High school student psychological assessment method and system based on artificial intelligence

Abstract

The invention provides a psychological assessment method and system for senior high school students based on artificial intelligence, and the method comprises the steps: enabling a testee to wear an intelligent bracelet in daily learning life, obtaining the daily physiological data of the testee through the bracelet, and obtaining the ranking of all examination scores of the testee from a school; performing data statistics, feature extraction and analysis on the acquired physiological data and score ranking data to obtain physiological and score feature data of the testee; the physiological and score feature data are imported into a high school student psychological assessment model to speculate a psychological assessment result of the subject, a psychological health score is generated, and the high school student psychological assessment model is obtained by training an artificial intelligence model through machine learning and deep learning and is updated through reinforcement learning. According to the method, the internal relation and rule among the physiological indexes, the academic performance and the psychological health state can be revealed, and the algorithm and accurate speculation of the psychological health state of the senior high school students are realized.

Description

Technical Field

[0001] The present invention relates to the field of mental health assessment, and in particular to a mental health assessment method and system based on physiological information monitoring of high school students. Background Art

[0002] Currently, high school students are in a critical period of puberty. Not only do they face heavy academic burdens and pressure from family, social life, and other aspects, but they are also constantly concerned about their academic performance. This makes mental health issues increasingly prominent among them and has become a social concern that cannot be ignored. Traditional mental health assessment methods, such as questionnaires and psychological tests, can help identify psychological problems to a certain extent, but their limitations are also quite obvious. Questionnaires often rely on students' self-reports and are greatly influenced by subjective factors. Students may hide or misreport their true feelings for various reasons, including concerns about their grades. Psychological tests usually take a long time to complete, and the results may be affected by a variety of factors, such as the testing environment, the student's mentality at the time, and anxiety about grades.

[0003] Furthermore, these methods often focus solely on the psychological level, overlooking the close connection between physiological factors, academic performance, and mental health. In reality, academic performance is not only an important indicator of student learning outcomes but also a crucial factor influencing their mental health. Long-term academic pressure and excessive anxiety about grades can lead to mental health issues such as anxiety and depression.

[0004] Given the above limitations, it is particularly urgent to explore a new method that can objectively, in real time, and comprehensively assess the mental health status of high school students. In recent years, with the advancement of science and technology and the deepening of biomedical research, people have gradually realized that there is a close and complex relationship between physiological information and mental health status. Studies have shown that physiological indicators such as heart rate variability, respiratory rate, and blood pressure are not only important signs of physical health, but also important windows for reflecting psychological state, including stress responses caused by academic performance. For example, a decrease in heart rate variability may be significantly associated with students' concerns about grades and increased anxiety; abnormal fluctuations in respiratory rate may indicate that the individual is in a state of tension caused by academic pressure; and a persistent increase in blood pressure may be related to long-term academic pressure and excessive focus on grades.

[0005] Therefore, by monitoring and analyzing high school students' physiological information in real time, combined with their academic performance data, daily behavior patterns, learning environment, and other multi-dimensional data, we are hoping to develop a new mental health assessment method. This method can comprehensively assess students' mental health by capturing physiological changes and fluctuations in academic performance in real time. It not only provides immediate feedback, helping students and parents promptly identify potential psychological issues caused by academic performance, but also provides scientific evidence for schools and mental health education institutions to guide them in developing more precise and effective intervention measures to alleviate students' academic stress and improve their mental health.

[0006] In summary, incorporating academic performance into a mental health assessment system and using physiological information to infer and evaluate the mental health of high school students is not only theoretically feasible but also has important practical significance and application value. It is expected to open up a new path for monitoring and intervening in the mental health of high school students, safeguarding the healthy development of adolescents, especially those under academic pressure. Summary of the Invention

[0007] The purpose of this invention is to propose a psychological assessment method and system for high school students based on artificial intelligence. This method can reveal the intrinsic connection and laws between physiological indicators, academic performance and mental health status, and realize the algorithmic and accurate inference of the mental health status of high school students.

[0008] The present invention is achieved through the following technical solutions:

[0009] A psychological assessment method for high school students based on artificial intelligence, comprising the following steps:

[0010] Step 1: The test subject wears a smart bracelet in daily study life, and the bracelet obtains the test subject's daily physiological data and the test subject's ranking in all exams from the school;

[0011] Step 2: performing data statistics, feature extraction, and analysis on the physiological data and performance ranking data obtained in step 1 to obtain the physiological and performance characteristic data of the test subject;

[0012] Step 3: Import the physiological and performance characteristic data obtained in step 2 into the high school student psychological assessment model to infer the subject's psychological assessment results and generate a mental health score. The high school student psychological assessment model is an artificial intelligence model obtained through machine learning and deep learning training, and is updated using reinforcement learning.

