A health management platform for children and adolescents in five health conditions

Abstract

This invention relates to the field of children and adolescents' mental health management technology, and more particularly to a five-cycle health management platform for children and adolescents, comprising an original segment module, a psychological index module, a resilience coefficient module, a crisis scoring module, and a graded early warning module. This invention manages students in special groups through psychological estimations in a first and second time window. By constructing specialized behavioral description systems for different disorder types, this invention can effectively identify and quantify clinically significant special behavioral patterns, reduce semantic loss caused by generic labels, and significantly improve assessment accuracy compared to existing technologies.

Description

Technical Field

[0001] This invention relates to the field of children and adolescents' mental health management technology, and in particular to a five-cycle health management platform for children and adolescents. Background Technology

[0002] The Five Healths encompass five dimensions of children and adolescents' health: weight, vision, mental health, skeletal health, and oral health. In the field of children and adolescents' mental health management, especially for special groups such as children with intellectual disabilities, autism spectrum disorders, and those in inclusive education programs, existing mental health assessment and intervention techniques have the following shortcomings: Traditional psychological assessment scales and behavioral observation tools are primarily designed for children with typical developmental characteristics. Their standardized items and scoring rules are difficult to adapt to the significant individual differences among children with special needs in language expression, social interaction, and emotion regulation, leading to low accuracy and high false positive rates in assessment results; existing behavioral analysis systems use generic action classification labels, which are difficult to effectively... Identifying and quantifying clinically significant abnormal behavioral patterns, such as stereotyped behaviors, self-stimulatory behaviors, and sensory-seeking behaviors unique to children with special needs, leads to a lack of information for inferring their mental state. Most existing technologies only focus on the students' own physiological and behavioral signals, failing to systematically incorporate the impact of environmental variables such as light, noise, and social density on the mental state of children with special needs. Because children with special needs are highly sensitive to environmental changes, ignoring environmental factors will seriously reduce the reliability of assessment and early warning. Existing early warning mechanisms are mostly based on the instantaneous state at the current moment, failing to utilize historical trends and future evolution predictions, thus failing to achieve proactive intervention and often missing the best opportunity for early intervention. Summary of the Invention

[0003] The purpose of this invention is to provide a five-cycle health management platform for children and adolescents to solve the problems mentioned in the background art.

[0004] The technical solution of this invention is as follows: A health management platform for children and adolescents' five-stage development cycle includes: The original segment module is used to preset the first time window, acquire the student's physiological data, environmental data and original behavior flow within the first time window, and construct the original behavior segment based on the first time window and the original behavior flow. The psychological index module is used to map the original behavioral fragments to the adaptive behavior code of the current student, calculate the four-dimensional psychological state vector of the student at the current time based on the adaptive behavior code, physiological data and environmental data, and output the comprehensive psychological index of the current student. The elasticity coefficient module is used to obtain the estimated value of the four-dimensional psychological state vector and the estimated psychological comprehensive index of the current student under the second time window based on the four-dimensional psychological state vector and the psychological comprehensive index; and to construct the dual-time psychological elasticity coefficient based on the actual value and the estimated value of the four-dimensional psychological state vector and the psychological comprehensive index. The crisis scoring module is used to calculate six membership functions of the current student based on the estimated value of the four-dimensional psychological state vector, the estimated comprehensive psychological index, and the estimated rate of change of the comprehensive psychological index; to obtain a fuzzy rule base based on the estimated value of the four-dimensional psychological state vector, the estimated comprehensive psychological index, and the estimated rate of change of the comprehensive psychological index; and to obtain the current student's psychological crisis score based on the fuzzy rule base and the six membership functions. The graded early warning module is used to obtain personalized psychological early warning thresholds for different students based on the dual-time psychological elasticity coefficients, and to classify and manage students' psychological crises by graded approach based on personalized psychological early warning thresholds and psychological crisis scores.

[0005] Preferably, the preset first time window, acquiring the student's physiological data, environmental data, and original behavior flow within the first time window, and constructing an original behavior segment based on the first time window and the original behavior flow, includes: setting the first time window according to the current student's five-key data collection frequency; acquiring the current student's physiological data, environmental data, and original behavior flow within the first time window; the physiological data including the student's heart rate cycle changes, skin conductance response, and respiratory rate; the environmental data including light intensity, noise intensity, and social density-related quantities; the original behavior flow including the student's hand skeletal joint sequence and hand trajectory in a time-continuous state; and constructing the current student's original behavior segment based on the first time window and the original behavior flow within the first time window using a boundary detection function; the original behavior segment includes a complete behavior unit of the current student.

[0006] Preferably, the step of mapping the original behavioral fragments to the adaptive behavior code of the current student, calculating the student's four-dimensional psychological state vector at the current time based on the adaptive behavior code, physiological data, and environmental data, and outputting the current student's comprehensive psychological index includes: mapping the original behavioral fragments of each student in the special student group to the current student's adaptive behavior code through an annotator; the adaptive behavior code includes classifying the original behavioral fragments through annotation to obtain four coding components of the current student; the four coding components include a motor component, a social component, a sensory component, and an emotional component; constructing a four-dimensional coupled model of the current student's psychological state based on the adaptive behavior code, physiological data, and environmental data; the four-dimensional coupled model of the psychological state includes calculating the current student's behavioral activation, physiological arousal, environmental support, and social willingness based on the four coding components in the adaptive behavior code; constructing the current student's behavioral activation, physiological arousal, environmental support, and social willingness into a four-dimensional psychological state vector of the current student at the current moment; and calculating the current student's comprehensive psychological index by weighted summation based on the four-dimensional psychological state vector, wherein the calculation formulas for behavioral activation, physiological arousal, environmental support, and social willingness are: ; ; ; ; in, For behavioral activation, Physiological arousal level For environmental support, For social willingness, The hyperbolic tangent function is used to map the integral result to the interval [0,1). For the current moment, The length of the integration time window. For integration variables, For at any time The student's instantaneous average speed of motion The sigmoid function is used to map linear combinations to the interval (0,1). The weighting coefficients for heart rate cycle variations. This is the weighting coefficient for the skin conductance response. For students exist Heart rate cycle changes over time. For students exist skin conductance levels at any given time For reference speed, For all students, the minimum and maximum heart rate cycle variations in historical data are given. For all students, the minimum and maximum skin conductance response levels in the historical data are given. for The ambient light intensity at any given time The preset upper limit of ambient light intensity, for The linear sound pressure level at time t. The preset maximum permissible sound pressure level, for The minimum Euclidean distance between students at any given time and other students. The preset maximum social distancing threshold, It is a multilayer perceptron. Original behavioral fragments social weight For the first A fragment of original behavior.

[0007] Preferably, the step of obtaining the current student's comprehensive psychological index by weighted summation based on the four-dimensional psychological state vector includes: performing a unidirectional analysis based on the four-dimensional psychological state vector; calculating the current student's comprehensive psychological index by weighted summation based on preset behavioral weights, physiological weights, environmental weights, and social weights of the unidirectional analysis; the comprehensive psychological index represents the current student's mental health level; and calculating the rate of change of the psychological index using a differential method based on the comprehensive psychological index, wherein the formula for calculating the comprehensive psychological index is: ; in, For behavioral activation, Physiological arousal level For environmental support, For social willingness, These are the weighting coefficients for behavioral activation, physiological arousal, environmental support, and social willingness. This is the Sigmoid function.

[0008] Preferably, the step of obtaining the estimated value of the four-dimensional psychological state vector and the estimated psychological comprehensive index of the current student in the second time window based on the four-dimensional psychological state vector and the psychological comprehensive index includes: constructing a psychological state attention network for a special student group on the five-cycle health management platform based on the student's four-dimensional psychological state vector at the current moment, wherein the psychological state attention network includes a teacher attention network and a student attention network; calculating the estimated residuals of the four encoded components in the four-dimensional psychological state vector through the psychological state attention network and the original behavioral flow, physiological data and environmental data; calculating the actual value of the four-dimensional psychological state vector at the current moment by calculating the four-dimensional psychological state vector and the estimated residuals of the four encoded components in the four-dimensional psychological state vector; obtaining the actual psychological comprehensive index and the actual rate of change of the psychological index at the current moment by using the actual value of the four-dimensional psychological state vector at the current moment; presetting a second time window based on the first time window; obtaining the estimated value of the four-dimensional psychological state vector in the second time window based on the actual value of the four-dimensional psychological state vector at the current moment and the psychological state transfer model; and calculating the estimated psychological comprehensive index and the estimated rate of change of the psychological comprehensive index in the second time window based on the estimated value of the four-dimensional psychological state vector in the second time window.

