Construction and dynamic intervention system of mental health digital twinborn body

Abstract

The invention relates to the technical field of psychological health service and digital twinning, and discloses a psychological health digital twinning construction and dynamic intervention system, which comprises a multi-modal data acquisition module, a psychological digital twinning modeling module, a psychological state prediction module, a dynamic intervention decision module and a feedback optimization module, each module realizes real-time data interaction through a distributed data bus; wherein the multi-modal data acquisition module is responsible for acquiring individual physiological, psychological, behavior and environment multi-dimensional data, and the psychological digital twin modeling module constructs a dynamic virtual mapping body of an individual psychological state based on the data. According to the method, real-time monitoring of the psychological state is realized through multi-modal data acquisition and twinborn dynamic mapping, and the hysteresis of traditional evaluation is overcome; on the basis of a two-stage matching algorithm, interference strategies are screened in combination with user static characteristics and dynamic states, and limitation of a homogenization scheme is avoided.

Description

Technical Field

[0001] The present invention relates to the field of mental health services and digital twin technology, and specifically to a mental health digital twin construction and dynamic intervention system. Background Art

[0002] Mental health issues have become a major challenge in the field of global public health. Traditional mental health services have many limitations: first, evaluation lags, relying on users to actively seek help and regular scale assessments, making it difficult to capture changes in mental state in real time (such as sudden increases in anxiety that are not discovered in time); second, intervention homogeneity, intervention plans based on group data cannot adapt to differences in individual psychological characteristics (such as introverted and extroverted users have completely different needs for social intervention); third, the lack of a feedback loop, the lack of accurate data support for intervention effect evaluation and program adjustment, resulting in low intervention efficiency (such as users who do not respond well to cognitive reconstruction technology still repeatedly use the same program).

[0003] In existing technologies, health digital twins mostly focus on physiological indicators (such as heart rate and blood pressure), without deeply integrating psychological characteristic data (such as cognitive patterns, emotional fluctuations, and behavioral habits), and are unable to fully map the physiological-psychological interaction relationship; there is a lack of dynamic modeling methods for the field of mental health, and it is impossible to simulate the evolution of mental states; the intervention mechanism is disconnected from the twin model; in view of this, we propose a mental health digital twin construction and dynamic intervention system. Summary of the Invention

[0004] The purpose of the present invention is to provide a mental health digital twin construction and dynamic intervention system to solve the problems raised in the above background technology.

[0005] To achieve the above-mentioned objectives, the present invention provides the following technical solutions: a system for constructing and dynamically intervening in a mental health digital twin, comprising a multimodal data acquisition module, a psychological digital twin modeling module, a psychological state prediction module, a dynamic intervention decision module and a feedback optimization module, wherein each module realizes real-time data interaction through a distributed data bus; wherein, the multimodal data acquisition module is responsible for acquiring multidimensional data of individual physiology, psychology, behavior and environment, the psychological digital twin modeling module constructs a dynamic virtual mapping body of the individual's psychological state based on the data, the psychological state prediction module simulates the evolution trend of the psychological state through the twin, the dynamic intervention decision module generates a personalized intervention plan based on the real-time state of the twin and the prediction results, and the feedback optimization module iteratively optimizes the system model and intervention strategy based on the intervention effect data.

[0006] Optionally, the multimodal data acquisition module includes:

[0007] Physiological data unit module: equipped with a wearable sensor device interface, supports the collection of heart rate variability (HRV), galvanic skin response (GSR), body temperature, respiratory rate, sleep structure and blood oxygen saturation (SpO2). The sampling frequency is 100Hz-500Hz for physiological signals and 1Hz for sleep data. The data is encrypted and stored after pre-processing (including filtering, denoising, and baseline correction);

[0008] Sleep structure includes sleep onset latency, deep sleep ratio, and awakening times; data preprocessing includes filtering, denoising, and baseline correction;

[0009] Psychological Assessment Unit: This module integrates a digital scale library, including depression assessment (PHQ-9, SDS), anxiety assessment (GAD-7, SAS), personality traits (Big Five Personality Inventory), cognitive function (Digit Span Test), and social function (Social Adaptation Scale). It supports three assessment modes: timed trigger (fixed time period daily), event trigger, and user-initiated. Scale data is automatically associated with timestamps and context tags.

