and the confidence level is higher than a preset threshold (e.g., confidence level > 0.8 parsed by a natural language processing model), then the label is directly used as the current context label. .

[0103] Priority of Conditions (Location and Audio): When there is no high-confidence schedule tag, the fusion result of the location tag and audio tag is used. The fusion rule is as follows: if the location tag (e.g., office) and the audio tag (e.g., meeting) are semantically consistent or compatible, the common tag is used; if the two conflict (e.g., location gym and audio conference), the decision is made based on the time context: if the current time belongs to a work time event parsed from the user's schedule, or is within a preset work time period, the location tag is used first; otherwise, the audio tag is used first.

[0104] Ambient light intensity data is not directly involved in label generation, but is only used for the specific parameters of subsequent adjustment intervention actions. The mapping logic is as follows: when the ambient light intensity is ≥500 lux (high light environment), the volume output of audio actions is adaptively increased and the brightness of the mobile terminal APP interface is reduced to improve perceptibility; when the ambient light intensity is ≤100 lux (low light environment), the audio volume is adaptively reduced and the interface brightness is increased to avoid interference; when the light environment is moderate, the default parameters are maintained.

[0105] The triggering and process of intervention decision-making and execution are as follows:

[0106] The core operating mode of the system is a continuously running closed loop. The system updates the pressure index and level at a fixed sampling period (e.g., 1 second). The pressure level determination module outputs the real-time level... Reaching or exceeding the preset intervention trigger level (i.e. When this occurs, the central collaborative controller immediately initiates the following decision-making and execution processes:

[0107] Context filtering: Based on the current context label output by the environment context awareness module. (e.g., in the office, while driving, or resting at home), select all actions marked as permissible in the given context from a predefined intervention action library to form a candidate action set. .

[0108] Personalization and performance pre-filtering: Query dynamic user profiles and perform the following two filtering steps in sequence:

[0109] Hard disable filtering: from the candidate action set Remove actions that users have explicitly disabled manually.

[0110] Performance Elimination Filtering: Removes labels that are not present in the current context. Below, historical average efficiency value Below the preset performance threshold (For example Stress index (unit: per second). Historical average performance value. It can be positive, zero, or negative, representing that the intervention is effective, ineffective, or has the opposite effect, respectively. When the value is negative or zero, it is considered an inefficient action and is removed.

[0111] After the above filtering, the optimized candidate action set is obtained. .

[0112] Performance Ranking and Selection: Query the action-context performance table maintained in the dynamic user profile. This table records the historical average performance value of each action under different contexts experienced by the user. .

[0113] Performance value The calculation formula is: .in, The stress index at the start of the intervention. This is the value taken when the stress index stabilizes after the intervention ends. The duration of the intervention. A positive value indicates that the greater the pressure decreases per unit time, meaning that the action is more effective in the current situation.

[0114] The intervention is terminated under one of the following conditions, whichever comes first:

[0115] 1) The preset action duration has expired;

[0116] 2) The system detected a personalized stress index. It has entered a stable state, that is, continuous Within each sampling period The absolute values ​​of all changes are less than the preset stability threshold. ,For example , After that, the system began calculations. .

[0117] The system optimizes the candidate action set. In the middle, select the current context label. Historical performance value The highest action This was the first action to be taken in this operation.

[0118] Execution and Real-time Feedback Adjustment: Execution And start timing. Within the time window after execution begins. Inside (e.g.) (seconds), calculate the average rate of decrease of the stress index. ;in, For intervention implementation Stress index after a certain period of time.

[0119] Validity assessment and learning: If (in For example, the preset minimum effective descent rate threshold. If the stress index is expressed in units of per second, the intervention is considered effective, and the system will execute the current intervention. ) and its actual effectiveness value Record and use this information to update the average performance value of the corresponding item in the action-context performance table. .

[0120] Invalid response strategy: If This will trigger the strategy adjustment process:

[0121] Initial adjustment (switch): Immediately interrupt the current action and switch to the optimized candidate action set. Select historical performance value The second highest action Execute and reset the evaluation timer.

[0122] Secondary adjustment (enhancement): If the action after switching is within the time window Internal pressure change rate Still below the threshold Then revert to the original action. and its intensity parameters Add a fixed step size (For example, a 10% increase in audio volume or a 15% increase in tactile vibration intensity). There is an upper limit to the intensity increase. To prevent discomfort.

[0123] Termination condition: If continuous Next (for example) If adjustments (including switching and enhancements) are still ineffective, stop the current intervention cycle, record the failure information of all attempted actions in this situation, and avoid repeated triggering in the short term.

