Psychological crisis multi-stage joint control method and system based on psychological large model

Abstract

The invention provides a psychological crisis multi-stage joint control method and system based on a psychological large model, and aims to realize real-time monitoring, accurate evaluation and intelligent intervention of psychological states through a multi-modal data fusion and deep learning technology. The system collects multi-source information such as texts, voices, videos, physiological signals and behavior data, performs cross-modal analysis by using models such as Transform, LSTM and CNN, constructs personalized psychological portraits, and analyzes and predicts the psychological state change trend in combination with a time sequence. According to the method, a psychological crisis dynamic grading model is adopted, the psychological state of a user is divided into a normal grade, a mild grade, a moderate grade and a severe grade, multi-grade intelligent intervention is provided based on different risk grades, and the multi-grade intelligent intervention comprises AI self-service adjustment, psychological counseling matching, social support enhancement, emergency medical intervention and the like. The psychological intervention strategy is optimized in combination with reinforcement learning, the intervention mode is dynamically adjusted according to user feedback, and individuation and adaptability are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of mental health monitoring, and specifically to a multi-level joint control method and system for mental crisis based on a large mental model. Background Art

[0002] Mental health management is an important issue in the global public health field at present. The continuous increase in the incidence of depression, anxiety, and suicide rate has posed a severe challenge to social stability and individual well-being. Especially in groups such as teenagers, working people, high-pressure occupations (such as medical staff, financial practitioners, military personnel), and the elderly, the incidence of mental problems is relatively high. However, the existing mental health management technologies still have many deficiencies in terms of real-time performance, data accuracy, personalized intervention, multi-level joint control, etc., and it is difficult to effectively respond to sudden mental crises.

[0003] Current mental health management mainly relies on means such as psychological assessment, interviews, psychological counseling, and medical intervention. Although these methods have provided some help for the screening and intervention of mental problems to a certain extent, there are still many limitations. First of all, the traditional mental health assessment methods are single, mainly relying on psychological measurement questionnaires, such as the SAS Self-Rating Anxiety Scale, SDS Self-Rating Depression Scale, PHQ-9 Depression Scale, GAD-7 Anxiety Scale, etc., or relying on the manual assessment of psychologists. However, these methods highly rely on the active participation of users, and the assessment results are greatly affected by the subjective answers of users, making it difficult to accurately reflect the real mental state. In addition, most of these assessments are carried out regularly, with a long assessment cycle, and it is difficult to capture the short-term emotional fluctuations of users, resulting in difficulty in timely warning and intervention when mental crises occur. More importantly, most traditional assessment methods lack the support of physiological data, such as heart rate variability (HRV), skin conductance response (EDA), blood oxygen saturation, etc., resulting in a lack of objective and scientific quantitative criteria for mental health assessment.

[0004] Secondly, the data sources of the existing mental health monitoring systems are relatively single, usually only relying on psychological questionnaires, interview records, medical files, etc., while ignoring multi-dimensional data such as the behavior patterns, voice characteristics, facial expressions, and physiological signals of users. For example, the text expressions on social media can reflect the emotional fluctuations of users, and the tone, speech rate, pitch, and pause patterns of voice can reveal mental problems such as anxiety and depression, while physiological signals (such as HRV, EDA, blood oxygen level) can objectively quantify the mental stress level of users. However, most of the existing mental health assessment systems have not effectively integrated these data, resulting in low accuracy and stability of mental state assessment. In addition, the current mental health management technologies mainly rely on a single data source and lack cross-modal data fusion, making the analysis results of mental health status lack comprehensiveness and prone to misjudgment or missed judgment.

[0005] In addition, existing mental health intervention models have problems of lag and insufficient personalization. Current psychological interventions mainly rely on methods such as psychological counseling, drug treatment, and cognitive behavioral therapy (CBT), and usually require users to actively make appointments for psychological counseling services. However, many users in psychological crises often do not seek help actively before the problems worsen, resulting in a lag in intervention. At the same time, traditional psychological intervention methods lack intelligence and personalization, and cannot dynamically adjust the intervention plan according to the user's real-time mental state. The intervention methods are single and lack adaptability. For example, for patients with anxiety disorder, the triggering factors and coping methods of different users may be different, but the existing intervention measures are often general and difficult to be optimized according to the specific situation of individuals. In addition, there is usually a problem of tight resources in psychological counseling services, making it difficult to cover all users in need. Especially in areas where the resources of psychiatrists are scarce, the accessibility of psychological intervention is relatively low.