[0010] Behavioral feature unit module: This module accesses the user's digital behavior data interface through authorization, and collects content including social media texts, app usage records, geographic location trajectories, voice interaction records, and motion data. The behavioral data is desensitized and then feature extracted.

[0011] Social media text includes word frequency and emotional tendencies; app usage records include daily usage duration and frequency of social, entertainment, and utility apps; geographic location trajectories include stop type and movement speed; voice interaction records include tone, speech speed, and the proportion of emotional words; exercise data includes step count, activity intensity, and exercise type;

[0012] Environmental data unit module: Obtain environmental parameters through IoT devices or third-party APIs, including light intensity (lux), ambient noise (dB), temperature (°C), humidity (%), air quality (AQI) and social environment characteristics. Environmental data is aligned with user behavior data in time and space and used as input for factors affecting psychological state. Social environment characteristics include estimated population density in homes, offices, and public places.

[0013] Optionally, the psychological digital twin modeling module includes:

[0014] Static feature modeling unit module: Extracts user demographic information, past psychological history, personality traits, and genetic risk factors to build a static feature vector library as the basic parameters of the twin. The parameter update cycle is every 3 months or when the user actively updates. The extracted user demographic information includes age, gender, education level, and occupation type; past psychological history includes diagnosis records, treatment history, and medication history; personality traits include scores on dimensions such as neuroticism and extroversion;

[0015] Dynamic state mapping unit module: uses a multimodal data fusion algorithm (deep neural network with attention mechanism) to map physiological, psychological, behavioral and environmental data into a four-dimensional dynamic model of "emotion-cognition-behavior-social function":

[0016] Emotional dimension module: quantified into three dimensions: pleasure (-10 to +10), arousal (0 to 10), and dominance (-10 to +10), and weighted calculation is performed by combining physiological signals (such as the ratio of low-frequency to high-frequency HRV to reflect arousal) with text sentiment analysis results;

[0017] Cognitive dimension module: includes cognitive bias types (12 categories such as black-and-white and catastrophizing), automatic thinking frequency (times, days), attention concentration (0 to 100%), and problem-solving ability score (0 to 100 points);

[0018] Behavioral dimension module: covers activity level (daily activity entropy), frequency of social interactions (number of face-to-face and online social interactions), sleep behavior (regularity index) and incidence of impulsive behavior;

[0019] Social function dimension module: evaluates work and study performance (fluctuations in task completion), interpersonal relationship quality (frequency of conflicts, frequency of support acquisition), and daily self-care ability;

[0020] Twin calibration unit module: uses a dynamic weight adjustment mechanism to calculate the deviation between the virtual psychological state output by the twin and the user's actual evaluation results every 24 hours (using the root mean square error RMSE). When the deviation value is greater than 15%, the weight coefficient of each input data is automatically adjusted. It also supports manual calibration by users (using the "My State Correction" function to input actual feelings). After calibration, the model parameters are updated in real time and the calibration log is recorded.

[0021] Optionally, the mental state prediction module includes:

[0022] Short-term prediction unit module: Deploys a time series prediction model (an improved LSTM combined with an attention mechanism), inputs the twin dynamic data of the past 72 hours, and outputs a four-dimensional state prediction curve for the next 48 hours, including the peak and valley time points of emotional fluctuations, the trend of changes in the intensity of cognitive biases, and the probability of behavioral pattern changes. The prediction results are expressed as a 95% confidence interval;

[0023] Crisis Warning Unit Module: Multiple crisis thresholds are preset. Level 1 warning (concern level) is when a depression score is ≥6 or anxiety score is ≥7 and lasts for 6 hours. Level 2 warning (intervention level) is when a depression score is ≥8 and accompanied by text and voice features related to self-harm. Level 3 warning (emergency level) is when a level 2 warning state lasts for 12 hours and behavioral features such as solitude and a sudden drop in activity are present. When a warning is triggered, the system automatically pushes a warning message to pre-set contacts (including trusted relatives and friends, and contracted counselors). The information includes the warning level, current status parameters, recent behavioral changes, and recommended response measures.

[0024] Long-term trend analysis unit module: Analyze the twin data of the past three months through time series clustering algorithms (such as DBSCAN), identify the periodic patterns of psychological states (such as anxiety peaks every Monday and seasonal depression tendencies), extract trend characteristic parameters (cycle length, amplitude, phase), generate trend reports seven days in advance, and mark high-risk periods and possible triggering factors.