[0009] Preferably, the mental state attention network includes a teacher attention network and a student attention network, comprising: using the teacher residuals of the four encoded components in the four-dimensional mental state vector as soft labels for the teacher attention network, and obtaining the estimated residuals of the four encoded components in the four-dimensional mental state vector through the student attention network, wherein the fusion calculation formula for the teacher attention network and the loss function formula for the student attention network are: ; ; ; ; ; ; in, The fusion feature matrix of the teacher attention network. For teacher residuals in teacher attention networks, Let be the loss function of the student attention network. Let KL divergence loss function be used. Let the mean squared error loss function be . For boundary regularization loss function, and To balance hyperparameters, , and The projection matrix of the cross-modal attention layer. , and The modal characteristic matrix, This represents the dimension of the attention layer, with a value of 64. For multilayer perceptron functions, This represents the total number of moments within the first time window. This represents the number of moments within the first time window; For the softmax function, This is the index of the four-dimensional mental state components in the residual vector. For distillation temperature parameters, The first in the student attention network output The estimated residuals of the component index, For the output of the teacher's attention network, the first Teacher residuals of component indexes, For students exist In the moment A four-dimensional mental state vector with component indexes. It is an absolute value.

[0010] Preferably, the step of constructing a dual-time psychological elasticity coefficient based on the actual and estimated values ​​of the four-dimensional psychological state vector and the psychological comprehensive index includes: calculating a first-time psychological elasticity coefficient under a first time window based on the actual value of the four-dimensional psychological state vector and the actual psychological comprehensive index; calculating a second-time psychological elasticity coefficient under a second time window based on the first-time psychological elasticity coefficient, the estimated value of the four-dimensional psychological state vector, and the estimated psychological comprehensive index; and calculating the dual-time psychological elasticity coefficient of the current student based on the first-time psychological elasticity coefficient, the second-time psychological elasticity coefficient, the first-time psychological weight, and the second-time psychological weight. The dual-time psychological elasticity coefficient is the psychological elasticity coefficient of the current student under the conditions of the first and second time windows. The calculation formulas for the first-time psychological elasticity coefficient, the second-time psychological elasticity coefficient, and the dual-time psychological elasticity coefficient are as follows: ; ; ; in, The first-time psychological resilience coefficient, This is the second time-based psychological elasticity coefficient. This is a two-time psychological elasticity coefficient. This refers to the number of moments within the first time window. For the Euclidean norm, This represents the total number of moments within the first time window. In order to be in A four-dimensional mental state vector at any given moment. This is the baseline four-dimensional mental state vector. For the second time window The estimated value of the four-dimensional mental state vector at time t. For error terms, Here is the uncertainty coefficient. This is a sensitivity hyperparameter. For recovery time, For the maximum deviation, For reference only.

[0011] Preferably, the step of calculating six membership functions for the current student based on the estimated value of the four-dimensional psychological state vector, the estimated psychological comprehensive index, and the estimated rate of change of the psychological comprehensive index, and obtaining a fuzzy rule base based on the estimated value of the four-dimensional psychological state vector, the estimated psychological comprehensive index, and the estimated rate of change of the psychological comprehensive index, includes: calculating the first membership degree, the second membership degree, the third membership degree, and the fourth membership degree based on the estimated value of the four-dimensional psychological state vector; calculating the fifth membership degree based on the estimated psychological comprehensive index; and calculating the sixth membership degree based on the estimated rate of change of the psychological comprehensive index; the first membership degree is the corresponding... The membership degree is calculated as follows: the first membership degree is the degree of membership of the behavioral activation estimate to the high-activation state; the second membership degree is the degree of membership of the corresponding physiological arousal estimate to the high-arousal state; the third membership degree is the degree of membership of the corresponding environmental support estimate to the low-support state; and the fourth membership degree is the degree of membership of the corresponding social willingness estimate to the low-willing state. All four membership degrees are calculated using a Gaussian membership function. Fuzzy rules for the current student are constructed using the four-dimensional psychological state vector estimate, the estimated psychological comprehensive index, and the estimated rate of change of the psychological comprehensive index. The fuzzy rule base determines the student's status as follows: if behavioral activation and physiological arousal are greater than the first threshold and environmental support and social willingness are less than the second threshold, the student is in a first rule crisis; if behavioral activation and physiological arousal are less than the second threshold, the student is in a second rule crisis; if the estimated psychological comprehensive index is less than the second threshold and the duration exceeds the psychological recovery time threshold, the student is in a third rule crisis; if the estimated psychological comprehensive index is less than the third threshold, the student is in a fourth rule crisis; if the estimated rate of change of the psychological comprehensive index is less than zero, the student is in a fifth rule crisis. The first, second, and third thresholds are the boundary conditions for estimating the four-dimensional psychological state vector and the estimated psychological comprehensive index; and the first threshold is greater than the second threshold, which is greater than the third threshold. Based on the distribution of corresponding indicators in the historical data of the student group, the upper quartile, median, and lower quartile are taken as the first, second, and third thresholds, respectively. The calculation formulas for the first, second, third, fourth, and fifth membership degrees, as well as the membership degree calculation formula for the estimated rate of change of the psychological comprehensive index, are as follows: ; ; in, For membership degree, The independent variables of the membership function include the four component estimates of the four-dimensional mental state vector estimate and the estimated psychological composite index. The central value of the membership function. The standard deviation of the membership function. To estimate the rate of change of the comprehensive psychological index, To estimate the sixth membership degree of the rate of change of the psychological composite index, The coefficient of variation, For the change offset, The rate of change of the comprehensive psychological index is used as a reference.

[0012] Preferably, obtaining the current student's psychological crisis score based on the fuzzy rule base and six membership functions includes: obtaining the current student's crisis confidence level based on the different rule-based crises the current student is in; calculating the crisis score when each rule-based crisis is activated using the crisis confidence level and the six membership functions; and taking the maximum crisis score when each rule is activated as the current student's psychological crisis score. The formulas for calculating the crisis score and the psychological crisis score are as follows: ; ; in, Assign crisis scores to each rule-related crisis. Assess the current psychological crisis among students. For the current rule crisis The set of membership degrees involved varies depending on the rule or crisis. The current student's crisis level weight is determined based on the current rule-based crisis level. For the rule crisis sequence number, These are the first rule crisis, the second rule crisis, the third rule crisis, the fourth rule crisis, and the fifth rule crisis.

[0013] Preferably, the step of obtaining personalized psychological early warning thresholds for different students based on the dual-time psychological elasticity coefficient, and classifying and managing students' psychological crises by using personalized psychological early warning thresholds and psychological crisis scores, further includes: calculating personalized psychological early warning thresholds based on the dual-time psychological elasticity coefficients, and determining the student's warning level based on the personalized psychological early warning thresholds and psychological crisis scores. The determination criteria are as follows: if the psychological crisis score is in the first range, the current student is at the green level, and the current student is considered psychologically normal; if the psychological crisis score is in the second range, the current student is at the blue level; if the psychological crisis score is in the third range, the current student is at the yellow level; if the psychological crisis score is in the fourth range... If the psychological crisis score is in the fifth range, the student is at the orange level; if the psychological crisis score is in the fifth range, the student is at the red level; if the psychological crisis score is greater than the personalized psychological warning threshold, the student's warning level is set to red. The first range corresponds to the green level, indicating that the student's psychological state is normal; the second range corresponds to the blue level, indicating that there is a slight psychological risk; the third range corresponds to the yellow level, indicating that there is a moderate psychological risk; the fourth range corresponds to the orange level, indicating that there is a high psychological risk; the fifth range corresponds to the red level, indicating that there is a serious psychological crisis. The first, second, third, and fourth ranges are obtained according to the equal-frequency binning method and the equal-proportional division method.