[0006] On the other hand, the current mental health management system lacks an effective multi-level joint control mechanism and is difficult to form an all-round mental health support network. Mental health problems usually involve multiple levels, including individuals, social networks, enterprises / schools, medical institutions, etc. However, the existing mental health management system mainly stays at the individual level and lacks a systematic linkage mechanism. For example, individual users often can only rely on self-regulation to relieve psychological pressure, but lack scientific guidance, and the social support system is not effectively utilized. The roles of relatives, friends, and colleagues in the mental health management of users are relatively limited. In addition, the mental health monitoring capabilities of enterprises and schools are weak, making it difficult to detect the mental health problems of employees or students in a timely manner. Psychological intervention measures are often passive rather than proactive prevention and intervention. In terms of medical institutions, due to limited resources of psychiatrists, traditional psychological counseling services are difficult to meet the needs of social mental health management, resulting in uneven distribution of mental health service resources and difficulty in covering all people in need.

[0007] Therefore, we urgently need a multi-level joint control method and system for psychological crises based on a psychological large model to solve the above problems. Summary of the Invention

[0008] The purpose of the present invention is to provide a multi-level joint control method and system for psychological crises based on a psychological large model to solve the problems raised in the above background technology.

[0009] To achieve the above purpose, the present invention provides the following technical solution: A multi-level joint control method for psychological crises based on a psychological large model, the method includes the following steps:

[0010] Multi-source data collection: Obtain information related to mental state through intelligent terminals (such as mobile phones, smart watches), wearable devices (such as heart rate monitors, electroencephalogram analyzers), social media interfaces, user interaction data (such as voice input, text chat, keyboard input behavior), and physiological monitoring devices (such as blood oxygen meters, sleep monitoring devices);

[0011] The collected data includes, but is not limited to: Text data: Analyze social media content, chat records, diaries, emails, etc., and extract emotional keywords, tone features, psychological tendencies, etc.;

[0012] Voice data: Based on Affective Computing, analyze intonation, speech rate, pitch, volume, pause pattern, and identify abnormal emotions, such as anxiety, depression, emotional out-of-control, etc.;

[0013] Video data: Detect facial expressions (such as smiling, frowning, dull eyes), eye movement patterns, body language, etc. through computer vision technology to evaluate the psychological state;

[0014] Physiological data: Collect parameters such as heart rate variability (HRV), electro-dermal activity (EDA), blood oxygen saturation, body temperature, sleep quality, etc., and analyze the physiological stress state;

[0015] Behavior data: Track the user's movement patterns, steps, social interaction frequency, APP usage, and analyze abnormal user behaviors, such as social avoidance, long-term low activity, etc.

[0016] Psychological large model calculation and analysis: Adopt a multi-modal deep learning model, combine technologies such as Transformer, LSTM, CNN, Bayesian network, etc., perform feature extraction, emotion calculation, psychological state prediction on multi-source data, and construct an individual psychological portrait;

[0017] Optimize the psychological state model using self-supervised learning to enhance the ability to identify abnormal behaviors;

[0018] Combine with the personalized emotion baseline, compare the deviation between the user's current state and the normal state, and calculate the psychological stress index.

[0019] Psychological crisis level classification: Adopt a multi-factor weighted evaluation model, and classify the user's psychological state according to parameters such as emotional state, physiological reaction, social pattern, behavior change, etc.:

[0020] Normal (low risk): No obvious emotional fluctuations, and social activities, sleep, and physiological parameters are all within the normal range;

[0021] Mild crisis (low to medium risk): There are mild emotional fluctuations, such as short-term anxiety and increased stress, but it has not affected normal life;

[0022] Moderate crisis (medium risk): Obvious emotional abnormalities occur, such as persistent anxiety, depression, isolation tendency, or reduced social interaction;

[0023] Severe crisis (high risk): Severe psychological crises are detected, such as self-harm, suicidal tendencies, extreme behaviors, or long-term social avoidance.