[0025] Optionally, the dynamic intervention decision module includes:

[0026] Intervention Strategy Library Module;

[0027] Emotional Regulation Modules: Breathing exercises (4-7-8 breathing method, box breathing), progressive muscle relaxation, mindfulness meditation (including guided audio and video, duration 5-30 minutes), and emotional visual diary (supporting text, voice, and image input);

[0028] Cognitive Restructuring Module: Automatic Thought Recording Sheet (guided by the three steps of identification, questioning, and substitution), Cognitive Bias Correction Training (interactive exercises targeting specific bias types), and Socratic Questioning Guidance.

[0029] Behavior Activation Module: Graded social tasks (from "brief greeting one person" to "attending a small gathering"), enjoyable activity planning (recommending activities based on user historical preferences), and sleep and rest adjustment plans (sleep hygiene education + sleep and rest supervision);

[0030] Matching algorithm unit modules: Light intervention recommendations (e.g., seasonal depression recommends 30 minutes of 10,000 lux light daily), noise control solutions (recommending noise reduction equipment or quiet places);

[0031] Each intervention technique is associated with the characteristics of the applicable population (e.g., relaxation techniques are suitable for people with high neuroticism), expected effects (magnitude of mood improvement, onset time), and contraindications (e.g., certain breathing exercises are not suitable for people with asthma);

[0032] Execution Algorithm Unit Module: Executes a two-stage matching logic. The first stage screens intervention technology types based on static user characteristics (such as personality trait scores and previous intervention responses). The second stage determines the specific intervention plan based on the current dynamic state (such as the dominant emotion type and cognitive bias type).

[0033] Intervention execution unit module: supports multi-channel and multi-format intervention push, the APP provides graphic, video guidance and interactive tasks (such as instant feedback after completing the cognitive record form), the smart hardware end (speakers, bracelets) provides voice guidance and synchronous monitoring of physiological signals (such as real-time display of heart rate changes during breathing training), and SMS and WeChat terminals send concise reminders; during the intervention process, it supports real-time adjustment of difficulty and records task completion (start rate, completion rate, quality score).

[0034] Optionally, the feedback optimization module includes:

[0035] Effect evaluation unit module: Establish a multi-dimensional evaluation system. 1 hour after the intervention, collect changes in physiological indicators (such as HRV increase and GSR decrease rate), 24 hours after the intervention, collect changes in psychological assessment data (difference in target dimension scores), 72 hours after the intervention, analyze changes in behavioral patterns (such as changes in the frequency of social interactions), and comprehensively calculate the intervention effect score (0-100 points). A score of 70 or more is considered effective, and 85 or more is considered excellent.

[0036] Strategy Iteration Unit: This module uses a reinforcement learning framework and uses intervention effectiveness scores as reward signals. When an intervention technique's effectiveness score is less than 60 points for three consecutive times, an alternative technique (the one with the highest correlation based on cosine similarity) is retrieved from the strategy library. The reason for the ineffectiveness is also recorded (e.g., "user resists the text task" or "inappropriate timing"). For the most effective techniques, their weight in the matching algorithm is increased, and implementation parameters are optimized.

[0037] System upgrade unit module: Global optimization is performed monthly. Based on the anonymized data of all users (individual identifiers are removed), the twin modeling algorithm (to improve prediction accuracy) and the intervention matching model (to improve efficiency) are updated using federated learning. User personalized parameters are retained during the upgrade process to ensure that system performance iteration is compatible with individual adaptability.

[0038] Optionally, the preprocessing process of the physiological data unit module includes:

[0039] Noise filtering: Wavelet transform is used to remove motion artifacts and electromagnetic interference;

[0040] Feature extraction: extract time domain features (mean, standard deviation, number of peaks), frequency domain features (energy ratio of low frequency band and high frequency band) and time-frequency domain features (wavelet entropy) from the original signal;

[0041] Anomaly detection: Identify abnormal physiological events (such as a sudden increase in heart rate >20 beats / minute) using the 3σ criterion, mark the time of occurrence and duration of the event, and correlate the environmental and behavioral data for the same period.