[0013] Furthermore, the physiological data include heart rate, blood pressure, respiratory rate, blood oxygen rate, sleep time, sleep quality and weight; the academic performance data include the overall school ranking and subject ranking of weekly tests, monthly tests, midterm and final tests.

[0014] Furthermore, the psychological assessment artificial intelligence model is obtained through machine learning and deep learning training, specifically including:

[0015] Step 2.1, obtain the physiological data and academic performance data of high school students in their daily life;

[0016] Step 2.2, using the holiday physiological data and grades as a benchmark, a machine learning clustering algorithm is performed to obtain the sample provider's mental health score;

[0017] Step 2.3: Use the physiological data and academic performance data obtained in step 2.1 and the mental health score data obtained in step 2.2 as a data set, and use k-fold cross-validation to divide the data set into k parts according to the set ratio, of which 1 part is the test set and the remaining k-1 parts are the training sets; use the training set data, the physiological data and academic performance data as the input values of the artificial intelligence model, and the mental health score as the output label of the artificial intelligence model to perform supervised training of deep learning, and then use cross-validation to test the trained artificial intelligence model through the test set test. If the test passes, the high school student psychological assessment model is obtained. If the test fails, execute step 2.4;

[0018] Step 2.4: Increase the number of training samples and repeat steps 2.1 to 2.3 until the test passes to obtain a high school student psychological assessment model.

[0019] Furthermore, after each academic year, the psychological assessment model for high school students is incrementally learned again to update the psychological assessment model for high school students.

[0020] Furthermore, the incremental learning adopts a regularization method to iteratively update the model.

[0021] Furthermore, the machine learning clustering algorithm in step 2.2 is a One-Class SVM clustering algorithm that combines dynamic threshold adjustment with time series features. The optimization formula is:

[0022]

[0023] Where: t is the dynamic threshold of the current time step t; ρ0 is the initial threshold, which is determined according to the mean of the holiday health benchmark data; β and γ are the global distance weight and time fluctuation weight respectively; ||x i -μ t || is the sample point x i To the data center μ t distance; is the time series volatility, T is the length of the historical time series, and N is the number of samples.

[0024] Furthermore, the kernel function of the One-Class SVM clustering algorithm is:

[0025] K(x i ,x j )=w1·K RBF (x i ,x j )+w2·K Linear (x i ,x j )+w3·K Poly (x i ,x j )

[0026]

[0027] K Linear (x i ,x j )=x i ·x j

[0028] K Poly (x i ,x j )=(x i ·x j +c) d

[0029]

[0030] Among them, K RBF (x i ,x j ) is the Gaussian kernel function, K Linear (x i ,x j ) is a linear kernel function, K Poly (x i ,x j ) is the polynomial kernel function, w k is the weight, Var k is the variance of the specific feature distribution, σ is the bandwidth parameter used to control the similarity between data points, c is the kernel constant, and d is the order of the polynomial kernel, which jointly handle the nonlinear distribution of data.

[0031] Furthermore, the artificial intelligence model is a deep neural network model, and the network structure includes an input layer, a feature interaction layer, a hidden layer, and an output layer. The feature interaction layer fuses multimodal features through a feature interaction module.

[0032] Furthermore, the multi-objective loss function for artificial intelligence model training is:

[0033] L=λ1·L Classification+λ2·L TemporalConsistency +λ3·L FeaturePenalty

[0034]

[0035] L TemporalConsistency =λ·||W f ||1

[0036] Among them, L Classification is the cross entropy loss based on mental health labels; L TemporalConsistency is the smoothing loss of time series data; L FeaturePenalty is the sparsity constraint missing of the input feature; λ1, λ2, λ3 are the weight parameters of each loss function, y i is the true label of the i-th sample, is the health score predicted by the model for the i-th sample.

[0037] Furthermore, the output of the high school student psychological assessment model is:

[0038] AnomalyScore=α·f physio (x)+(1-α)·f score (x)

[0039] f physio (x) = w physio ·φ(x physio )-ρ t

[0040] f score (x) = w score ·φ(x score )-ρ t

[0041] Among them, w physio is the physiological data weight, w score is the weight of the performance data, φ(x physio ) is the representation of physiological data mapped to high-dimensional space through kernel function, φ(x score ) is the representation of the performance data mapped to the high-dimensional space through the kernel function, ρ t is the dynamic threshold of the current time step t, α is the proportional coefficient, f physio (x) is the test value of physiological data, f score (x) is the test value of the performance data.

[0042] Furthermore, the deep neural network model uses a Sigmoid activation function, converts it into a percentage system, and outputs a mental health score.