[0024] Psychological crisis early warning and multi-level intervention: Mild crisis: AI self-regulation, such as meditation training, music therapy, psychological adjustment courses, cognitive behavioral training (CBT);

[0025] Moderate crisis: Intelligent matching of psychological counselors or psychiatrists to provide online or offline psychological counseling and establish a social support system;

[0026] Severe crisis: Activate the emergency intervention mechanism, notify family members, medical institutions or psychological first aid teams, and provide real-time monitoring and remote psychological assistance.

[0027] Dynamic adaptive optimization: Combining reinforcement learning (RL) and federated learning (FL) technologies, continuously optimize the psychological large model according to user feedback to improve the evaluation accuracy;

[0028] Combined with personalized data fine-tuning, optimize the psychological state evaluation and intervention strategies.

[0029] The feature lies in that multi-source data collection adopts feature fusion and denoising processing, including:

[0030] Multi-modal data feature alignment: Normalize the features of information from different data sources, such as uniformly converting text, speech, video, and physiological data into a comparable feature vector space;

[0031] Abnormal data filtering: Based on statistical analysis and machine learning models, remove low-quality or abnormal data, such as sensor misreadings and atypical behaviors;

[0032] Data augmentation: For rare psychological state data, improve the generalization ability of the model through synthetic data and data balancing strategies (such as the SMOTE algorithm).

[0033] As a preferred technical solution of the present invention, multi-source data collection adopts feature fusion and denoising processing, including:

[0034] Multi-modal data feature alignment: Normalize the features of information from different data sources, such as uniformly converting text, speech, video, and physiological data into a comparable feature vector space;

[0035] Abnormal data filtering: Based on statistical analysis and machine learning models, remove low-quality or abnormal data;

[0036] Data augmentation: For rare psychological state data, improve the generalization ability of the model through synthetic data and data balancing strategies.

[0037] As a preferred technical solution of the present invention, the psychological large model adopts cross-modal fusion technology and performs joint modeling through Transformer and deep neural networks, including:

[0038] Text Emotion Analysis: Conduct NLP analysis based on BERT, GPT, or RoBERTa models to extract semantic emotion features;

[0039] Speech Emotion Detection: Analyze audio signals using the CNN-LSTM architecture to identify emotion changes in speech;

[0040] Video Emotion Recognition: Use computer vision tools such as OpenFace and DLib to analyze facial expressions, eye movements, and micro-expression features of the head;

[0041] Physiological Signal Analysis: Combine an LSTM network to process physiological data sequences and predict stress indices and levels of mental fatigue.

[0042] As a preferred technical solution of the present invention, the psychological crisis assessment module adopts a multi-layer risk prediction model, including:

[0043] Short-term Psychological State Prediction: Analyze the emotional fluctuation trend based on data in the past 24 hours;

[0044] Medium-term Psychological State Prediction: Combine emotional, social, and behavioral data from the past week to predict the mental health trend;

[0045] Long-term Psychological State Prediction: Use a time series model to analyze the psychological state in the past few months and identify long-term psychological risks.

[0046] As a preferred technical solution of the present invention, the multi-level joint control mechanism includes a social support system that intelligently analyzes the social network trusted by the user and provides psychological support suggestions with the user's permission.

[0047] As a preferred technical solution of the present invention, the psychological large model has a real-time feedback adjustment function and can automatically adjust psychological intervention strategies according to the user's immediate feedback

[0048] The present invention proposes a multi-level joint control system for psychological crisis based on a psychological large model, including the following modules:

[0049] (1) Multimodal Data Acquisition Module: Obtain the user's text, speech, video, physiological signals, and behavioral data through intelligent terminals, wearable devices, social media interfaces, and psychological assessment tools;

[0050] Adopt data fusion technology to perform feature alignment, denoising processing, and data enhancement on information from different data sources to improve the accuracy of psychological state analysis.