[0042] Optionally, the two-stage matching logic of the matching algorithm unit module is specifically as follows:

[0043] Phase 1 (Type Screening): Input the user's static feature vector (including age, personality trait score, and previous intervention response history) and use the random forest classification model to predict the response probability to interventions in the emotion regulation module, cognitive restructuring module, and behavior activation module. Select technology types with a response probability ≥ 60%;

[0044] Phase 2 (Parameter Optimization): For the selected technology types, intervention parameters are optimized based on current dynamic state parameters (such as emotional arousal and cognitive load). For example, for users with high arousal anxiety, the breathing exercise rhythm is adjusted to "inhale for 4 seconds - hold for 7 seconds - exhale for 8 seconds" and the baseline duration is set at 5 minutes. For users with high cognitive load, the step description of the cognitive restructuring task is simplified and examples are added to guide the user.

[0045] Matching constraints: exclude technologies that conflict with user contraindications (e.g., prohibiting high-intensity breathing training for users with a history of cardiovascular disease), and give priority to technology types with a historical user completion rate greater than 70% to ensure the safety and feasibility of the intervention plan.

[0046] Compared with the existing technology, the present invention provides a system for constructing and dynamically intervening in a mental health digital twin, which has the following beneficial effects:

[0047] 1. The construction and dynamic intervention system of the mental health digital twin realizes real-time monitoring of mental state through multimodal data collection and twin dynamic mapping, overcoming the lag of traditional assessment. Based on a two-stage matching algorithm, it combines the user's static characteristics with the dynamic state to screen intervention strategies, avoiding the limitations of homogenized solutions.

[0048] 2. The construction and dynamic intervention system of the mental health digital twin establishes a complete closed loop of data collection-twin mapping-prediction-intervention-feedback. Through reinforcement learning, the system is continuously optimized, and the intervention efficiency is improved by more than 30%. The three-level early warning mechanism can predict psychological crises 48 hours in advance, and combined with multi-channel push, it reduces the risk of extreme events. DETAILED DESCRIPTION

[0049] The following will be combined with the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0050] The present invention provides a technical solution: a mental health digital twin construction and dynamic intervention system, including a multimodal data acquisition module, a psychological digital twin modeling module, a psychological state prediction module, a dynamic intervention decision module and a feedback optimization module, each module realizes real-time data interaction through a distributed data bus; among them, the multimodal data acquisition module is responsible for acquiring individual physiological, psychological, behavioral and environmental multidimensional data, the psychological digital twin modeling module constructs a dynamic virtual mapping body of the individual's psychological state based on the data, the psychological state prediction module simulates the evolution trend of the psychological state through the twin, the dynamic intervention decision module generates a personalized intervention plan according to the real-time state of the twin and the prediction results, and the feedback optimization module iteratively optimizes the system model and intervention strategy based on the intervention effect data.

[0051] The multimodal data acquisition module includes: physiological data unit module, psychological assessment unit module, behavioral characteristics unit module and environmental data unit module.

[0052] The physiological data unit module is equipped with a wearable sensor device interface and supports the collection of heart rate variability (HRV), galvanic skin response (GSR), body temperature, respiratory rate, sleep structure (including sleep onset latency, deep sleep ratio, and number of awakenings), and blood oxygen saturation (SpO2). The sampling frequency is 100Hz-500Hz for physiological signals and 1Hz for sleep data. The data is encrypted and stored after pre-processing (including filtering, denoising, and baseline correction). The psychological assessment unit module integrates a digital scale library, including depression assessment (PHQ-9, SDS), anxiety assessment (GAD-7, SAS), personality traits (Big Five Personality Scale), cognitive function (Digit Span Test), and social function (Social Adaptation Scale). It supports three assessment modes: timed triggering (fixed time of day), event triggering (such as when abnormal physiological signals are detected), and user-initiated. The scale data is automatically associated with timestamps and context tags.

[0053] The behavioral feature unit module accesses the user's digital behavior data interface through authorization, and collects content including social media text (word frequency, emotional tendency), APP usage records (daily usage time and frequency of social, entertainment, and tool apps), geographic location trajectory (stop point type, movement speed), voice interaction records (tone, speaking speed, proportion of emotional words) and exercise data (number of steps, activity intensity, exercise type). The behavioral data is desensitized and then feature extracted; the environmental data unit module obtains environmental parameters through IoT devices or third-party APIs, including light intensity (lux), ambient noise (dB), temperature (℃), humidity (%), air quality (AQI) and social environment characteristics (such as population density estimation at home, office, and public places). The environmental data is aligned with the user behavior data in time and space and then input as factors affecting the psychological state.