[0043] An artificial intelligence-based psychological assessment system for high school students, including:

[0044] Data collection unit: the test subject wears a smart bracelet in daily study life, and the bracelet collects the test subject's daily physiological data and obtains the test subject's ranking in all exams from the school;

[0045] The data processing unit performs data statistics, feature extraction and analysis on the acquired physiological data and performance ranking data to obtain the physiological and performance characteristic data of the test subjects;

[0046] The mental health assessment unit imports physiological and performance characteristic data into the high school student psychological assessment model to infer the subject's psychological assessment results and generate a mental health score. The high school student psychological assessment model is an artificial intelligence model obtained through machine learning and deep learning training, and is updated using reinforcement learning.

[0047] Compared with the prior art, the method provided by the present invention has the following beneficial technical effects:

[0048] 1. This invention uses a psychological assessment instrument to monitor the heart rate, blood pressure, respiratory rate, sleep quality and other data of each student to analyze the mental health index. This solves the problem of the current reliance on a single indicator to evaluate the mental health index. It breaks the deficiency that a single dimension cannot comprehensively evaluate the mental health index. It fully considers the influence of key physiological factors, thereby improving the accuracy and completeness of mental health assessment.

[0049] 2. This invention uses physiological indicator data combined with artificial intelligence algorithms to analyze each student's mental health status and derive a mental health score, thereby achieving a personalized assessment of the student's mental state. This allows educators to subsequently formulate more appropriate intervention measures, thereby providing targeted intervention based on the student's specific problems. This can more effectively improve students' physical and mental states, thereby enhancing their quality of life and learning outcomes.

[0050] 3. This invention utilizes artificial intelligence algorithms and iteratively updates models in real time based on post-processing data. Compared to traditional statistical analysis methods, artificial intelligence-based algorithms can autonomously learn and adjust features based on data, thereby continuously improving model performance. This capability enables artificial intelligence-based algorithms to flexibly respond to complex and changing environments and continuously optimize. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 Flowchart of a psychological assessment method for high school students based on artificial intelligence. DETAILED DESCRIPTION

[0052] In this section, the method provided by the present invention will be further described in conjunction with specific examples. It should be noted that the examples are intended only to illustrate the present invention and are not intended to limit the scope of the present invention. Those skilled in the art may make various changes or modifications to the present invention that do not substantially change the present invention, which also fall within the scope defined by the claims appended hereto.

[0053] Example 1

[0054] Combine Figure 1 This embodiment provides a high school student psychological assessment method based on artificial intelligence, comprising the following steps:

[0055] Step 1: The test subject wears a smart bracelet in his daily study life, obtains the test subject's daily physiological data through the bracelet, and obtains the test subject's ranking in all exams from the school.

[0056] Step 2: Extract and analyze the physiological data and performance ranking data obtained in step 1 to obtain the physiological and performance characteristic data of the test subjects.

[0057] Step 3: The physiological and performance characteristic data obtained in step 2 are imported into the psychological assessment artificial intelligence model to infer the subject's psychological assessment results and generate a mental health score. The psychological assessment artificial intelligence model is obtained through machine learning and deep learning training.

[0058] The artificial intelligence training of the obtained high school student psychological assessment deep learning model includes the following steps:

[0059] Step 2.1: Obtain physiological data and academic performance data of high school students who provide artificial intelligence training samples in daily life; the physiological data includes physiological data such as heart rate, blood pressure, respiratory rate, blood oxygen rate, sleep time, sleep quality, and weight of the sample providers in daily life; the academic performance data includes the sample providers' comprehensive ranking in the school and ranking in each subject in weekly tests, monthly tests, midterm tests, and final tests.

[0060] Step 2.2: Using the test subjects' physiological data and grades during weekends, winter and summer vacations as a benchmark, a machine learning clustering algorithm is performed to obtain the sample provider's mental health score.

[0061] Step 2.3: Use k-fold cross-validation to divide the physiological data and academic performance data of the sample provider obtained in step 2.1 and the mental health score data of the sample provider obtained in step 2.2 into k parts according to a certain ratio, of which 1 part is the test set and the remaining k-1 parts are training sets; use the training set data, the physiological data and academic performance data of the sample provider as the model input values, and the mental health score of the sample provider as the model output label to perform supervised training of deep learning, and then use cross-validation to test the trained artificial intelligence model. After the test, a psychological assessment model for high school students is obtained.

[0062] Step 2.4: If the trained artificial intelligence model fails the test of the test set data, increase the number of training samples and repeat steps 2.1 to 2.3 until the test passes to obtain a high school student psychological assessment model.

[0063] Step 2.5: After each academic year, incremental learning is performed on the high school student psychological assessment model. After steps 2.1 to 2.3, an iterative high school student psychological assessment model is obtained to avoid sample drift.