[0051] (2) Psychological Large Model Calculation Module: Adopt a multi-modal deep learning model based on technologies such as Transformer, LSTM, CNN, and Bayesian networks to perform feature extraction, emotion calculation, and psychological state modeling on the data;

[0052] Calculate the psychological stress index in combination with the user's personalized emotional baseline, and predict the trend of the psychological state based on real-time data.

[0053] (3) Psychological crisis grading and assessment module: Adopt a multi-factor weighted assessment model to calculate the psychological crisis level according to parameters such as text emotion, physiological state, social pattern, and behavior change;

[0054] Combine time series prediction to conduct short-term (24 hours), medium-term (one week), and long-term (several months) trend analysis of the user's psychological state;

[0055] Adopt Bayesian network and logistic regression models to conduct dynamic grading and anomaly detection of the psychological state.

[0056] (4) Intelligent intervention module: Self-help intervention system: Generate personalized psychological adjustment suggestions based on a large model, such as mindfulness training, cognitive behavioral therapy (CBT), social activity recommendations, etc.;

[0057] Mental health support system: Intelligent match psychological counselors and provide online or offline consultation appointment services;

[0058] Social support recommendation: Analyze the user's social network and recommend establishing connections with close contacts or support groups;

[0059] Emergency intervention mechanism: When detecting a high-risk psychological crisis, send a warning to the user's designated contacts, medical institutions, or emergency teams, and initiate remote psychological intervention.

[0060] (5) Adaptive learning module: Combine reinforcement learning (RL) to continuously optimize the psychological state assessment model according to user feedback and improve the accuracy of personalized intervention;

[0061] Adopt federated learning (FL) technology to optimize the psychological large model using distributed data while protecting user privacy and improving its generalization ability.

[0062] (6) Data security and privacy protection module: Use blockchain encryption technology to store the user's mental health data to ensure data integrity and privacy;

[0063] Combine differential privacy (DP) technology to prevent the leakage of user data during transmission or storage;

[0064] Adopt a distributed computing architecture so that the processing of mental health data can be completed on local or edge computing devices to improve data security.

[0065] (7) Platform Support and Scalability: The system supports the collaboration between cloud and edge computing and can run on smartphones, smartwatches, PCs, and IoT devices;

[0066] It has cross-platform data synchronization function, supports users to seamlessly switch between multiple devices, and ensures the continuity of mental health management.

[0067] As a preferred technical solution of the present invention, the mental crisis grading and evaluation module combines the mental health database and individual historical data to dynamically adjust the risk level of the user's mental state and improve the evaluation accuracy.

[0068] As a preferred technical solution of the present invention, the intelligent intervention module optimizes psychological adjustment suggestions based on reinforcement learning and can personalized adjust the intervention plan according to user feedback to improve the intervention effect.

[0069] As a preferred technical solution of the present invention, the data security and privacy protection module adopts blockchain and differential privacy technologies to improve the storage security and access control of user mental health data.

[0070] Compared with the prior art, the beneficial effects of the present invention are:

[0071] Through the multi-modal data fusion technology, the present invention combines data such as text, voice, video, and physiological signals to improve the accuracy of mental state evaluation, breaking through the limitations of traditional mental health management that relies on a single data source, has a single evaluation method, and lacks real-time performance. By using a deep learning model to comprehensively analyze the user's emotions, behavior patterns, social interactions, and physiological signals, a personalized mental portrait is established, and the change trend of the mental state is predicted through time series analysis, making the identification of mental crises more scientific, objective, and real-time, so that mental health risks can be detected earlier.

[0072] The present invention constructs an intelligent mental health intervention system. Based on the mental crisis grading model, it adopts multi-level intelligent intervention means such as AI self-regulation, psychological counseling matching, social support guidance, and emergency medical intervention to provide personalized and precise mental intervention plans. The system combines reinforcement learning (RL) to optimize the intervention strategy and dynamically adjusts the mental intervention method according to user feedback to improve the adaptability and effectiveness of the intervention. At the same time, the system has an active monitoring and reminder function, and users can obtain mental health services without actively seeking help, thus effectively reducing the lag of mental intervention and improving the timeliness and effectiveness of coping with mental crises.