[0014] The beneficial effects of this invention are as follows: 1. This invention constructs a specialized behavioral description system for different types of disorders, which can effectively identify and quantify special behavioral patterns with clinical significance, reduce semantic loss caused by general labels, and significantly improve the accuracy of assessment compared with existing technologies; 2. This invention collects environmental parameters such as light intensity, noise, and social density through a system and incorporates them into an analysis model. This allows it to distinguish between state fluctuations caused by environmental changes and genuine mental health risks, significantly reducing the false alarm rate in actual classroom environments. 3. This invention constructs an individualized state evolution prediction model, which not only assesses the current psychological state, but also predicts the trend of change in the next few minutes, thus upgrading the early warning system from post-event response to pre-event prevention. 4. The intervention prescriptions automatically generated by reinforcement learning in this invention can be dynamically optimized based on students' historical response records. The recommended intervention actions are more personalized and can be continuously optimized and upgraded over time, effectively improving long-term adherence to the intervention. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of the structure of the five-cycle health management platform for children and adolescents of the present invention; Figure 2 This is a schematic diagram of the multi-attention network of the five-cycle health management platform for children and adolescents of the present invention. Detailed Implementation

[0016] The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments, but this is not intended to limit the present invention.

[0017] A health management platform for children and adolescents' five-stage development cycle, such as Figure 1 and 2 As shown, it includes: The original segment module is used to preset the first time window, acquire the student's physiological data, environmental data and original behavior flow within the first time window, and construct the original behavior segment based on the first time window and the original behavior flow. The psychological index module is used to map the original behavioral fragments to the adaptive behavior code of the current student, calculate the four-dimensional psychological state vector of the student at the current time based on the adaptive behavior code, physiological data and environmental data, and output the current student's comprehensive psychological index. The elasticity coefficient module is used to obtain the estimated value of the four-dimensional psychological state vector and the estimated psychological comprehensive index of the current student under the second time window based on the four-dimensional psychological state vector and the psychological comprehensive index; and to construct the dual-time psychological elasticity coefficient based on the actual value and the estimated value of the four-dimensional psychological state vector and the psychological comprehensive index. The crisis scoring module is used to calculate six membership functions of the current student based on the estimated value of the four-dimensional psychological state vector, the estimated comprehensive psychological index, and the estimated rate of change of the comprehensive psychological index; to obtain a fuzzy rule base based on the estimated value of the four-dimensional psychological state vector, the estimated comprehensive psychological index, and the estimated rate of change of the comprehensive psychological index; and to obtain the current student's psychological crisis score based on the fuzzy rule base and the six membership functions. The graded early warning module is used to obtain personalized psychological early warning thresholds for different students based on dual-time psychological elasticity coefficients, and to classify and manage students' psychological crises by graded and classified them through personalized psychological early warning thresholds and psychological crisis scores.

[0018] A first time window is set based on the current collection frequency of students' five-key data. Physiological data, environmental data, and raw behavioral flow of the current student within the first time window are obtained based on the first time window. Physiological data includes the student's heart rate cycle changes, skin conductance response, and respiratory rate. Environmental data includes light intensity, noise intensity, and social density-related quantities. Raw behavioral flow includes the student's hand skeletal joint sequence and hand trajectory in a time-continuous state. Based on the first time window and the raw behavioral flow within the first time window, a boundary detection function is used to construct the current student's raw behavioral segment. The raw behavioral segment includes a complete behavioral unit of the current student.

[0019] It should be explained that physiological data is collected in real time by wearable smart bracelets worn by students, specifically including changes in heart rate cycles, skin conductance, and respiratory rate. Heart rate cycles are represented by the root mean square of the difference between adjacent normal heartbeats, and skin conductance is represented by real-time values ​​of skin conductivity. Environmental data is collected synchronously by multiple sensors deployed in the classroom, specifically including light intensity, noise intensity, and social density-related quantities. The first time window is obtained by pre-setting the collection frequency of student five-key data; in this embodiment, the collection frequency is set to ten frames per second, and the length of the first time window is set to ten seconds. The calculation formulas for the original behavior flow and boundary detection function are as follows: ; ; in, For the original behavior flow, The three-dimensional coordinates of 25 key points, Two-dimensional coordinates of 21 key hand points. For logical OR, To identify behavioral pattern mutations based on Bayesian change point detection, For indicator functions, For the Frobenius norm, For a moment The coordinate matrix of the skeletal joints has dimensions of 25*3. For a moment The coordinate matrix of the skeletal joints, The motion threshold is set to 0.3. The criteria for determining the original behavior segment are: the student's hand skeletal joint sequence and hand trajectory in a continuous time state. The hand skeletal joint sequence contains the three-dimensional spatial coordinates of 21 key points of each hand. The sampling frequency is consistent with the frame rate of the first time window. The hand trajectory consists of the motion paths of the above 21 key points. The change curve of the student's hand position within the first time window is recorded. The boundary detection function includes motion threshold detection and Bayesian change point detection. The specific implementation of motion threshold detection is: for each frame, calculate the Frobenius norm difference between the current frame and the previous frame's skeletal joint coordinate matrix. Thresholding based on the Frobenius norm difference In comparison, the specific calculation method for Bayesian variable point detection is as follows: A 20-dimensional behavioral feature vector is defined, including the hand center position, velocity and acceleration, head orientation angle, and limb joint angles. A preset sliding time window with a length of 30 frames is used. The Bayesian factor is calculated using the behavioral feature vector, where the formula for calculating the Bayesian factor is: ; in, Bayesian factor, Assuming there are no changing points among all data points within the sliding time window, Assuming there exists a turning point among all the data within the sliding time window, Assumption The marginal likelihood, Assumption The marginal likelihood is determined based on the Bayes factor and the log-Bayes factor threshold. If the Bayes factor is greater than the log-Bayes factor threshold (which is a fixed value of 2.0), the Bayesian change point detection identifies a behavioral pattern mutation with a value of 1. The results of motion threshold detection and Bayesian change point detection are then logically ORed to construct a boundary detection function.

[0020] The original behavioral fragments of each student in the special student group are mapped to adaptive behavior codes using an annotation tool. Adaptive behavior coding involves classifying the original behavioral fragments through annotation to obtain four coding components for the current student: motor, social, sensory, and emotional components. Based on the adaptive behavior codes, physiological data, and environmental data, a four-dimensional coupled model of the current student's psychological state is constructed. This model calculates the student's behavioral activation, physiological arousal, environmental support, and social willingness based on the four coding components. These components are then used to construct a four-dimensional psychological state vector for the current student at the current moment. A weighted summation of this four-dimensional psychological state vector yields the student's comprehensive psychological index. The formulas for calculating behavioral activation, physiological arousal, environmental support, and social willingness are as follows: ; ; ; ; in, For behavioral activation, Physiological arousal level For environmental support, For social willingness, The hyperbolic tangent function is used to map the integral result to the interval [0,1). For the current moment, The length of the integration time window. For integration variables, For at any time The student's instantaneous average speed of motion The sigmoid function is used to map linear combinations to the interval (0,1). The weighting coefficients for heart rate cycle variations. This is the weighting coefficient for the skin conductance response. For students exist Heart rate cycle changes over time. For students exist skin conductance levels at any given time For reference speed, For all students, the minimum and maximum heart rate cycle variations in historical data are given. For all students, the minimum and maximum skin conductance response levels in the historical data are given. for The ambient light intensity at any given time The preset upper limit of ambient light intensity, for The linear sound pressure level at time t. The preset maximum permissible sound pressure level, for The minimum Euclidean distance between students at any given time and other students. The preset maximum social distancing threshold, It is a multilayer perceptron. Original behavioral fragments social weight For the first A fragment of original behavior.