[0064] The physiological characteristic data in step 2 and step 2.1 include: heart rate, blood pressure, respiratory rate, blood oxygen rate, sleep time, sleep quality, and body weight (BMI), which are counted on a daily basis.

[0065] The performance data in Steps 2 and 2.1 above include the sample provider's overall school ranking and subject ranking in weekly, monthly, midterm, and final exams, calculated on a weekly basis. Year-on-year and month-on-month comparisons of the overall school ranking and subject rankings are also taken.

[0066] The unsupervised machine learning algorithm used in step 2.2 is the One-Class SVM algorithm. This algorithm maps the data into a high-dimensional feature space, so that normal data points are surrounded by a hyperplane (decision plane). This hyperplane is designed to keep normal data points as close as possible to the hyperplane, while abnormal data points are further away from the hyperplane. Because the traditional One-Class SVM uses a fixed threshold ρ, it is suitable for statically distributed data. However, in actual mental health assessments, physiological data and performance data have obvious dynamic characteristics.

[0067] The One-Class SVM algorithm in step 2.2 adopts an improved method that combines dynamic threshold adjustment with time series features. The traditional One-Class SVM is prone to misjudgment or classification performance degradation when the data changes dynamically. Therefore, the improved algorithm introduces a dynamic threshold adjustment formula so that the decision threshold ρ t Automatically adjust with time step t:

[0068]

[0069] in:

[0070] ρ t : dynamic threshold of the current time step t;

[0071] ρ0: initial threshold, determined according to the mean of the holiday health benchmark data;

[0072] β and γ: global distance weight and time fluctuation weight respectively;

[0073] ||x i -μ t ||: sample point x i To the data center μ t The distance between the two groups captures the overall characteristics of the group's health distribution;

[0074] Time series volatility is used to measure short-term changes in mental health status. The final model decision function is:

[0075] f(x)=w·φ(x)-ρ t

[0076] The improved One-Class SVM algorithm in step 2.2 adopts adaptive kernel function design and improves the modeling ability of multimodal data through the dynamic fusion of Gaussian kernel, linear kernel and polynomial kernel: K(x i ,x j )=

[0077] w1·K RBF (x i ,x j )+w2·K Linear (x i ,x j )+w3·K Poly (x i ,x j )

[0078] in:

[0079] (1) Gaussian kernel function:

[0080]

[0081] (2) Linear kernel function:

[0082] K Linear (x i ,x j )=x i ·x j

[0083] (3) Polynomial kernel function:

[0084] K Poly (x i ,x j )=(x i ·x j +c) d

[0085] (4) Dynamic adjustment mechanism of core weights:

[0086]

[0087] Var k is the variance of the distribution of a specific feature.

[0088] The improved One-Class SVM model in step 2.2 was trained using holiday physiological data to ensure that the model accurately captured the normal distribution of holiday mental health status. During the testing phase, the physiological data and academic performance data were combined for detection and fusion.

[0089] (1) Detecting psychological anomalies using physiological data: Using the One-Class SVM model of physiological features to detect whether the sample deviates from the normal psychological state:

[0090] f physio (x) = w·φ(x physio )-ρ t

[0091] If f physio (x)≤0, it is judged as psychological abnormality.

[0092] (2) Detecting psychological anomalies using grade data: Using semester grade data, a separate grade clustering model is trained to detect abnormal grade fluctuations and determine whether the psychological state deviates from the normal state.

[0093] f score (x) = w·φ(x score )-ρ t

[0094] If f score (x)≤0, it is judged as psychological abnormality.

[0095] (3) Fusion test results: Fusion of the two test results:

[0096] AnomalyScore=α·f physio (x)+(1-α)·f score (x)

[0097] Where α is the weight ratio of physical and academic performance. If the AnomalyScore is less than 0, the mental health label is 0, otherwise it is 1.

[0098] The artificial intelligence model used for supervised training in step 2.3 is a deep neural network model. The network structure includes an input layer, a feature interaction layer, a hidden layer, and an output layer. Deep neural networks can learn the deep features of data. By leveraging the high-speed computing power of neural networks, they can quickly find optimal solutions to complex problems, thereby avoiding errors caused by human feature selection.

[0099] The deep neural network in step 2.3 is trained using the following optimized multi-objective loss function:

[0100] L=λ1·L Classification +λ2·L TemporalConsistency +λ3·L FeaturePenalty

[0101] in:

[0102] (1)L Classification : Cross-entropy loss based on mental health labels;

[0103]

[0104] in:

[0105] N: is the sample size;

[0106] y i : is the true label of the i-th sample, which takes the value of 0 (unhealthy) or 1 (healthy);

[0107] is the health score predicted by the model for the i-th sample;

[0108] (2)L TemporalConsistency : It is the smoothing loss of time series data, which maintains the stability of the prediction results of consecutive time steps;

[0109]

[0110] in:

[0111] T: is the length of the time series;

[0112] is the predicted health score at time step t;

[0113] is the predicted health score at time step t-1.