[0073] The present invention constructs a multi-level joint control mechanism to achieve collaborative linkage between individuals, social networks, enterprises / schools, and medical institutions. The system can not only help users to adjust their own psychology, but also guide users to seek support from relatives, friends, and colleagues when necessary, or automatically match professional psychological counseling resources to ensure that mental health issues receive timely attention. When the psychological crisis worsens, the system can also automatically notify emergency contacts or medical institutions to improve the global responsiveness of mental health management. Through this multi-level intelligent mental health management system, the present invention effectively improves the efficiency of psychological crisis identification, early warning, and intervention, and can be widely used in personal health management, corporate employee care, school mental health monitoring, and social psychological service systems, providing a more intelligent and efficient solution for social psychological health management. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] Figure 1 This is a flow chart of the multi-level cascade control method for psychological crisis based on the psychological big model proposed by the present invention. DETAILED DESCRIPTION

[0075] The following is a combination of the embodiments of the present invention Figure 1 The technical solutions in the embodiments of the present invention are clearly and completely described. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. All other embodiments derived by persons of ordinary skill in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0076] See Figure 1 This invention provides a multi-level cascaded psychological crisis control method and system based on a large psychological model. Through multimodal data acquisition, deep learning modeling, psychological crisis assessment, multi-level intelligent intervention, and adaptive learning optimization, it aims to achieve real-time monitoring, accurate assessment, and effective intervention of users' mental health. This system is suitable for personal mental health management, corporate employee care, psychological diagnosis and treatment in medical institutions, school mental health monitoring, and social psychological service systems.

[0077] The core functional modules of the system include: Multimodal data acquisition module: collects text, voice, video, physiological signals and behavioral data to build a user psychological state data set;

[0078] Psychological model calculation module: Analyzes the user's psychological state based on cross-modal fusion technology based on deep learning and establishes individual psychological portraits;

[0079] Psychological crisis grading assessment module: uses a multi-factor weighted assessment method to dynamically grade psychological states and combines time series to predict mental health trends;

[0080] Intelligent Intervention Module: Provide multi-level intelligent intervention solutions based on the psychological crisis level, including AI self-regulation, psychological counseling, social support, and emergency medical intervention;

[0081] Adaptive Learning Module: Optimize the psychological large model by combining reinforcement learning and federated learning to improve the accuracy of personalized assessment;

[0082] Data Security and Privacy Protection Module: Use blockchain and differential privacy technologies to ensure the security of user data.

[0083] Multi-modal Data Acquisition Module: (1) Data sources: Smart terminals: mobile phones, smart watches, tablets, PCs;

[0084] Wearable devices: heart rate monitors, blood oxygen meters, sleep trackers, electroencephalogram detectors;

[0085] Social media interfaces: Weibo, Twitter, WeChat, Facebook, etc.;

[0086] Physiological monitoring devices: blood pressure monitors, pulse sensors, sweat sensing devices;

[0087] Office / learning environment monitoring: smart desks, mouse and keyboard interaction records, remote conferencing systems.

[0088] Collected data types: Text data: social media posts, message chat records, emails, diaries;

[0089] Natural Language Processing (NLP) extracts users' emotional expressions, tone changes, and negative emotion word frequencies;

[0090] Monitor statements where users express emotions such as anxiety, depression, and helplessness.

[0091] Voice data: Analyze calls, voice assistant interactions, remote conferences, etc.; Use CNN-LSTM to analyze voice features such as pitch, speech rate, pitch, and pauses; Identify emotions such as anxiety (rapid), depression (low), and anger (increased volume).

[0092] Video data: Capture facial expressions, eye movement trajectories, and body movements through cameras; Use tools such as OpenFace and DLib to detect micro-expressions, such as frowning, biting lips, and dull eyes; Calculate the frequency of eye drift, blink rate, and facial muscle tension.