[0021] It needs to be explained that the specific methods for obtaining the upper limit of ambient light intensity are as follows: The maximum value is obtained based on the national building lighting design standards and the glare evaluation system for lighting projects. Because it is for children with special disabilities, the maximum value is reduced by a factor of 10 through individualized calibration experiments. The specific method for obtaining the maximum permissible sound pressure level is based on the building design standards for special education schools. The specific method for obtaining the maximum social distance threshold is to obtain the minimum social distance by statistically analyzing historical data from the past month. The specific method for obtaining the reference speed is to take 10 consecutive moments, calculate the instantaneous average motion speed at each moment, and use the average value as the reference speed. The annotation includes a spatiotemporal graph convolutional network and a dictionary embedding layer. The spatiotemporal graph convolutional network consists of 6 convolutional modules, each containing a graph convolutional layer and a temporal convolutional layer. The graph convolutional layer is defined using 25 nodes from the human body's natural connectivity graph. The parameters of the temporal convolutional layer are a one-dimensional convolutional kernel with a size of 5 and a stride of 1. The feature channels of the 6 convolutional modules are 64, 64, and 128, respectively. 128, 256, 256; The dictionary embedding layer represents the number of categories for different obstacle types. Its function is to convert discrete behavior category symbols into learnable dense vector representations and obtain independent embedding matrices for different obstacle types. The specific calculation method for classifying the original behavior fragments is as follows: extract the skeleton and hand sequences from the fragments and normalize the coordinates; calculate the feature vectors through a spatiotemporal graph convolutional encoder; select the corresponding dictionary embedding layer according to the student's obstacle type to obtain the category label as the component name; and select the category with the highest probability as the encoding value of each component. The perceptron's structural parameters are a first fully connected layer, a second fully connected layer, and an output fully connected layer. The first fully connected layer has an input dimension of 8 and an output dimension of 16, with the activation function being ReLU. The second fully connected layer has an input dimension of 16 and an output dimension of 8, with the activation function being ReLU. The output fully connected layer has an input dimension of 8 and an output dimension of 1, with the activation function being Sigmoid. The loss function is the mean squared error of the training data, the optimizer is Adam, and the learning rate is 0.001; The specific method for obtaining the maximum and minimum values ​​of heart rate cycle variation and skin conductance response level is to perform data statistics on historical data within one month, sort the heart rate cycle variation and skin conductance response level in the historical data, and obtain the maximum and minimum values ​​of heart rate cycle variation and skin conductance response level. The labeler is a pre-trained obstacle type perception labeler, specifically optimized for the behavioral characteristics of students with disabilities. For each original behavioral segment, the labeler obtains four encoded components of the current student through a classification method. The four encoded components include a motor component, a social component, a sensory component, and an emotional component, among which the motor component... The components are used to describe the student's somatic movement patterns in the behavioral segment, the social component to describe the student's social interaction tendencies in the behavioral segment, the sensory component to describe the student's sensory regulation behavior in the behavioral segment, and the emotional component to describe the student's emotional regulation state in the behavioral segment. The instantaneous average movement velocity is obtained by constructing a three-dimensional coordinate vector of skeletal joints and calculating it through the time difference of the skeletal joint coordinates. The weighting coefficients of heart rate cycle variation and skin conductance response are obtained by pre-setting them through offline experiments and then fitting them with the least squares method based on historical averages. The specific calculation formula is as follows: ; in, The weighting coefficients for heart rate cycle variations. This is the weighting coefficient for the skin conductance response. For the number of trials, This is the current test number. For the Sigmoid function, To assess the true physiological arousal level by psychological experts, the weighting coefficient for heart rate cycle changes is set to 0.4, and the weighting coefficient for skin conductance is set to 0.6 in this embodiment. The maximum social distance threshold is obtained by fine-tuning it based on the current student's body size, with a range of 4 to 5 meters. The values ​​for behavioral activation, physiological arousal, environmental support, and social willingness are all between zero and one. Behavioral activation is a quantitative indicator that measures the intensity of physical movement and the degree of behavioral activity of students within a specified time window; the higher the value, the more intense the student's activity. Physiological arousal is a comprehensive indicator that measures the activation level of the student's autonomic nervous system; the higher the value, the higher the level of physiological arousal. Environmental support is a comprehensive indicator that measures the degree of support the current physical environment provides for the student's mental health; the higher the value, the more friendly the environment. Social willingness is an indicator that measures the student's tendency to actively engage in social interaction with others; the higher the value, the stronger the student's willingness to actively socialize.

[0022] Based on the four-dimensional psychological state vector, a unidirectional analysis is performed. Using pre-defined behavioral, physiological, environmental, and social weights, a weighted summation method is used to calculate the current student's comprehensive psychological index. This comprehensive psychological index represents the student's current mental health level. Furthermore, based on this index, the rate of change is calculated using a differential method. The formula for calculating the comprehensive psychological index is as follows: ; in, For behavioral activation, Physiological arousal level For environmental support, For social willingness, These are the weighting coefficients for behavioral activation, physiological arousal, environmental support, and social willingness. This is the Sigmoid function.

[0023] It needs to be explained that the state transition model includes a nonlinear mapping function pre-trained using historical student data. This function accepts the current four-dimensional psychological state vector as input and outputs a psychological transition state model describing the evolution of the state. The input is the current psychological transition state, and the output is the predicted value of the psychological transition state after a preset time period. The four-dimensional psychological state vector and preset behavioral, physiological, environmental, and social weights are used to calculate the current student's comprehensive psychological index through a weighted summation. The specific methods for obtaining the behavioral, physiological, environmental, and social weights are as follows: they are preset by fitting historical data, with the behavioral weight set to 0.3, the physiological weight to 0.25, the environmental weight to 0.2, and the social weight to 0.25, and the sum of the four weights being 1. The comprehensive psychological index ranges from 0 to 1, and the rate of change of the psychological index is defined as the first derivative of the comprehensive psychological index with respect to time.

[0024] Based on the students' four-dimensional psychological state vector at the current moment, a psychological state attention network for a special student group on the Five-fold Healthy Life Cycle Management Platform is constructed. The psychological state attention network includes teacher attention network and student attention network. The estimated residuals of the four encoded components in the four-dimensional psychological state vector are calculated through the psychological state attention network and the original behavioral flow, physiological data and environmental data. The actual value of the four-dimensional psychological state vector at the current moment is obtained by calculating the four-dimensional psychological state vector and the estimated residuals of the four encoded components in the four-dimensional psychological state vector. The actual psychological comprehensive index and the actual psychological index change rate at the current moment are obtained through the actual value of the four-dimensional psychological state vector at the current moment. A second time window is preset according to the first time window. The estimated value of the four-dimensional psychological state vector under the second time window is obtained based on the actual value of the four-dimensional psychological state vector at the current moment and the psychological state transfer model. The estimated psychological comprehensive index and the estimated psychological comprehensive index change rate under the second time window are calculated based on the estimated value of the four-dimensional psychological state vector under the second time window.

[0025] It needs to be explained that the estimated residuals of the four encoded components are a four-dimensional vector. Each component corresponds to the behavioral activation residual, physiological arousal residual, environmental support residual, and social willingness residual, respectively. The actual value of the four-dimensional mental state vector equals the theoretically calculated value of the four-dimensional mental state vector plus the estimated residuals of the four encoded components. The theoretically calculated value is obtained by using the formulas for behavioral activation, physiological arousal, environmental support, and social willingness. For example, if the behavioral activation in the theoretically calculated value is 0.50, and the behavioral activation residual output by the student's attention network is 0.12, then the actual value of behavioral activation is 0.62. The actual rate of change of the psychological index is calculated using the numerical difference method in calculus: subtracting the actual psychological comprehensive index of the previous moment from the actual psychological comprehensive index of the current moment, and then dividing by the time interval between the two moments. The first time window has a set length of 10 seconds and is used to collect real-time data up to the current moment. The second time window is obtained by doubling the first time window. The length of the second time window is greater than that of the first time window. In this embodiment, the second time window is preset to be 60 seconds. The psychological state transfer model uses the actual value of the four-dimensional psychological state vector at the current moment as the initial condition and recursively calculates the state at future moments according to the preset state evolution law. Starting from the current moment, it recursively calculates 60 times with a step size of one second to obtain the estimated value of the four-dimensional psychological state vector for each second from the current moment to the next 60 seconds. For example, if the actual value of the behavioral activation at the current moment is 0.62 and is on the rise, the psychological state transfer model will predict that the behavioral activation will continue to rise to 0.75 in the next 30 seconds and then slowly fall back due to fatigue and environmental adjustment.