[0054] The psychological digital twin modeling module includes: static feature modeling unit module, dynamic state mapping unit module, emotion dimension module, cognitive dimension module, behavior dimension module, social function dimension module and twin calibration unit module.

[0055] The static feature modeling unit module extracts user demographic information (age, gender, education level, occupation type), past psychological history (diagnosis records, treatment history, medication history), personality traits (scores of dimensions such as neuroticism and extroversion), and genetic risk factors (family history of mental illness) to construct a static feature vector library as the basic parameters of the twin. The parameter update cycle is every 3 months or when the user actively updates. The dynamic state mapping unit module uses a multimodal data fusion algorithm (deep neural network with attention mechanism) to map physiological, psychological, behavioral and environmental data into a four-dimensional dynamic model of "emotion-cognition-behavior-social function":

[0056] The emotional dimension module is quantified into three dimensions: pleasure (-10 to +10), arousal (0 to 10), and dominance (-10 to +10), and is weighted and calculated by combining physiological signals (such as the ratio of low-frequency to high-frequency HRV to reflect arousal) with the results of text sentiment analysis. The cognitive dimension module includes cognitive bias types (12 categories such as black-and-white and catastrophizing), automatic thinking frequency (times, days), attention concentration (0 to 100%), and problem-solving ability score (0 to 100 points). The behavioral dimension module covers activity level (daily activity entropy), frequency of social interactions (number of face-to-face and online social interactions), sleep behavior (regularity index), and the incidence of impulsive behavior. The social function dimension module: evaluates work and learning performance (fluctuations in task completion), interpersonal relationship quality (frequency of conflicts, frequency of support acquisition), and daily self-care ability.

[0057] The twin calibration unit module adopts a dynamic weight adjustment mechanism to calculate the deviation between the virtual psychological state output by the twin and the actual evaluation results of the user every 24 hours (using the root mean square error RMSE). When the deviation value is greater than 15%, the weight coefficient of each input data is automatically adjusted. At the same time, it supports manual calibration by users (entering actual feelings through the "My State Correction" function). After calibration, the model parameters are updated in real time and the calibration log is recorded.

[0058] Among them, the psychological state prediction module includes: a short-term prediction unit module, a crisis warning unit module and a long-term trend analysis unit module.

[0059] The short-term prediction unit module deploys a time series prediction model (an improved LSTM combined with an attention mechanism), inputs the twin dynamic data of the past 72 hours, and outputs a four-dimensional state prediction curve for the next 48 hours, including the peak and trough time points of mood fluctuations, the trend of changes in the intensity of cognitive biases, and the probability of behavioral pattern changes. The prediction results are expressed as a 95% confidence interval; the crisis warning unit module presets multiple crisis thresholds. The first-level warning (attention level) is a depression dimension score ≥6 points or an anxiety dimension score ≥7 points and lasts for 6 hours. The second-level warning (intervention level) is a depression score ≥8 points accompanied by self-harm-related text and voice features. The third-level warning (emergency level) is a second-level warning state that lasts for 12 hours and has behavioral characteristics such as solitude and a sudden drop in activity. When the warning is triggered, the system automatically pushes the warning information to the preset contacts (including the user's trusted relatives and friends, and contracted consultants). The information content includes the warning level, current state parameters, recent behavioral changes, and recommended response measures.

[0060] The long-term trend analysis unit module uses time series clustering algorithms (such as DBSCAN) to analyze twin data from the past three months, identify periodic patterns in psychological states (such as anxiety peaks every Monday and seasonal depression tendencies), extract trend characteristic parameters (cycle length, amplitude, phase), and generate trend reports seven days in advance, marking high-risk periods and possible triggering factors.

[0061] The dynamic intervention decision module includes: intervention strategy library module, emotion regulation module, cognitive reconstruction module, behavior activation module, matching algorithm unit module and intervention execution unit module.