[0114] (3)L FeaturePenalty: The sparsity constraint of the input features is missing, which is used to reduce the interference of irrelevant features;

[0115] L TemporalConsistency =λ·||W f ||1

[0116] in:

[0117] W f : is the input feature weight matrix;

[0118] |W f ||1: W f The L1 norm of

[0119] λ: Regularization strength parameter for sparsity loss.

[0120] (4)λ1,λ2,λ3: are the weight parameters of each loss function, which are dynamically adjusted based on the validation set.

[0121]

[0122] in:

[0123] L k : is the current value of the k-th loss function;

[0124] λ k : is the corresponding dynamic weight;

[0125]

[0126] This adjustment method dynamically assigns weights based on the current value of each part of the loss function. Larger loss values are assigned smaller weights to prevent a certain part of the loss function from over-dominantly affecting the training and affecting the final result.

[0127] In the k-fold cross validation in step 2.3, k is set to 5, and a total of 5 cross validations are performed.

[0128] In step 2.5, incremental learning uses a regularization method to iteratively update the model. Every year, students' physiological and psychological data are collected again, and steps 1-3 are repeated. The LwF (Learning without Forgetting) algorithm is used to update the model without using old task data. By incorporating a distillation loss into the loss function, old knowledge is retained, eliminating the need to retrain the entire model and saving costs.

[0129] Furthermore, the key task of this assessment method is to evaluate the mental health of high school students based on physiological and academic performance data; the data provided by the sample providers are the same as those of the test subjects.

[0130] This embodiment provides an artificial intelligence-based psychological assessment system for high school students, including:

[0131] Data collection unit: the test subject wears a smart bracelet in daily study life, and the bracelet collects the test subject's daily physiological data and obtains the test subject's ranking in all exams from the school;

[0132] The data processing unit performs data statistics, feature extraction and analysis on the acquired physiological data and performance ranking data to obtain the physiological and performance characteristic data of the test subjects;

[0133] The mental health assessment unit imports physiological and performance characteristic data into the high school student psychological assessment model to infer the subject's psychological assessment results and generate a mental health score. The high school student psychological assessment model is an artificial intelligence model obtained through machine learning and deep learning training, and is updated using reinforcement learning.

[0134] This method integrates physiological monitoring, academic performance analysis, and artificial intelligence technology to conduct in-depth mining and comprehensive analysis of collected physiological information and academic performance data. By constructing a precise algorithmic model, this method can reveal the inherent connections and patterns between physiological indicators, academic performance, and mental health status, achieving algorithmic and accurate inference of the mental health status of high school students.

[0135] Compared to traditional psychological assessment methods, the AI-based assessment method proposed in this paper offers significant advantages. It not only overcomes the potential impact of test takers' potential pandering and disguised behaviors on assessment results, but also effectively avoids issues such as poorly structured and inadequately scientific assessments caused by the evaluator's personal inexperience. Furthermore, because the entire assessment process is automated and intelligent, eliminating the need for manual tracking and judgment, it significantly improves the efficiency and accuracy of assessments and reduces assessment costs. Schools and educators can promptly identify students' psychological problems, implement targeted intervention and counseling measures, effectively prevent the occurrence and deterioration of psychological problems, and provide strong support for students' healthy growth.

[0136] Example 2

[0137] Based on specific experimental data, this embodiment provides an AI-based psychological assessment method for high school students. This method uses physiological and academic data through an AI model to obtain a high school student's mental health score, allowing educators to subsequently develop more appropriate intervention measures. The detailed steps are as follows:

[0138] Step 1: Select a test school. Test subjects wear smart bracelets during their daily study life. The bracelets collect their daily physiological data, and the school obtains their test score rankings. A year's worth of data is selected as the sample set. Let d represent the number of days, d = 1, 2, 3, ... m. Let s represent the student numbers, s = 1, 2, 3, ... n. Assume the experimental subjects are 1,000 students.

[0139] Step 2: Extract and analyze the physiological data and performance ranking data obtained in step 1 to obtain the physiological and performance characteristic data of the test subjects.