[0026] The teacher residuals of the four encoded components in the four-dimensional mental state vector are used as soft labels for the teacher attention network. The estimated residuals of the four encoded components in the four-dimensional mental state vector are obtained through the student attention network. The fusion calculation formula for the teacher attention network and the loss function formula for the student attention network are as follows: ; ; ; ; ; ; in, The fusion feature matrix of the teacher attention network. For teacher residuals in teacher attention networks, Let be the loss function of the student attention network. Let KL divergence loss function be used. Let the mean squared error loss function be . For boundary regularization loss function, and To balance hyperparameters, , and The projection matrix of the cross-modal attention layer. , and The modal characteristic matrix, This represents the dimension of the attention layer, with a value of 64. For multilayer perceptron functions, This represents the total number of moments within the first time window. This represents the number of moments within the first time window; For the softmax function, This is the index of the four-dimensional mental state components in the residual vector. For distillation temperature parameters, The first in the student attention network output The estimated residuals of the component index, For the output of the teacher's attention network, the first Teacher residuals of component indexes, For students exist In the moment A four-dimensional mental state vector with component indexes. It is an absolute value.

[0027] It should be explained that the specific method for obtaining the balancing hyperparameters is to use the validation set to search for hyperparameters within a fixed range during the training process of the teacher-student distillation network. and A grid search is performed to select the combination that minimizes the estimation error of the student network's mental state on the validation set. In this embodiment, the values ​​are 0.5 and 0.1. The distillation temperature parameter controls the smoothness of the soft labels during knowledge distillation. Specifically, it is obtained by selecting the value that minimizes the KL divergence on the validation set from {1.0, 2.0, 3.0, 4.0, 5.0} through cross-validation. In this embodiment, the value is 2.0. Three modality feature matrices are extracted based on the three types of data. This is the behavioral modality feature matrix, with dimensions equal to the length of the first time window multiplied by the feature dimension. This is the physiological modality feature matrix, with dimensions equal to the length of the first time window multiplied by the feature dimension. The environmental modality feature matrix has a dimension equal to the length of the first time window multiplied by the feature dimension. The meaning of the teacher attention network includes the behavioral modality feature matrix. The result after query projection is used as the query, with the physiological modality feature matrix as the basis. The result after key projection is used as the key. The similarity between the query and the key is calculated and normalized to obtain the attention weight. Then, the attention weight is used to apply the value projection to the environmental modality feature matrix. We perform a weighted summation to obtain a fusion feature matrix that incorporates information from the three modalities.

[0028] The first-time psychological elasticity coefficient under the first time window is calculated based on the actual value of the four-dimensional psychological state vector and the actual psychological comprehensive index. The second-time psychological elasticity coefficient under the second time window is calculated based on the first-time psychological elasticity coefficient, the estimated value of the four-dimensional psychological state vector, and the estimated psychological comprehensive index. The dual-time psychological elasticity coefficient of the current student is calculated based on the first-time psychological elasticity coefficient, the second-time psychological elasticity coefficient, the first-time psychological weight, and the second-time psychological weight. The dual-time psychological elasticity coefficient represents the student's psychological elasticity coefficient under both the first and second time windows. The formulas for calculating the first-time psychological elasticity coefficient, the second-time psychological elasticity coefficient, and the dual-time psychological elasticity coefficient are as follows: ; ; ; in, The first-time psychological resilience coefficient, This is the second time-based psychological elasticity coefficient. This is a two-time psychological elasticity coefficient. This refers to the number of moments within the first time window. For the Euclidean norm, This represents the total number of moments within the first time window. In order to be in A four-dimensional mental state vector at any given moment. This is the baseline four-dimensional mental state vector. For the second time window The estimated value of the four-dimensional mental state vector at time t. For error terms, Here is the uncertainty coefficient. This is a sensitivity hyperparameter. For recovery time, For the maximum deviation, For reference only.

[0029] It needs to be explained that the maximum deviation is the maximum deviation between the four-dimensional mental state vector and the baseline four-dimensional mental state vector. Specifically, it is obtained by statistically analyzing historical data to find the maximum difference between the two four-dimensional mental state vectors as the maximum deviation. The reference time is obtained by setting a fixed value of 1. The recovery time is the time required to recover from the current deviation state to the baseline state. Specifically, in a safe and controlled environment, a standardized stress-induced task is used to significantly deviate the student's state from the baseline. The student's four-dimensional mental state vector is monitored in real time, and the peak deviation and its occurrence time are recorded. The stress source is removed, and the student is allowed to recover naturally. Monitoring continues until the Euclidean distance between the four-dimensional mental state vector and the baseline four-dimensional mental state vector is first less than a preset distance threshold. This current time is recorded as the experimental recovery time. Based on statistical analysis of the experimental recovery times of students with similar disabilities, the average experimental recovery time is calculated as the recovery time. The distance threshold is preset based on clinical experience, and in this embodiment, it is set to 0.2. The first-time psychological elasticity coefficient is the average rate at which students recover from a deviation from the baseline state to the baseline state within the first time window, and the second-time psychological elasticity coefficient is the average rate at which students recover from a deviation from the baseline state to the baseline state within the second time window. The values ​​of the first-time psychological elasticity coefficient, the second-time psychological elasticity coefficient, and the dual-time psychological elasticity coefficient are all between 0 and 1. The error term is obtained by looking up a table, and in this embodiment, it is set to 0.001. The uncertainty coefficient is obtained by calculating the ratio of the trace of the process noise covariance of the current state to the trace of the state covariance. The larger the uncertainty coefficient, the less reliable the estimation of the current state is. The value range is between 0 and 1. The second-time psychological elasticity coefficient is the ratio obtained by comparing the degree of deviation of the state at the future time with the degree of deviation of the state at the current time. If the future deviation is less than the current deviation, the ratio is less than 1, indicating that the student's state is regressing to the baseline. The sensitivity hyperparameter controls the sensitivity of the weights in the dual-time elasticity coefficient fusion process to the uncertainty coefficient. The sensitivity hyperparameter is obtained by repeatedly simulating the correlation between the dual-time elasticity coefficients and the subsequent actual recovery effect on a historical dataset, and selecting the sensitivity hyperparameter that maximizes the correlation. In this embodiment, the sensitivity hyperparameter is set to 5. In this embodiment, if the current uncertainty coefficient is 0.3, then... The value is 0.7. The sensitivity hyperparameter multiplied by 0.7 equals 3.5. Therefore, the first time weight is 0.97, the second time weight is 0.03, and the dual-time elasticity coefficient is approximately 0.97 multiplied by the first time weight plus 0.03 multiplied by the second time weight. If the first time weight is 0.06 and the second time weight is 0.045, then the dual-time elasticity coefficient is approximately 0.0596, which is slightly lower than the first time weight, reflecting a slight reduction in the future recovery capability.

[0030] The first, second, third, and fourth membership degrees are calculated based on the estimated four-dimensional psychological state vector values. The fifth membership degree is calculated based on the estimated psychological comprehensive index, and the sixth membership degree is calculated based on the estimated rate of change of the psychological comprehensive index. The first membership degree represents the degree of membership of the corresponding behavioral activation estimate to a high-activation state; the second membership degree represents the degree of membership of the corresponding physiological arousal estimate to a high-arousal state; the third membership degree represents the degree of membership of the corresponding environmental support estimate to a low-support state; and the fourth membership degree represents the degree of membership of the corresponding social willingness estimate to a low-willing state. All four membership degrees are calculated using a Gaussian membership function. A fuzzy rule base for the current student is constructed using the estimated four-dimensional psychological state vector values, the estimated psychological comprehensive index, and the estimated rate of change of the psychological comprehensive index. The fuzzy rule base is determined as follows: if behavioral activation and physiological arousal are greater than the first threshold and environmental support and social willingness are less than the second threshold, then when… The previous student is in a first-rule crisis; if the behavioral activation and physiological arousal are less than the second threshold, the current student is in a second-rule crisis; if the estimated psychological comprehensive index is less than the second threshold and the duration exceeds the psychological recovery time threshold, the current student is in a third-rule crisis; if the estimated psychological comprehensive index is less than the third threshold, the current student is in a fourth-rule crisis; if the estimated rate of change of the psychological comprehensive index is less than zero, the current student is in a fifth-rule crisis; the first, second, and third thresholds are the decision boundaries for the estimated value of the four-dimensional psychological state vector and the estimated psychological comprehensive index; and the first threshold is greater than the second threshold, which is greater than the third threshold; based on the distribution of the corresponding indicators in the historical data of the student group, the upper quartile, median, and lower quartile are taken as the first, second, and third thresholds, respectively; the calculation formulas for the first, second, third, fourth, and fifth membership degrees, as well as the membership degree calculation formula for the estimated rate of change of the psychological comprehensive index, are as follows: ; ; in, For membership degree, The independent variables of the membership function include the four component estimates of the four-dimensional mental state vector estimate and the estimated psychological composite index. The central value of the membership function. The standard deviation of the membership function. To estimate the rate of change of the comprehensive psychological index, To estimate the sixth membership degree of the rate of change of the psychological composite index, The coefficient of variation, For the change offset, The rate of change of the comprehensive psychological index is used as a reference.