[0062] Among them, the emotion regulation module includes breathing training (4-7-8 breathing method, box breathing), progressive muscle relaxation, mindfulness meditation (including guided audio and video, duration 5 minutes to 30 minutes), and emotional visualization diary (supporting text, voice, and image input); the cognitive reconstruction module includes automatic thought record sheet (three-step guidance of identification-questioning-substitution), cognitive bias correction training (interactive exercises for specific types of bias), and Socratic questioning guidance; the behavioral activation module includes graded social tasks (from "brief greeting with 1 person" to "participating in a small gathering"), pleasant activity planning (recommended activities based on user historical preferences), and work and rest adjustment plans (sleep hygiene education + work and rest supervision); the matching algorithm unit module includes light intervention recommendations (such as seasonal depression recommends 10,000 lux light for 30 minutes per day) and noise control plans (recommended noise reduction equipment or quiet places); each intervention technology is associated with the characteristics of the applicable population (such as relaxation techniques suitable for people with high neuroticism), expected effects (the extent of mood improvement and onset time), and contraindications (such as certain breathing exercises are not suitable for asthma patients);

[0063] The execution algorithm unit module implements a two-stage matching logic. The first stage screens the type of intervention technology based on the user's static characteristics (such as personality trait scores and previous intervention responses). The second stage determines the specific intervention plan based on the current dynamic state (such as the type of emotion dominance and the type of cognitive bias). The intervention execution unit module supports multi-channel and multi-form intervention push. The APP side provides graphic, video guidance and interactive tasks (such as instant feedback after completing the cognitive record form). The smart hardware side (speakers, bracelets) provides voice guidance and synchronous monitoring of physiological signals (such as real-time display of heart rate changes during breathing training). SMS and WeChat sides send concise reminders. During the intervention process, it supports real-time adjustment of difficulty and records task completion (start rate, completion rate, quality score).

[0064] The feedback optimization module includes: effect evaluation unit module, strategy iteration unit module and system upgrade unit module.

[0065] The effectiveness evaluation unit module establishes a multi-dimensional evaluation system. It collects changes in physiological indicators (such as the increase in HRV and the rate of decrease in GSR) one hour after the intervention, collects changes in psychological assessment data (difference in target dimension scores) 24 hours later, and analyzes changes in behavioral patterns (such as changes in the frequency of social interactions) 72 hours later. The intervention effect score (0-100 points) is comprehensively calculated. A score ≥70 is considered effective, and a score ≥85 is considered excellent. The strategy iteration unit module adopts a reinforcement learning framework and uses the intervention effect score as a reward signal. When an intervention technology has an effective score of less than 60 points for three consecutive times, an alternative technology is retrieved from the strategy library (the technology with the highest correlation based on cosine similarity calculation), and the invalid reason label (such as "user resistance to text tasks" and "inappropriate timing") is recorded. For excellent technologies, their weight in the matching algorithm is increased, and implementation parameters are optimized. The system upgrade unit module performs global optimization every month. Based on the anonymized data of all users (removing individual identifiers), it uses federated learning to update the twin modeling algorithm (to improve prediction accuracy) and the intervention matching model (to improve efficiency). User personalized parameters are retained during the upgrade process to ensure that system performance iteration is compatible with individual adaptability.

[0066] It is worth noting that the preprocessing process of the physiological data unit module includes:

[0067] Noise filtering, using wavelet transform to remove motion artifacts and electromagnetic interference;

[0068] Feature extraction: extracting time domain features (mean, standard deviation, number of peaks), frequency domain features (energy ratio of low frequency band and high frequency band) and time-frequency domain features (wavelet entropy) from the original signal;

[0069] Anomaly detection uses the 3σ criterion to identify abnormal physiological events (such as a sudden increase in heart rate >20 beats / minute), mark the time of occurrence and duration of the event, and correlate the environmental and behavioral data for the same period.

[0070] The two-stage matching logic of the matching algorithm unit module is as follows:

[0071] In the first stage (type screening), the user's static feature vector (including age, personality trait score, and previous intervention response history) is input. The random forest classification model is used to predict the response probability to interventions in the emotion regulation module, cognitive reconstruction module, and behavior activation module. The technology types with a response probability ≥ 60% are screened.

[0072] In the second phase (parameter optimization), intervention parameters are optimized based on the current dynamic state parameters (such as emotional arousal and cognitive load) for the selected technology types. For example, for users with high arousal anxiety, the breathing exercise rhythm is adjusted to "inhale for 4 seconds - hold for 7 seconds - exhale for 8 seconds" and the basic duration is set to 5 minutes. For users with high cognitive load, the step description of the cognitive restructuring task is simplified and examples are added to guide the user.