[0093] Physiological data: Collect heart rate variability (HRV), skin conductance response (EDA), sleep quality, and blood oxygen saturation; Combine machine learning to predict users' stress levels and psychological fatigue degrees.

[0094] Behavioral data: Monitor APP usage frequency, social interaction times, keyboard typing rhythm, and mouse click speed;

[0095] Detect motion patterns, such as prolonged sitting and lack of exercise, and prompt the risk of mental burnout or depression.

[0096] Psychological large model computing module: Text analysis: Use pre-trained models such as BERT and GPT to analyze the emotional trend of user texts; Speech analysis: Use the CNN-LSTM fusion model to detect user emotional fluctuations; Video analysis: Perform facial expression recognition based on computer vision technology, and combine frameworks such as OpenFace to improve accuracy; Physiological signal analysis: Process physiological data based on LSTM and calculate the mental stress index.

[0097] Psychological crisis grading and assessment module: Use the Bayesian network + time series analysis model to calculate the psychological crisis level:

[0098] Normal (low risk):

[0099] No obvious emotional fluctuations, active in social activities, and normal physiological parameters;

[0100] The behavior pattern is stable, and no excessive stress, anxiety or depression tendency is detected.

[0101] Mild crisis (low to medium risk): Short-term emotional fluctuations intensify, such as reduced social interaction and negative voice emotions;

[0102] Slightly reduced HRV and slightly increased EDA, indicating mild stress or anxiety.

[0103] Moderate crisis (medium risk): Voice, text and video data show signs of anxiety and depression;

[0104] Physiological data deviate from the normal range for a long time, such as continuously high EDA and low HRV.

[0105] Severe crisis (high risk): Detect text content with self-harm or suicide tendencies;

[0106] Extreme emotions appear in voice data, such as crying, anger, and low intonation;

[0107] Facial expression analysis indicates a long-term lack of emotional changes (numb state).

[0108] Intelligent intervention module:

[0109] (1) Mild crisis intervention: AI pushes psychological courses, music therapy, and meditation training; It is recommended to adjust the work and rest schedule and improve living habits.

[0110] (2) Moderate crisis intervention: Recommend a psychologist and arrange remote psychological counseling; Establish a support system through social networks and encourage users to seek help actively.

[0111] (3) Severe crisis intervention: Urgently notify the designated contacts (family members, friends) of the user;

[0112] Initiate remote medical support and recommend psychiatrists or hospital resources.

[0113] Adaptive learning module: Combine reinforcement learning (RL) and adjust the intervention strategy based on user feedback;

[0114] Adopt federated learning (FL) to optimize the psychological large model and protect user privacy.

[0115] Data security and privacy protection module: Use blockchain technology to encrypt and store user data to ensure immutability; Use differential privacy (DP) technology to prevent data leakage.

[0116] Example 1: Personal mental health management

[0117] Data collection: The smartwatch records the user's heart rate data, and the mobile phone collects social media interaction situations;

[0118] Psychological analysis: Calculate HRV based on LSTM and predict the trend of psychological stress;

[0119] Crisis grading: If the user's social activity decreases by 50% and the HRV drops by 20%, it is evaluated as a moderate crisis;

[0120] Intervention measures: Recommend that the user participate in relaxation training; Provide psychological counseling appointments.

[0121] Example 2: Mental health monitoring of enterprise employees

[0122] Data collection: The smart workstation detects the employee's working hours and mouse and keyboard interaction data;

[0123] Psychological analysis: Calculate the results of email sentiment analysis and detect the stress trend;

[0124] Crisis grading:

[0125] If the overtime hours exceed 2 hours for three consecutive days and the voice sentiment is negative, it is evaluated as a mild crisis;

[0126] If the employee shows a low mood for a consecutive week, it is upgraded to a moderate crisis;

[0127] Intervention measures: Mild crisis: Prompt to rest; Moderate crisis: Enterprise psychological counseling appointment; Severe crisis: HR intervention and provide psychological assistance resources.