[0031] It should be explained that the reference psychological composite index change rate is the average of the estimated psychological composite index change rates over historical periods. The change coefficient is the weighting of changes in the estimated psychological composite index change rate, and the change offset is the offset when the estimated psychological composite index change rate changes. The change coefficient is obtained by fitting the data through multiple sample trainings and minimizing the mean squared error. The sample data is obtained from historical data over a period of one week. The fitting calculation method is as follows: ; in, For the total number of training iterations, This is the training sequence number. The confidence level for the decrease given by experts ranges from [0,1], and in this embodiment, the coefficient of change is 5. The specific method for obtaining the change offset is to collect the estimated rate of change of the psychological comprehensive index at all times in historical data and the event labels of the psychological comprehensive index decreasing by more than 0.05 within the next 10 minutes, and fit them through a logistic regression model. In this embodiment, the value is 0.005. The first threshold, the second threshold, and the third threshold are the judgment ranges for the estimated value of the four-dimensional psychological state vector and the estimated psychological comprehensive index. The estimated value of the four-dimensional psychological state vector and the estimated psychological comprehensive index have the same range. The first threshold is greater than the second threshold, which is greater than the third threshold. The specific method for obtaining the first threshold, the second threshold, and the third threshold is to collect the estimated values ​​of behavioral activation, physiological arousal, environmental support, social willingness, and psychological comprehensive index at all effective times in the past 30 days for this special student group, and merge all historical values ​​of the five indicators of all students into a total sample set. Then, the total sample set is sorted in ascending order; the lower quartile is the 0.25×Nth quartile after sorting, serving as the third threshold, where N is the total number of samples; the median is the 0.50×Nth quartile, serving as the second threshold; and the upper quartile is the 0.75×Nth quartile, serving as the first threshold. The quartiles are automatically recalculated and the three thresholds are updated every month. The four-dimensional psychological state vector estimate includes four components: behavioral activation estimate, physiological arousal estimate, environmental support estimate, and social willingness estimate. The first membership degree corresponds to the degree of membership of the behavioral activation estimate to the high-activation state; the second membership degree corresponds to the degree of membership of the physiological arousal estimate to the high-arousal state; the third membership degree corresponds to the degree of membership of the environmental support estimate to the low-support state; and the fourth membership degree corresponds to the degree of membership of the social willingness estimate to the low-willing state. The first, second, third, and fourth membership degrees are all calculated using a Gaussian membership function, the general form of which is: In the formula These are the estimated values ​​for the corresponding components; The central value of the membership function; The standard deviation is taken as the standard deviation. The first membership parameter was obtained by collecting historical behavioral activation data from 100 special needs students in daily classrooms, recess activities, and stressful situations. The distribution of high-activation states (i.e., behavioral activation marked as significantly active by experts) was statistically analyzed, and the 50th percentile of this distribution was used as the center value. The standard deviation was taken as the estimated standard deviation of this distribution. The second membership parameter was obtained by analyzing skin conductance and heart rate variability data collected from students' wearable devices, combined with high-arousal moments of tension and excitement marked by teachers, and fitted using maximum likelihood estimation to obtain Gaussian parameters. The third membership parameter was obtained by clustering the environmental support of the environmental periods that triggered students' problem behaviors in historical data, using the low-support cluster centers as the center values. The fourth membership parameter was taken from clinical research... The mean of the distribution of social willingness corresponding to social avoidance behavior in students with autism spectrum disorder was used as the center value. The center value and standard deviation were periodically updated based on historical data of specific student groups and recalculated using unsupervised clustering or expert-annotated data fitting to improve the individual adaptability of the membership function. The specific method for obtaining the center value of the membership function was as follows: historical data fitting was performed on the center value and it was preset. The estimated center value of the psychological comprehensive index was 0.4, the estimated center value of the four-dimensional psychological state vector was 0.8, and the standard deviation was obtained by offline experimental calibration with a value of 0.15. The sixth membership degree was obtained by the membership degree calculation formula of the estimated rate of change of the psychological comprehensive index. The sixth membership degree was used to judge the fuzzy set of the decreasing trend of the rate of change.

[0032] Based on the different rule-based crises currently experienced by the students, the crisis confidence level of the current students is obtained. The crisis score at the activation of each rule crisis is calculated using the crisis confidence level and six membership functions. The maximum crisis score at the activation of each rule is taken as the current student's psychological crisis score. The formulas for calculating the crisis score and the psychological crisis score are as follows: ; ; in, Assign crisis scores to each rule-related crisis. Assess the current psychological crisis among students. For the current rule crisis The set of membership degrees involved varies depending on the rule or crisis. The current student's crisis level weight is determined based on the current rule-based crisis level. For the rule crisis sequence number, These are the first rule crisis, the second rule crisis, the third rule crisis, the fourth rule crisis, and the fifth rule crisis.

[0033] It should be explained that the specific method for obtaining the crisis level weight is as follows: In this embodiment, the crisis confidence level of the rule crisis includes: if the current student meets the first rule crisis, the crisis confidence level is 0.9; if the current student meets the second rule crisis, the crisis confidence level is 0.7; if the current student meets the third rule crisis, the crisis confidence level is 0.75; if the current student meets the fourth rule crisis, the crisis confidence level is 0.95; if the current student meets the fifth rule crisis, the crisis confidence level is 0.8. The first rule crisis involves behavioral activation membership, physiological arousal membership, and environmental factors. The first rule involves membership in environmental support and social willingness, multiplied by the minimum of these four membership values. The second rule involves membership in behavioral activation and physiological arousal, multiplied by the minimum of these two membership values. The third rule involves membership in estimated psychological comprehensive index, multiplied by 0.75. The fourth rule involves membership in estimated psychological comprehensive index, multiplied by 0.95. The fifth rule involves membership in estimated psychological comprehensive index change rate, multiplied by 0.80. If multiple rule crises are satisfied simultaneously, the average crisis confidence level is used.

[0034] Personalized psychological early warning thresholds are calculated based on dual-time psychological elasticity coefficients. A student's warning level is determined by combining these thresholds with a psychological crisis score. The criteria are as follows: if the psychological crisis score falls within the first range, the student is at the green level, indicating normal psychological health; if it falls within the second range, the student is at the blue level; if it falls within the third range, the student is at the yellow level; if it falls within the fourth range, the student is at the orange level; if it falls within the fifth range, the student is at the red level. If the psychological crisis score exceeds the personalized psychological early warning threshold, the student's warning level is set to red. The first range corresponds to the green level, indicating a normal psychological state; the second range corresponds to the blue level, indicating a slight psychological risk; the third range corresponds to the yellow level, indicating a moderate psychological risk; the fourth range corresponds to the orange level, indicating a high psychological risk; and the fifth range corresponds to the red level, indicating a severe psychological crisis. The first, second, third, and fourth ranges are obtained using the equal-frequency binning method and the proportional division method.