[0124] The final guidance instructions output by the intelligent guidance strategy generation engine are encapsulated as structured control commands and sent to the execution terminal via an AES-256 encrypted channel. The execution terminal includes a bone conduction audio player, a haptic feedback vest, an ambient lighting adjustment device, and a mobile terminal application interface. The bone conduction audio player plays binaural beat audio with specific frequency combinations according to instructions, with the frequency difference between the left and right ears set between 4Hz and 8Hz, corresponding to the theta brainwave frequency band, or plays guided breathing training voice, with the speaking speed controlled at a rhythm of six breaths per minute; the haptic feedback vest has sixteen built-in micro-vibration motors, symmetrically distributed on both sides of the spine on the back, and activates a specific area of ​​the motor array according to instructions to simulate deep pressure tactile stimulation, with the vibration frequency set at 30Hz, the amplitude at 70% of the maximum value, and the duration at two minutes; the ambient lighting adjustment device connects to smart lights via the Zigbee protocol, adjusting the color temperature to the range of 2700K to 3000K and reducing the illuminance to 50 to 100 lux; the mobile terminal application interface presents interactive cognitive reconstruction exercises, such as guiding users to identify and challenge automated negative thoughts, or initiating a mindfulness meditation guided process, including two stages: body scan and breath awareness.

[0125] The system in this embodiment runs continuously as a background service. Its core process is a closed-loop control loop based on real-time data streams, which includes the following continuous stages: Continuous sensing and calculation: The system synchronously performs multimodal physiological signal acquisition, preprocessing, feature calculation, and stress index calculation at a fixed frequency (e.g., 1Hz). With pressure level renew.

[0126] Decision trigger: Whenever stress level If the anxiety level is updated to 2 (moderate anxiety) or higher, the intervention decision-making process is triggered immediately.

[0127] Decision-making and execution: The system determines decisions based on the current stress index. ,grade Contextual tags In addition, dynamic user profiles are generated and executed based on the aforementioned steps of context filtering, execution, and real-time feedback adjustment. .

[0128] Concurrent monitoring and feedback learning: While performing actions, the system continues to periodically calculate... and Within the time window after the action begins. Internally, assess the intervention effect (i.e., calculate the rate of decrease in pressure). and with threshold Compare).

[0129] After the action is completed, the evaluation results (effective or ineffective) and the calculated performance value are used as the basis for the assessment. Update the action-scenario performance table. Afterward, the system immediately returns to the continuous sensing and computation phase, preparing to respond to the next escalation of stress levels. The system power management module dynamically reduces sensor sampling frequency or pauses non-core computing modules by recognizing user states such as prolonged inactivity or device idleness, achieving low-power operation rather than relying entirely on a multi-state machine.

[0130] System initialization, calibration, and adaptive learning: Personal baseline calibration period: Upon first use, the system enters a baseline calibration mode lasting at least 24 hours. In this mode, the system only collects and stores multimodal physiological signals, while prompting the user to input simple subjective state labels (such as relaxed, neutral, stressed) at random intervals.

[0131] After the calibration period, the system calculates various physiological characteristics based on the collected data. Individual baseline mean and standard deviation The pressure index sequence during the calibration period was calculated and used as the initial threshold. to The settings.

[0132] The dynamic user profile and performance table are initialized as follows:

[0133] User profiles store user static attributes, contextual preferences (such as disabling audio-related actions), and action-context performance tables.

[0134] Initial efficacy values ​​of all predefined intervention actions in all situations Set to the same default value (e.g.) ).

[0135] Online learning and updates:

[0136] Stress Index Baseline Update: Weekly, the system recalculates the baseline values ​​for each characteristic using physiological data from the past week when the user marked the user as relaxed. and To adapt to possible long-term physiological changes, the system adds a baseline version label to each record in the action-context performance table after the baseline update. When querying the average performance value in subsequent queries, a weighted mixed update is used, with the old baseline version record having a weight of [missing information]. The weight recorded in the new baseline version is .

[0137] At the same time, the system initiates a week-long parallel learning period, during which performance data under the new baseline will be given higher weight to accelerate knowledge transfer, reduce the influence weight of the old baseline data on current decisions, and prioritize the new performance data generated after the update.

[0138] Dynamic threshold adjustment: Every month, the system filters all stress index P(t) data that users have marked as relaxed in the past month, sorts them in ascending order of value, and recalculates their 70th, 85th, 92nd, and 98th percentiles, which are used as new candidate values ​​for T1, T2, T3, and T4, respectively. The updated suggestions are presented to users through the mobile terminal application interface and take effect after user confirmation.

[0139] Action-Context Efficacy Table Update: After each intervention action is completed, the system updates the average efficacy value of the corresponding item in the Action-Context Efficacy Table based on the actual effect of the intervention. The updated formula is: .in, Forgetting factor (range of values) ); This represents the updated historical average performance value; This represents the historical average performance value before the update. The efficacy value calculated based on this intervention, . The stress index at the start of the intervention. This represents the value of the stress index when it stabilizes after the intervention. The duration of the intervention. This mechanism allows the system to continuously adapt to the user's response characteristics to specific actions.