[0128] Based on the psychological large model, the present invention constructs a complete psychological crisis joint control system, covering data collection, computational analysis, psychological crisis assessment, intelligent intervention and adaptive optimization, providing an efficient and personalized mental health management solution for individuals, enterprises and social institutions, improving the ability to identify psychological crises and enhancing the effect of psychological intervention.

[0129] The content not described in detail in this specification belongs to the prior art well-known to those skilled in the art. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent replacements for some of the technical features. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A multi-level joint control method for psychological crises based on a large psychological model, characterized in that, The method includes the following steps: Multi-source data collection: Obtain information related to mental state through intelligent terminals, wearable devices, social media interfaces, user interaction data, and physiological monitoring devices; The collected data includes but is not limited to: Text data: Analyze social media content, chat records, diaries, and emails to extract emotion keywords, tone features, and psychological tendencies; Voice data: Based on affective computing, analyze intonation, speech rate, pitch, volume, and pause patterns to identify abnormal emotions; Video data: Detect facial expressions, eye movement patterns, and body language through computer vision technology to evaluate the mental state; Physiological data: Collect heart rate variability, skin conductance response, blood oxygen saturation, body temperature, and sleep quality parameters to analyze the physiological stress state; Behavioral data: Track the user's movement patterns, steps, social interaction frequency, and APP usage to analyze abnormal user behavior; Mental large model calculation and analysis: Adopt a multi-modal deep learning model, combine Transformer, LSTM, CNN, and Bayesian network technologies to extract features, calculate emotions, and predict the mental state of multi-source data, and construct an individual mental profile; Optimize the mental state model using self-supervised learning to enhance the ability to identify abnormal behaviors; Combine the personalized emotion baseline, compare the deviation between the user's current state and the normal state, and calculate the mental stress index; Mental crisis level classification: Adopt a multi-factor weighted evaluation model to classify the user's mental state according to emotion state, physiological reaction, social pattern, and behavior change parameters: Normal: No obvious emotional fluctuations, and social activities, sleep, and physiological parameters are all within the normal range; Mild crisis: There are mild emotional fluctuations, such as short-term anxiety and increased stress, but it has not affected normal life; Moderate crisis: Obvious emotional abnormalities occur, such as persistent anxiety, depression, isolation tendency, or reduced social interaction; Severe crisis: Severe mental crises are detected, such as self-harm, suicide tendency, extreme behaviors, or long-term social avoidance; Mental crisis early warning and multi-level intervention: Mild crisis: AI self-help adjustment, such as meditation training, music therapy, psychological adjustment courses, and cognitive behavior training; Moderate crisis: Intelligent matching of psychological counselors or psychiatrists to provide online or offline psychological counseling and establish a social support system; Severe crisis: Activate the emergency intervention mechanism, notify family members, medical institutions, or psychological first aid teams, and provide real-time monitoring and remote psychological assistance; Dynamic adaptive optimization: Combine reinforcement learning and federated learning technologies to continuously optimize the mental large model according to user feedback, and fine-tune it with personalized data to optimize the mental state assessment and intervention strategies.

2. The multi-level joint control method for psychological crisis based on the psychological large model according to claim 1, characterized in that Feature fusion and denoising processing are adopted for multi-source data collection, including: Multi-modal data feature alignment: Standardize the features of information from different data sources, such as uniformly converting text, voice, video, and physiological data into a comparable feature vector space; Abnormal data filtering: Based on statistical analysis and machine learning models, remove low-quality or abnormal data; Data augmentation: For rare mental state data, improve the generalization ability of the model through synthetic data and data balancing strategies.

3. The multi-level joint control method for psychological crisis based on the psychological large model according to claim 1, characterized in that, The psychological large model adopts cross-modal fusion technology and conducts joint modeling through Transformer and deep neural networks, including: Text emotion analysis: Conduct NLP analysis based on models such as BERT, GPT, or RoBERTa to extract semantic emotion features; Speech emotion detection: Analyze audio signals using the CNN-LSTM architecture to identify emotion changes in speech; Video emotion recognition: Use computer vision tools such as OpenFace and DLib to analyze facial expressions, eye movements, and micro-expression features of the head; Physiological signal analysis: Combine the LSTM network to process physiological data sequences and predict stress index and mental fatigue level.