[0035] It should be explained that the first, second, third, fourth, and fifth ranges are based on the psychological crisis scoring criteria, with the first range being less than the second, third, fourth, and fifth ranges. The specific method for obtaining the first, second, third, fourth, and fifth ranges is to collect all valid psychological crisis scoring records of this specific student group over the past 90 days, collecting data from each student every 5 minutes. These data are then merged into a total sample set, and significant outliers are removed to obtain a clean sample set. The initial quantiles are calculated using the equal-frequency binning method, which involves first sorting the sample set by values ​​from smallest to largest to determine five levels, with each level expected to have a 20% sample share. The 20th, 40th, 60th, and 80th quantiles are then calculated. It is the value of the 0.2 multiplied by the Nth digit; that is... It is the value of 0.4 multiplied by N digits; that is... It is the value of 0.6 multiplied by N digits; that is... The value is 0.8 multiplied by N digits. The initial range is divided into: the first range. Second range Third range Fourth Scope Fifth Scope The proportional division method directly divides the [0,1] interval into 5 equal subintervals, each with a length of 0.2. A weighted fusion method is used to determine the final range boundary; let the boundary obtained by the equal-frequency method be... The boundary obtained by the proportional method is The final boundary is In the formula The equilibrium coefficient has a value range of [0,1]. In this embodiment, it is taken as... The personalized psychological warning threshold is the minimum crisis score required for a student to receive a red alert in the psychological assessment. When a student's psychological crisis score exceeds the personalized psychological warning threshold, the system will forcibly trigger a red-level warning. The personalized psychological warning threshold ranges from 0.5 to 0.8; the psychological crisis score ranges from 0 to 1. If the psychological crisis score is in the first range, the student is in the green level, indicating that the student's current mental health is good and no intervention is needed, only routine recording. If the psychological crisis score is in the second range, the student is in the blue level, indicating that the student has a slight mental health risk and does not yet require professional intervention, but teachers need to monitor the student's condition weekly. If the psychological crisis score is in the third range... If the score falls within the range of four, the student is at the yellow level, indicating a moderate level of mental health risk. Regular intervention from school counselors and resource teachers is required, such as weekly counseling and behavioral intervention. If the score is in the fourth range, the student is at the orange level, indicating a high level of mental health risk. In addition to counselor intervention, parents should be notified promptly for collaborative home-school intervention. If the score is in the fifth range, the student is at the red level, indicating a high level of mental health risk and a state of mental crisis. An immediate crisis intervention team should be activated, a professional psychologist should be called for an emergency assessment, and referral to a hospital's psychiatric or psychological department is recommended. The formula for calculating the personalized mental health warning threshold is: ; in, For personalized psychological early warning thresholds, The minimum personalized psychological early warning threshold is set to 0.5. This is the psychological elasticity coefficient over two time periods.

[0036] The above embodiments are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Therefore, all equivalent changes made to the content described in the claims of the present invention should be included within the scope of the claims of the present invention.

Claims

1. A health management platform for children and adolescents' five-stage development cycle, characterized in that, include: The original segment module is used to preset the first time window, acquire the student's physiological data, environmental data and original behavior flow within the first time window, and construct the original behavior segment based on the first time window and the original behavior flow. The psychological index module is used to map the original behavioral fragments to the adaptive behavior code of the current student, calculate the four-dimensional psychological state vector of the student at the current time based on the adaptive behavior code, physiological data and environmental data, and output the comprehensive psychological index of the current student. The elasticity coefficient module is used to obtain the estimated value of the four-dimensional psychological state vector and the estimated psychological comprehensive index of the current student under the second time window based on the four-dimensional psychological state vector and the psychological comprehensive index; and to construct the dual-time psychological elasticity coefficient based on the actual value and the estimated value of the four-dimensional psychological state vector and the psychological comprehensive index. The crisis scoring module is used to calculate six membership functions of the current student based on the estimated value of the four-dimensional psychological state vector, the estimated comprehensive psychological index, and the estimated rate of change of the comprehensive psychological index; to obtain a fuzzy rule base based on the estimated value of the four-dimensional psychological state vector, the estimated comprehensive psychological index, and the estimated rate of change of the comprehensive psychological index; and to obtain the current student's psychological crisis score based on the fuzzy rule base and the six membership functions. The graded early warning module is used to obtain personalized psychological early warning thresholds for different students based on the dual-time psychological elasticity coefficients, and to classify and manage students' psychological crises by graded approach based on personalized psychological early warning thresholds and psychological crisis scores.

2. The five-cycle health management platform for children and adolescents according to claim 1, characterized in that, The preset first time window acquires the student's physiological data, environmental data, and original behavior flow within the first time window. Based on the first time window and the original behavior flow, an original behavior segment is constructed. This includes: setting the first time window based on the current student's five-key data acquisition frequency; acquiring the current student's physiological data, environmental data, and original behavior flow within the first time window; the physiological data including the student's heart rate cycle changes, skin conductance response, and respiratory rate; the environmental data including light intensity, noise intensity, and social density-related quantities; and the original behavior flow including the student's hand skeletal joint sequence and hand trajectory in a continuous time state. Based on the first time window and the original behavior flow within the first time window, an original behavior segment of the current student is constructed using a boundary detection function. The original behavior segment includes a complete behavioral unit of the current student.

3. The five-cycle health management platform for children and adolescents according to claim 1, characterized in that, The process of mapping the original behavioral fragments to the adaptive behavior code of the current student, calculating the student's four-dimensional psychological state vector at the current time based on the adaptive behavior code, physiological data, and environmental data, and outputting the current student's comprehensive psychological index includes: mapping the original behavioral fragments of each student in the special student group to the current student's adaptive behavior code through a labeler; the adaptive behavior code includes classifying the original behavioral fragments through labeling to obtain four coding components of the current student, including a motor component, a social component, a sensory component, and an emotional component; constructing a four-dimensional coupled model of the current student's psychological state based on the adaptive behavior code, physiological data, and environmental data; the four-dimensional coupled model of the psychological state includes calculating the current student's behavioral activation, physiological arousal, environmental support, and social willingness based on the four coding components in the adaptive behavior code; constructing the current student's behavioral activation, physiological arousal, environmental support, and social willingness into a four-dimensional psychological state vector of the current student at the current time; and calculating the current student's comprehensive psychological index by weighted summation based on the four-dimensional psychological state vector, wherein the calculation formulas for behavioral activation, physiological arousal, environmental support, and social willingness are as follows: ； ； ; ； in, For behavioral activation, Physiological arousal level For environmental support, For social willingness, The hyperbolic tangent function is used to map the integral result to the interval [0,1). For the current moment, The length of the integration time window. For integration variables, For at any time The student's instantaneous average speed of motion The sigmoid function is used to map linear combinations to the interval (0,1). The weighting coefficients for heart rate cycle variations. This is the weighting coefficient for the skin conductance response. For students exist Heart rate cycle changes over time. For students exist skin conductance levels at any given time For reference speed, For all students, the minimum and maximum heart rate cycle variations in historical data are given. For all students, the minimum and maximum skin conductance response levels in the historical data are given. for The ambient light intensity at any given time The preset upper limit of ambient light intensity, for The linear sound pressure level at time t. The preset maximum permissible sound pressure level, for The minimum Euclidean distance between students at any given time and other students. The preset maximum social distancing threshold, It is a multilayer perceptron. Original behavioral fragments social weight For the first A fragment of original behavior.

4. The five-cycle health management platform for children and adolescents according to claim 3, characterized in that, The step of obtaining the current student's comprehensive psychological index by weighted summation based on the four-dimensional psychological state vector includes: performing a unidirectional analysis based on the four-dimensional psychological state vector; calculating the current student's comprehensive psychological index by weighted summation based on preset behavioral weights, physiological weights, environmental weights, and social weights of the unidirectional analysis; the comprehensive psychological index represents the current student's mental health level; and calculating the rate of change of the psychological index using a differential method based on the comprehensive psychological index, wherein the formula for calculating the comprehensive psychological index is: ； in, For behavioral activation, Physiological arousal level For environmental support, For social willingness, These are the weighting coefficients for behavioral activation, physiological arousal, environmental support, and social willingness. This is the Sigmoid function.

5. The five-cycle health management platform for children and adolescents according to claim 1, characterized in that, The process of obtaining the estimated value of the four-dimensional psychological state vector and the estimated psychological comprehensive index of the current student in the second time window based on the four-dimensional psychological state vector and the psychological comprehensive index includes: constructing a psychological state attention network for a special student group on the five-cycle health management platform based on the student's four-dimensional psychological state vector at the current moment. The psychological state attention network includes a teacher attention network and a student attention network. The estimated residuals of the four encoded components in the four-dimensional psychological state vector are calculated using the psychological state attention network and the original behavioral flow, physiological data, and environmental data. The actual value of the four-dimensional psychological state vector at the current moment is obtained by calculating the four-dimensional psychological state vector and the estimated residuals of the four encoded components in the four-dimensional psychological state vector. The actual psychological comprehensive index and the actual rate of change of the psychological index at the current moment are obtained using the actual value of the four-dimensional psychological state vector at the current moment. A second time window is preset according to the first time window. The estimated value of the four-dimensional psychological state vector in the second time window is obtained based on the actual value of the four-dimensional psychological state vector at the current moment and the psychological state transfer model. The estimated psychological comprehensive index and the estimated rate of change of the psychological comprehensive index in the second time window are calculated based on the estimated value of the four-dimensional psychological state vector in the second time window.