[0073] Match the constraints, exclude technologies that conflict with the user's contraindications (such as prohibiting high-intensity breathing training for users with a history of cardiovascular disease), and give priority to technology types with a user's historical completion rate of >70% to ensure the safety and feasibility of the intervention plan.

[0074] As an application of this embodiment:

[0075] This embodiment achieves real-time monitoring of psychological state through multimodal data collection and twin dynamic mapping, overcoming the lag of traditional assessment; based on a two-stage matching algorithm, it combines user static features with dynamic state screening intervention strategies to avoid the limitations of homogenized solutions.

[0076] The above generally describes the present invention in detail. However, it is obvious to those skilled in the art that modifications or improvements may be made based on the present invention. Therefore, modifications or improvements that do not depart from the spirit of the present invention are within the scope of protection of the present invention.

Claims

1. A mental health digital twin construction and dynamic intervention system, characterized by: include: Multimodal data acquisition module, responsible for acquiring multidimensional data on individual physiology, psychology, behavior and environment; A psychological digital twin modeling module that constructs a dynamic virtual mapping of an individual's psychological state based on the data; The psychological state prediction module simulates the evolution trend of psychological state through twins; Dynamic intervention decision module generates personalized intervention plans based on the twin's real-time status and prediction results; Feedback optimization module iteratively optimizes system models and intervention strategies based on intervention effect data.

2. The mental health digital twin construction and dynamic intervention system according to claim 1 is characterized in that: The multimodal data acquisition module includes: Physiological data unit module: equipped with a wearable sensor device interface, it supports the collection of heart rate variability, galvanic skin response, body temperature, respiratory rate, sleep structure and blood oxygen saturation. The sampling frequency is 100Hz-500Hz for physiological signals and 1Hz for sleep data. The data is encrypted and stored after pre-processing. Psychological assessment unit module: integrated digital scale library, including depression assessment, anxiety assessment, personality traits, cognitive function and social function; Behavioral feature unit module: This module accesses the user's digital behavior data interface through authorization, and collects content including social media texts, app usage records, geographic location trajectories, voice interaction records, and motion data. The behavioral data is desensitized and then feature extracted. Environmental data unit module: Obtain environmental parameters through IoT devices or third-party APIs, including light intensity, ambient noise, temperature, humidity, air quality, and social environment characteristics. Environmental data is aligned with user behavior data in time and space and then used as input for factors affecting psychological state.

3. The mental health digital twin construction and dynamic intervention system according to claim 1 is characterized in that: The psychological digital twin modeling module includes: Static feature modeling unit module: extracts user demographic information, past psychological history, personality traits and genetic risk factors, and constructs a static feature vector library as the basic parameters of the twin. The parameter update cycle is every 3 months or when the user actively updates; Dynamic state mapping unit module: Using a multimodal data fusion algorithm, it maps physiological, psychological, behavioral, and environmental data into a four-dimensional dynamic model of "emotion-cognition-behavior-social function": Emotional dimension module: quantified into three dimensions: pleasure, arousal, and dominance, and weighted calculation based on physiological signals and text sentiment analysis results; Cognitive dimension module: includes cognitive bias types, automatic thinking frequency, attention concentration and problem-solving ability scores; Behavioral dimension module: covers activity level, frequency of social interaction, sleep behavior and incidence of impulsive behavior; Social function dimension module: assesses work and study performance, interpersonal relationship quality and daily self-care ability; Twin calibration unit module: adopts a dynamic weight adjustment mechanism to calculate the deviation between the virtual psychological state output by the twin and the actual evaluation results of the user every 24 hours. When the deviation value is greater than 15%, the weight coefficient of each input data is automatically adjusted. At the same time, it supports user manual calibration. After calibration, the model parameters are updated in real time and the calibration log is recorded.