4. The multi-level joint control method for psychological crisis based on the psychological large model according to claim 1, characterized in that The psychological crisis assessment module adopts a multi-layer risk prediction model, including: Short-term psychological state prediction: Analyze the emotion fluctuation trend based on data in the past 24 hours; Medium-term psychological state prediction: Combine emotion, social, and behavioral data in the past week to predict the mental health trend; Long-term psychological state prediction: Use time series models to analyze the psychological state in the past few months and identify long-term psychological risks.

5. The method for multi-level joint control of psychological crisis based on the psychological large model according to claim 1, wherein The multi-level joint control mechanism includes a social support system, which intelligently analyzes the social network trusted by the user and provides psychological support suggestions with the user's permission.

6. The multi-level joint control method for psychological crisis based on the psychological large model according to claim 1, wherein The psychological large model has a real-time feedback adjustment function and automatically adjusts psychological intervention strategies according to the user's immediate feedback.

7. A multi-level joint control system for psychological crisis based on a large psychological model, characterized in that, It includes the following modules: (1) Multi-modal data collection module: Obtain the user's text, speech, video, physiological signals, and behavioral data through intelligent terminals, wearable devices, social media interfaces, and psychological assessment tools; Adopt data fusion technology to perform feature alignment, denoising processing, and data enhancement on information from different data sources to improve the accuracy of psychological state analysis; (2) Psychological large model calculation module: Adopt a multi-modal deep learning model based on technologies such as Transformer, LSTM, CNN, and Bayesian network to extract features, calculate emotions, and model psychological states from the data; Combine the user's personalized emotion baseline, calculate the psychological stress index, and predict the trend of psychological state based on real-time data; (3) Psychological crisis grading assessment module: Adopt a multi-factor weighted assessment model to calculate the psychological crisis level according to text emotion, physiological state, social pattern, and behavior change parameters; Combine time series prediction to conduct short-term, medium-term, and long-term trend analysis of the user's psychological state; Adopt Bayesian network and logistic regression models to dynamically grade and detect abnormalities in the psychological state; (4) Intelligent intervention module: Self-help intervention system: Generate personalized psychological adjustment suggestions based on the large model; Mental health support system: Intelligently match psychological counselors and provide online or offline consultation appointment services; Social support Recommendation: Analyze the user's social network and recommend establishing connections with close contacts or support groups; Emergency intervention mechanism: When detecting a high-risk psychological crisis, send warnings to the user's designated contacts, medical institutions, or emergency teams, and initiate remote psychological intervention; (5) Adaptive learning module: Combine reinforcement learning to continuously optimize the psychological state assessment model according to user feedback and improve the accuracy of personalized intervention; Adopt federated learning technology to optimize the psychological large model using distributed data while protecting user privacy, and improve its generalization ability; (6) Data security and privacy protection module: Use blockchain encryption technology to store users' mental health data to ensure data integrity and privacy; Combine differential privacy technology to prevent the leakage of user data during transmission or storage; Adopt a distributed computing architecture to enable the processing of mental health data to be completed on local or edge computing devices, improving data security. (7) Platform support and scalability: The system supports the collaboration between the cloud and edge computing and runs on smartphones, smartwatches, PCs, and IoT devices; It has cross-platform data synchronization function and supports seamless switching of users between multiple devices.

8. The system according to claim 7, wherein The psychological crisis grading and evaluation module combines the mental health database and individual historical data to dynamically adjust the risk level of the user's mental state and improve the evaluation accuracy.

9. The system according to claim 7, wherein The intelligent intervention module optimizes psychological adjustment suggestions based on reinforcement learning and personalized adjusts the intervention plan according to user feedback to improve the intervention effect.

10. The system according to claim 7, wherein The data security and privacy protection module uses blockchain and differential privacy technologies to improve the storage security and access control of users' mental health data.

Images