[0140] For example, if audio guidance fails to respond repeatedly in a meeting setting, the recommendation priority for this action in similar scenarios will be reduced. All calibration operations are performed locally, without involving the cloud uploading of users' raw physiological data, ensuring compliance with GDPR and China's Personal Information Protection Law regarding privacy.

[0141] A dual-buffer communication mechanism is used between the central co-controller and the edge computing units. The preprocessing engine and cross-modal feature extraction layer in the edge computing units are deployed on a dedicated neural network acceleration chip, which supports INT8 quantization inference with a single forward propagation latency of no more than fifteen milliseconds. The central co-controller runs on a low-power application processor core, and its operating system employs a real-time scheduling strategy, assigning the highest priority to tasks that determine stress levels, ensuring a response time strictly less than one hundred milliseconds. The system power management module dynamically adjusts the power supply to each component based on the current state; for example, it shuts down the microphone array and ambient light sensor in idle states, maintaining only the low-power monitoring mode of the inertial measurement unit.

[0142] Example 2: See Figure 3 This embodiment discloses a mental stress assessment and intelligent counseling system for implementing the mental stress assessment and intelligent counseling method described above, including:

[0143] The signal acquisition module, deployed in wearable devices, is used to continuously acquire multimodal physiological signals;

[0144] The edge computing unit, connected to the signal acquisition module, is used for preprocessing physiological signals, feature extraction, and personalized stress index calculation.

[0145] The central collaborative controller, connected to the edge computing unit, is used to execute stress level determination, context fusion, intervention decision-making, and learning update logic;

[0146] A context-aware interface, connected to the central collaborative controller, is used to acquire environmental and schedule information; an execution terminal, connected to the central collaborative controller, is used to execute intervention actions.

[0147] The edge computing unit and the central collaborative controller form an edge-central collaborative processing architecture. The central collaborative controller assigns the highest priority to the pressure level determination task, ensuring that the response time from the pressure level exceeding the threshold to the start decision is less than 100 milliseconds.

[0148] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0149] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. It should be noted that any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for assessing and intelligently managing mental stress, characterized in that, Includes the following steps: S1, Signal Acquisition and Personalized Modeling: Multimodal physiological signals of users are continuously acquired through wearable devices. After preprocessing and feature extraction, the signals are normalized based on the user's personal physiological baseline and then weighted and fused to generate a real-time personalized stress index. S2, Real-time status determination and triggering: The personalized stress index is compared with a set of personalized dynamic thresholds and mapped to discrete stress levels; When the pressure level reaches the preset intervention trigger level, the intervention decision process is initiated; S3, Context Awareness and Fusion: Acquires environmental and schedule information in parallel, processes it for privacy protection, and then fuses it according to preset priority rules to generate the current context label; S4, Intelligent Decision Making and Action Selection: Based on the current context label, the predefined intervention action library is filtered for context; combined with user preferences in the dynamic user profile and historical intervention effectiveness data, effectiveness filtering and ranking are performed to select the optimal intervention action; S5, Closed-loop execution and feedback learning: Execute the optimal intervention action and monitor the changes in the personalized stress index in real time after execution; evaluate the intervention effect based on the stress change rate, adjust the intervention strategy in real time according to the evaluation results, and update the historical intervention effectiveness data to achieve self-optimization.

2. The method for assessing and intelligently managing mental stress according to claim 1, characterized in that, In step S1, generating a real-time personalized stress index specifically includes: Extract physiological features from multimodal physiological signals, including at least heart rate variability, skin conductance, electromyographic activity, respiratory rate, and range of motion. Use formula Personalized Z-score normalization was performed on each physiological characteristic, among which, Let be the real-time value of the i-th feature. and The individual baseline mean and standard deviation are calculated based on data from when users self-labeled themselves as relaxed during the calibration period; Indicates the first A physiological characteristic at any time Standardized eigenvalues; The normalized features are then summed using a weighted average. The personalized stress index is obtained. ,in, The total number of physiological characteristics; Indicates the first Weight coefficients for each feature; weight coefficients Based on preset rule settings.

3. The method for assessing and intelligently managing mental stress according to claim 1, characterized in that, In step S2, the personalized dynamic threshold is determined in the following way: During the calibration period, a series of stress index sample values ​​were calculated using fixed weighting coefficients and individual baseline parameters; The sample values ​​are sorted in ascending order, and their 70th, 85th, 92nd, and 98th percentiles are calculated and assigned to four threshold values ​​respectively. ,satisfy ; By comparing personalized stress indices The pressure level will be determined by the threshold value. The value is determined to be a discrete value between 0 and 4, where Intervention is triggered at the appropriate time.