6. The five-cycle health management platform for children and adolescents according to claim 5, characterized in that, The mental state attention network includes a teacher attention network and a student attention network, comprising: using the teacher residuals of the four encoded components in the four-dimensional mental state vector as soft labels for the teacher attention network; and obtaining the estimated residuals of the four encoded components in the four-dimensional mental state vector through the student attention network. The fusion calculation formula for the teacher attention network and the loss function formula for the student attention network are as follows: ； ； ； ； ; ； in, The fusion feature matrix of the teacher attention network. For teacher residuals in teacher attention networks, Let be the loss function of the student attention network. Let KL divergence loss function be used. Let the mean squared error loss function be . For boundary regularization loss function, and To balance hyperparameters, , and The projection matrix of the cross-modal attention layer. , and The modal characteristic matrix, This represents the dimension of the attention layer, with a value of 64. For multilayer perceptron functions, This represents the total number of moments within the first time window. This represents the number of moments within the first time window; For the softmax function, This is the index of the four-dimensional mental state components in the residual vector. For distillation temperature parameters, The first in the student attention network output The estimated residuals of the component index, For the output of the teacher's attention network, the first Teacher residuals of component indexes, For students exist In the moment A four-dimensional mental state vector with component indexes. It is an absolute value.

7. The five-cycle health management platform for children and adolescents according to claim 1, characterized in that, The construction of a dual-time psychological elasticity coefficient based on the actual and estimated values ​​of the four-dimensional psychological state vector and the psychological comprehensive index includes: calculating the first-time psychological elasticity coefficient under the first time window based on the actual value of the four-dimensional psychological state vector and the actual psychological comprehensive index; calculating the second-time psychological elasticity coefficient under the second time window based on the first-time psychological elasticity coefficient, the estimated value of the four-dimensional psychological state vector, and the estimated psychological comprehensive index; and calculating the current student's dual-time psychological elasticity coefficient based on the first-time psychological elasticity coefficient, the second-time psychological elasticity coefficient, the first-time psychological weight, and the second-time psychological weight. The dual-time psychological elasticity coefficient represents the current student's psychological elasticity coefficient under the first and second time windows. The calculation formulas for the first-time psychological elasticity coefficient, the second-time psychological elasticity coefficient, and the dual-time psychological elasticity coefficient are as follows: ； ； ； in, The first-time psychological resilience coefficient, This is the second time-based psychological elasticity coefficient. This is a two-time psychological elasticity coefficient. This refers to the number of moments within the first time window. For the Euclidean norm, This represents the total number of moments within the first time window. In order to be in The four-dimensional mental state vector at any given moment. The baseline is a four-dimensional mental state vector. For the second time window The estimated value of the four-dimensional mental state vector at time t. For the error term, Here is the uncertainty coefficient. This is a sensitivity hyperparameter. For recovery time, For the maximum deviation, For reference only.

8. The five-cycle health management platform for children and adolescents according to claim 1, characterized in that, The process of calculating six membership functions for the current student based on the estimated value of the four-dimensional psychological state vector, the estimated psychological comprehensive index, and the estimated rate of change of the psychological comprehensive index, and obtaining a fuzzy rule base based on the estimated value of the four-dimensional psychological state vector, the estimated psychological comprehensive index, and the estimated rate of change of the psychological comprehensive index, includes: calculating the first, second, third, and fourth membership degrees based on the estimated value of the four-dimensional psychological state vector; calculating the fifth membership degree based on the estimated psychological comprehensive index; and calculating the sixth membership degree based on the estimated rate of change of the psychological comprehensive index; the first membership degree is the corresponding behavioral stimulus. The first membership degree is the degree of membership of the activity estimate to the high-activation state; the second membership degree is the degree of membership of the corresponding physiological arousal estimate to the high-arousal state; the third membership degree is the degree of membership of the corresponding environmental support estimate to the low-support state; and the fourth membership degree is the degree of membership of the corresponding social willingness estimate to the low-willing state. All four membership degrees are calculated using a Gaussian membership function. A fuzzy rule base for the current student is constructed using the four-dimensional psychological state vector estimate, the estimated psychological comprehensive index, and the estimated rate of change of the psychological comprehensive index. The fuzzy rule base determines the following: if behavioral activation and physiological arousal are greater than the first threshold and environmental support and social willingness are less than the second threshold, the current student is in a first rule crisis; if behavioral activation and physiological arousal are less than the second threshold, the current student is in a second rule crisis; if the estimated psychological comprehensive index is less than the second threshold and the duration exceeds the psychological recovery time threshold, the current student is in a third rule crisis; if the estimated psychological comprehensive index is less than the third threshold, the current student is in a fourth rule crisis; if the estimated rate of change of the psychological comprehensive index is less than zero, the current student is in a fifth rule crisis. The first, second, and third thresholds are the determination boundaries for the estimated four-dimensional psychological state vector and the estimated psychological comprehensive index; and the first threshold is greater than the second threshold, which is greater than the third threshold. Based on the distribution of corresponding indicators in the historical data of the student group, the upper quartile, median, and lower quartile are taken as the first, second, and third thresholds, respectively. The calculation formulas for the first, second, third, fourth, and fifth membership degrees, as well as the membership degree calculation formula for the estimated rate of change of the psychological comprehensive index, are as follows: ； ； in, For membership degree, The independent variables of the membership function include the four component estimates of the four-dimensional mental state vector estimate and the estimated psychological composite index. The central value of the membership function. The standard deviation of the membership function. To estimate the rate of change of the psychological composite index, To estimate the sixth membership degree of the rate of change of the psychological composite index, The coefficient of variation, For the change offset, The rate of change of the comprehensive psychological index is used as a reference.

9. A five-cycle health management platform for children and adolescents according to claim 1, characterized in that, The step of obtaining the current student's psychological crisis score based on the fuzzy rule base and six membership functions includes: obtaining the current student's crisis confidence level based on the different rule-based crises the current student is in; calculating the crisis score when each rule-based crisis is activated using the crisis confidence level and the six membership functions; and taking the maximum crisis score when each rule is activated as the current student's psychological crisis score. The formulas for calculating the crisis score and the psychological crisis score are as follows: ； ； in, Assign crisis scores to each rule-related crisis. Assess the current psychological crisis among students. For the current rule crisis The set of membership degrees involved varies depending on the rule or crisis. The current student's crisis level weight is determined based on the current rule-based crisis level. For the rule crisis sequence number, These are the first rule crisis, the second rule crisis, the third rule crisis, the fourth rule crisis, and the fifth rule crisis.

10. A five-cycle health management platform for children and adolescents according to claim 1, characterized in that, The process of obtaining personalized psychological early warning thresholds for different students based on the dual-time psychological elasticity coefficient, and classifying and managing students' psychological crises through personalized psychological early warning thresholds and psychological crisis scores, further includes: calculating personalized psychological early warning thresholds based on the dual-time psychological elasticity coefficients, and determining the student's warning level based on the personalized psychological early warning thresholds and psychological crisis scores. The determination criteria are as follows: if the psychological crisis score is in the first range, the current student is at the green level, and the current student is considered psychologically normal; if the psychological crisis score is in the second range, the current student is at the blue level; if the psychological crisis score is in the third range, the current student is at the yellow level; if the psychological crisis score is in the fourth ... The current student is at the orange level; if the psychological crisis score is in the fifth range, the current student is at the red level; if the psychological crisis score is greater than the personalized psychological warning threshold, the current student's warning level is set to red. The first range corresponds to the green level, indicating that the student's psychological state is normal; the second range corresponds to the blue level, indicating that there is a slight psychological risk; the third range corresponds to the yellow level, indicating that there is a moderate psychological risk; the fourth range corresponds to the orange level, indicating that there is a high psychological risk; the fifth range corresponds to the red level, indicating that there is a serious psychological crisis. The first, second, third, and fourth ranges are obtained according to the equal-frequency binning method and the equal-proportional division method.

Images