[0140] Furthermore, the artificial intelligence training of the obtained high school student psychological assessment deep learning model includes the following steps:

[0141] Step 2.1: Obtain the physiological data and academic performance data of the high school student who provides the AI training sample in daily life. The physiological data features include the heart rate (HR), blood pressure (BP), respiratory rate (RR), blood oxygen rate (BOR), sleep time (ST), sleep quality (SQ), and body weight (BMI) of the sample provider in daily life, and calculate the data on a daily basis. The features F1 = [HR, BP, RR, SpO2, ST, SQ, BMI] are as follows:

[0142]

[0143] The performance data includes the sample provider's overall school ranking and subject ranking in weekly, monthly, midterm and final exams. Statistics are collected on a weekly basis. The overall school ranking (SWR) and subject ranking (SSR) are also taken. c Ranking year-on-year and quarter-on-quarter, c represents different course numbers, c = 1, 2, ... n. For example, the year-on-year value of the school ranking Month-on-month value Final score data characteristics and

[0144] F2=[SWR qoq ,SWR yoy ,SSR 1qoq ,SSR 1yoy ...SSR nqoq ,SSR nyoy ]. Features are as follows:

[0145]

[0146] Finally, we get the physiological feature F1 and the performance feature F2. Thus we get the data sets D1 and D2, D1 sd =[HR sd ,BP sd ,...,SQ sd , BMIsd ], Where s represents the student number, s = 1, 2, 3...n, d represents the day number d = 1, 2, 3...m, and the score data w can be converted based on d.

[0147] Step 2.2: Since the psychological assessment model is a supervised algorithm, it is necessary to obtain the psychological assessment of the test subject as the label Y sd , s represents the student number, d represents the date. First take Y sd The values are 0 and 1. 1 represents a healthy psychological assessment, and 0 represents an unhealthy psychological assessment.

[0148] To obtain label Y sd The unsupervised clustering algorithm, one-class SVM, is used. A single-class support vector machine uses only one type of data for modeling, aiming to identify abnormal data points that differ from the normal pattern. The algorithm maps the data into a high-dimensional feature space, so that normal data points are surrounded by a hyperplane (decision boundary). This hyperplane is designed to keep normal data points as close as possible to the hyperplane, while abnormal data points are far away from the hyperplane.

[0149] Because traditional one-class SVMs use a fixed threshold ρ, they are suitable for statically distributed data. However, in actual mental health assessments, physiological data and performance data exhibit significant dynamic changes. Therefore, a one-class SVM clustering algorithm combining dynamic threshold adjustment with time series features is employed. The optimization formula is:

[0150]

[0151] in:

[0152] ρ t : dynamic threshold of the current time step t;

[0153] ρ0: initial threshold, determined according to the mean of the holiday health benchmark data;

[0154] β and γ: global distance weight and time fluctuation weight respectively;

[0155] ||x i -μ t ||: sample point x i To the data center μ t distance;

[0156] Time series volatility, used to measure short-term changes in mental health status.

[0157] The dynamic One-Class SVM clustering algorithm is based on the characteristics of multimodal data and designs an adaptive kernel function for data fusion. The kernel function formula is:

[0158] K(x i ,x j )=w1·K RBF (x i ,x j )+w2·K Linear (x i ,x j )+w3·K Poly (x i ,x j )

[0159] in:

[0160] (1) Gaussian kernel function:

[0161]

[0162] (2) Linear kernel function:

[0163] K Linear (x i ,x j )=x i ·x j

[0164] (3) Polynomial kernel function:

[0165] K Poly (x i ,x j )=(x i ·x j +c) d

[0166] (4) Dynamic adjustment mechanism of core weights:

[0167]

[0168] Var k is the variance of the distribution of a specific feature.

[0169] Since the test subjects are generally in a good mental state during holidays, the test subjects' physiological and performance data during weekends, winter and summer vacations are used as the benchmark. The data set DS is taken from step 2.1, DS = [D1, D2, ... D h ], where h represents the holiday date. We then used holiday physiological data for training to ensure the model accurately captured the normal distribution of holiday health status. During the testing phase, we combined it with academic performance data for fusion verification, which included the following steps:

[0170] (1) Detecting psychological anomalies using physiological data: Using the One-Class SVM model of physiological features to detect whether the sample deviates from the normal psychological state:

[0171] f physio (x) = w physio ·φ(x physio )-ρ t

[0172] If f physio (x)≤0, it is judged as psychological abnormality.

[0173] (2) Detecting psychological anomalies using grade data: Using semester grade data, a separate grade clustering model is trained to detect abnormal grade fluctuations and determine whether the psychological state deviates from the normal state.

[0174] f score (x) = w score ·φ(x score )-ρ t

[0175] If f score (x)≤0, it is judged as psychological abnormality.