4. The mental health digital twin construction and dynamic intervention system according to claim 1 is characterized in that: The mental state prediction module includes: Short-term prediction unit module: Deploys a time series prediction model, inputs the twin dynamic data of the past 72 hours, and outputs a four-dimensional state prediction curve for the next 48 hours, including the peak and valley time points of emotional fluctuations, the trend of changes in cognitive bias intensity, and the probability of behavioral pattern transition. The prediction results are expressed as a 95% confidence interval; Crisis Warning Unit Module: Multiple crisis thresholds are preset. Level 1 warning is when a depression score is ≥6 or anxiety score is ≥7 and lasts for 6 hours. Level 2 warning is when a depression score is ≥8 and accompanied by text or voice features related to self-harm. Level 3 warning is when the level 2 warning state lasts for 12 hours and behavioral features such as solitude and a sudden drop in activity are present. When a warning is triggered, the system automatically pushes a warning message to the preset contact, including the warning level, current status parameters, recent behavioral changes, and recommended response measures. Long-term trend analysis unit module: Analyzes the twin data of the past three months through time series clustering algorithm, identifies the periodic patterns of psychological states, extracts trend characteristic parameters, generates trend reports seven days in advance, and marks high-risk periods and possible triggering factors.

5. The mental health digital twin construction and dynamic intervention system according to claim 1 is characterized in that: The dynamic intervention decision module includes: Intervention Strategy Library Module; Emotional regulation modules: breathing training, progressive muscle relaxation, mindfulness meditation, and emotional visualization diary; Cognitive reconstruction modules: automatic thought record sheet, cognitive bias correction training, and Socratic questioning guidance; Behavior activation modules: hierarchical social tasks, enjoyable activity planning, and work and rest adjustment plans; Matching algorithm unit module: lighting intervention suggestions, noise control solutions; Execution algorithm unit module: Executes a two-stage matching logic. The first stage selects the intervention technology type based on the user's static characteristics, and the second stage determines the specific intervention plan based on the current dynamic status. Intervention execution unit module: supports multi-channel and multi-format intervention push, the APP side provides graphic, video guidance and interactive tasks, the smart hardware side provides voice guidance and synchronous monitoring of physiological signals, and SMS and WeChat sides send concise reminders; during the intervention process, it supports real-time adjustment of difficulty and records task completion.

6. The mental health digital twin construction and dynamic intervention system according to claim 1 is characterized in that: The feedback optimization module includes: Effect evaluation unit module: Establish a multi-dimensional evaluation system, collect physiological index changes 1 hour after intervention, collect psychological assessment data changes 24 hours after intervention, analyze behavioral pattern changes 72 hours after intervention, and comprehensively calculate the intervention effect score. A score of 70 or more is considered effective, and 85 or more is considered excellent. Strategy Iteration Unit Module: This module uses a reinforcement learning framework and uses the intervention effect score as a reward signal. When an intervention technique's effective score is less than 60 points for three consecutive times, an alternative technique is retrieved from the strategy library, and the reason for the ineffectiveness is recorded. For the effective technique, its weight in the matching algorithm is increased, and the implementation parameters are optimized. System upgrade unit module: Global optimization is performed every month. Based on the anonymized data of all users, the twin modeling algorithm and intervention matching model are updated using federated learning. User personalized parameters are retained during the upgrade process to ensure that system performance iteration is compatible with individual adaptability.

7. The mental health digital twin construction and dynamic intervention system according to claim 2 is characterized in that: The preprocessing process of the physiological data unit module includes: Noise filtering: Wavelet transform is used to remove motion artifacts and electromagnetic interference; Feature extraction: extract time domain features, frequency domain features and time-frequency domain features from the original signal; Anomaly detection: Identify abnormal physiological events using the 3σ criterion, mark the time of occurrence and duration of the event, and correlate contemporaneous environmental and behavioral data.

8. The mental health digital twin construction and dynamic intervention system according to claim 5 is characterized in that: The two-stage matching logic of the matching algorithm unit module is specifically as follows: Phase 1: Input the user's static feature vector and use the random forest classification model to predict the probability of responding to interventions in the emotion regulation module, cognitive restructuring module, and behavior activation module. Select technology types with a response probability ≥ 60%; Phase 2: Optimize intervention parameters based on the current dynamic state parameters for the selected technology types. For example, for users with high arousal anxiety, adjust the breathing exercise rhythm to "inhale for 4 seconds - hold for 7 seconds - exhale for 8 seconds" and set a base duration of 5 minutes. For users with high cognitive load, simplify the step description of cognitive restructuring tasks and add examples to guide. Matching constraints: exclude technologies that conflict with user contraindications, and give priority to technology types with a user historical completion rate greater than 70% to ensure the safety and feasibility of the intervention plan.