4. The method for assessing and intelligently managing mental stress according to claim 1, characterized in that, In step S3, the generation of the context label depends on the fusion of at least two of the following information sources, and is performed in the following priority order: Highest priority: Standardized event category tags parsed from user calendar events; Second highest priority: Predefined location labels determined by geofencing technology based on user location information; Priority Supplement: Semantic tags identified from environmental sounds using a local audio event detection model.

5. The method for assessing and intelligently managing mental stress according to claim 4, characterized in that, The user location information and calendar event category information are processed with differential privacy perturbation before fusion, wherein the location information addition follows a Laplace distribution. Noise, scale parameters The accuracy of the geographical range of the required protected location is set; the schedule event categories are set by probability. Perform random flipping, that is, in each processing cycle, with probability... Replace the true category with a value randomly selected from the other possible categories.

6. The method for assessing and intelligently managing mental stress according to claim 1, characterized in that, In step S4, the performance filtering and ranking specifically includes: Query the action-context efficacy table maintained in the dynamic user profile. This table records the historical average efficacy value of each intervention action for users in different contexts. ; The historical average performance value The calculation method is as follows: ,in The stress index at the start of the intervention. The value to be taken when the pressure index reaches a stable state after the intervention ends, and the stable state is determined by predefined numerical conditions. The duration of the intervention; From the candidate actions filtered by the context, actions with historical average effectiveness values ​​lower than a preset effectiveness threshold are removed, and from the remaining actions, the action with the highest historical average effectiveness value in the current context is selected as the optimal intervention action.

7. The method for assessing and intelligently managing mental stress according to claim 2, characterized in that, In step S5, the assessment of intervention effectiveness and strategy update based on the rate of change of pressure includes: Preset time window after performing the selected intervention action Internally, the time of intervention start is used as Calculate the average rate of decrease of the pressure index. ; in, For intervention implementation Stress index after a period of time; like ,in If the minimum effective rate of descent is a preset threshold, the intervention is deemed effective, and the formula is used. Update the historical average efficacy value of the corresponding item in the action-situation efficacy table, where Forgetting factor, This is the efficacy value calculated for this intervention; This represents the updated historical average performance value; This represents the historical average performance value before the update; like If the intervention fails to achieve the expected results, the action switching or intensity enhancement strategy should be implemented immediately.

8. The method for assessing and intelligently managing mental stress according to claim 1, characterized in that, The method also includes a multi-timescale system adaptive learning step: Short-term learning: After each intervention, update the action-situation efficacy scale; Chung-Shi Learning: Regularly recalculates individual physiological baseline parameters using newly tagged relaxation state data from users. and After the baseline is updated, perform at least one of the following operations on the action-context performance table: (a) All intervention events recorded under the old baseline conditions are marked as historical data, and a weighted mixed update formula is used when calculating the new historical average efficacy value: ； in This represents the average performance value under the old baseline. This represents the average performance value under the new baseline. The weighting coefficients for historical data satisfy the following conditions: ; Or (b) clear the performance table and restart recording and learning based on the new baseline; Long-term learning: Regularly recalculate the stress level classification threshold based on the user's newly generated relaxation state stress index data, generate updated suggestions, and take effect after user confirmation.

9. The method for assessing and intelligently managing mental stress according to claim 1, characterized in that, The predefined intervention action library includes at least one of the following types of actions: binaural beat or guided breathing training played through bone conduction audio, deep pressure stimulation simulated through a haptic feedback device, color temperature and illuminance adjusted through an ambient lighting adjustment device, and interactive cognitive reconstruction or mindfulness meditation guidance presented on a mobile terminal application interface.

10. A mental stress assessment and intelligent counseling system, used to implement the mental stress assessment and intelligent counseling method as described in any one of claims 1-9, characterized in that, include: The signal acquisition module, deployed in wearable devices, is used to continuously acquire multimodal physiological signals; The edge computing unit, connected to the signal acquisition module, is used for preprocessing physiological signals, feature extraction, and personalized stress index calculation. The central collaborative controller, connected to the edge computing unit, is used to execute stress level determination, context fusion, intervention decision-making, and learning update logic; A context-aware interface, connected to the central collaborative controller, is used to acquire environmental and schedule information; An execution terminal, connected to the central collaborative controller, is used to execute intervention actions; The edge computing unit and the central collaborative controller form an edge-central collaborative processing architecture. The central collaborative controller assigns the highest priority to the pressure level determination task, ensuring that the response time from the pressure level exceeding the threshold to the start decision is less than 100 milliseconds.

Images