[0176] (3) Fusion test results: Fusion of the two test results:

[0177] Anomaly Score=α·f physio (x)+(1-α)·f score (x)

[0178] Cut off the holiday part of the data in step 2.1 and train the improved One-Class SVM to obtain

[0179] Anomaly Score=0.7·f physio (x)+0.3·f score (x)

[0180] The rest of the dataset D is evaluated by the discriminant model Anomaly Score to obtain the student psychological evaluation result Y sd The final data set D = [D1, D2, Y sd ].

[0181] Step 2.3: Label the physiological data and performance data of the sample provider obtained in step 2.1 and the mental health assessment result of the sample provider obtained in step 2.2 sd, according to the 10-fold cross-validation, the data set is divided into 5 parts according to a certain ratio, one of which is the test set and the remaining 4 are training sets; using the training set data, the physiological data and performance data of the sample provider are used as the model input values, and the mental health score of the sample provider is used as the model output label, and supervised training of deep learning is performed.

[0182] Furthermore, the artificial intelligence model for supervised training in step 2.3 is a deep neural network model. The network architecture includes a modality feature extraction layer, a feature fusion layer, and a health status prediction layer. The specific steps are as follows:

[0183] (1) Data preparation

[0184] Source of physiological data D1: Data obtained in step 2.1, including physiological data recorded daily for 365 days a year by 1,000 students (e.g., collected via smart bracelets). Specific characteristics include heart rate (HR), blood pressure (BP), respiratory rate (RR), blood oxygen rate (BOR), sleep time (ST), sleep quality (SQ), and body mass index (BMI). Data example:

[0185] Xscore=[[72,120,80,16,97,7.5,85,23.5],[78,125,85,17,96,6.8,80,24.0]…].

[0186] Source of score data D2: Data obtained in step 2.1, including comprehensive and individual subject rankings for weekly exams. Specific features include overall ranking, Chinese ranking, math ranking, English ranking, year-on-year and month-on-month change rates. Example data: Xscore = [[42, 2.5, 0, 30, 0, 0.67, 25, 2, 0, 55, -1.6, 0.4]…].

[0187] (2) Deep learning model design

[0188] Network input:

[0189] Physiological feature F1 is 8-dimensional, and performance feature F2 is 12-dimensional;

[0190] Network structure design:

[0191] 1. The input dimension d is 20.

[0192] 2. Feature interaction layer:

[0193] The feature interaction module is used to capture the synergistic relationship between physiological and performance features. The specific formula is:

[0194] F inter (x) = x + σ (W physio ·x physio+W score ·x score )

[0195] in:

[0196] W physio , W score : is the weight matrix of the interaction layer;

[0197] x physio , x score : Physiological and performance characteristics respectively.

[0198] 3. Hidden layer:

[0199] Multi-layer fully connected network, each layer contains 256, 128 and 64 neurons, using ReLU activation function:

[0200] h i =RELU(W i ·h i-1 +b i )

[0201] 4. Output layer:

[0202] Generate a mental health score using the Sigmoid activation function:

[0203] HealthScore=σ(W o ·h L +b0)·100

[0204] (3) Optimized multi-objective loss function

[0205] The total loss function is:

[0206] L=λ1·L Classification +λ2·L TemporalConsistency +λ3·L FeaturePenalty

[0207] in:

[0208] (1)L Classification : Cross-entropy loss based on mental health labels;

[0209]

[0210] (2)L TemporalConsistency : It is the smoothing loss of time series data, which maintains the stability of the prediction results of consecutive time steps;

[0211]

[0212] (3)L FeaturePenalty : The sparsity constraint of the input features is missing, which is used to reduce the interference of irrelevant features;

[0213] L TemporalConsistency =λ·||W f ||1

[0214] (4)λ1,λ2,λ3: are the weight parameters of each loss function, which are dynamically adjusted based on the validation set.

[0215]

[0216] (4) Model training and verification

[0217] Training process:

[0218] 1. Randomly split the data into a training set (80%) and a validation set (20%).

[0219] 2. Use the Adam optimizer with a learning rate of 0.001 and a batch size of 64.

[0220] 3. Iterate training until the validation set loss stops decreasing (early stopping mechanism).

[0221] Experimental results:

[0222] Classification accuracy: 98.8%.

[0223] F1 score: 97.9%.

[0224] Feature reconstruction error: 0.035.

[0225] Regularization loss: 0.0015.

[0226] Cross-validation is then used to test the trained psychological assessment model, ultimately resulting in a high school student psychological assessment model. By inputting the student's current physiological and academic performance data, a student's mental health score can be derived.

[0227] Step 2.4: If the trained artificial intelligence model fails the test of the test set data, increase the number of training samples and repeat steps 2.1 to 2.3 until the test passes to obtain a high school student psychological assessment model.

[0228] Step 2.5: Because students' physical conditions fluctuate annually and new students enter the classroom, to prevent model drift, which can lead to a large number of false positives in the psychological assessment model, the high school student psychological assessment model is incrementally retrained after each academic year, repeating steps 1-3 and employing regularization to iteratively update the model. The LwF (Learning without Forgetting) algorithm is used to update the model without using old task data. By incorporating a distillation loss into the loss function, old knowledge is retained, eliminating the need to retrain the entire model and saving costs.

[0229] Step 3: Import the physiological and performance characteristic data obtained in Step 2 into the psychological assessment artificial intelligence model to infer the psychological assessment results of the subjects and generate a mental health score. This can help understand the students' current mental state, provide a scientific basis for schools and mental health education institutions, and guide them to formulate more accurate and effective intervention measures to relieve students' learning pressure and improve their mental health level.

[0230] The method provided by the present invention can not only overcome the accuracy and unreliability problems in the prior art caused by the inability to avoid the test subjects' catering and disguise, but also effectively avoid the defects of poor structuring and unscientific and comprehensive evaluation caused by the tester's personal experience. At the same time, it is also a fully intelligent testing method that does not require the tester to manually track and judge the test subjects, thereby achieving accuracy and high reliability in the assessment of the mental health status of high school students.

[0231] The above contents are merely examples and explanations of the concept of the present invention. Those skilled in the art may make various modifications or additions to the described specific embodiments or replace them in a similar manner. As long as they do not deviate from the concept of the invention or exceed the scope defined by the present invention, they should all fall within the scope of protection of the present invention.

Claims

1. A psychological assessment method for high school students based on artificial intelligence, characterized by: Including steps: Step 1: The test subject wears a smart bracelet in daily study life, and the bracelet obtains the test subject's daily physiological data and the test subject's ranking in all exams from the school; Step 2: performing data statistics, feature extraction, and analysis on the physiological data and performance ranking data obtained in step 1 to obtain the physiological and performance characteristic data of the test subject; Step 3: Import the physiological and performance characteristic data obtained in step 2 into the high school student psychological assessment model to infer the subject's psychological assessment results and generate a mental health score. The high school student psychological assessment model is an artificial intelligence model obtained through machine learning and deep learning training, and is updated using reinforcement learning.

2. The artificial intelligence-based psychological assessment method for high school students according to claim 1, characterized in that: The physiological data include heart rate, blood pressure, respiratory rate, blood oxygen rate, sleep time, sleep quality and weight; the academic performance data include the overall school ranking and subject ranking of weekly tests, monthly tests, midterm and final tests.

3. The artificial intelligence-based psychological assessment method for high school students according to claim 1, characterized in that: The psychological assessment artificial intelligence model is obtained through machine learning and deep learning training, specifically including: Step 2.1, obtain the physiological data and academic performance data of high school students in their daily life; Step 2.2, using the holiday physiological data and grades as a benchmark, a machine learning clustering algorithm is performed to obtain the sample provider's mental health score; Step 2.3: Use the physiological data and academic performance data obtained in step 2.1 and the mental health score data obtained in step 2.2 as a data set, and use k-fold cross-validation to divide the data set into k parts according to the set ratio, of which 1 part is the test set and the remaining k-1 parts are the training sets; use the training set data, the physiological data and academic performance data as the input values of the artificial intelligence model, and the mental health score as the output label of the artificial intelligence model to perform supervised training of deep learning, and then use cross-validation to test the trained artificial intelligence model through the test set test. If the test passes, the high school student psychological assessment model is obtained. If the test fails, execute step 2.4; Step 2.4: Increase the number of training samples and repeat steps 2.1 to 2.3 until the test passes to obtain a high school student psychological assessment model.

4. The artificial intelligence-based psychological assessment method for high school students according to claim 3, characterized in that: After each academic year, the psychological assessment model for high school students is incrementally learned and updated.

5. The artificial intelligence-based psychological assessment method for high school students according to claim 4, characterized in that: The incremental learning adopts a regularization method to iteratively update the model.

6. The artificial intelligence-based psychological assessment method for high school students according to claim 3, characterized in that: The machine learning clustering algorithm in step 2.2 is a One-Class SVM clustering algorithm that combines dynamic threshold adjustment with time series features. The optimization formula is: Where: t is the dynamic threshold of the current time step t; ρ0 is the initial threshold, which is determined according to the mean of the holiday health benchmark data; β and γ are the global distance weight and time fluctuation weight respectively; ||x i -μ t || is the sample point x i To the data center μ t distance; is the time series volatility, T is the length of the historical time series, and N is the number of samples.

7. The artificial intelligence-based psychological assessment method for high school students according to claim 3, characterized in that: The kernel function of the One-Class SVM clustering algorithm is: K(x i ,x j )=w1·K RBF (x i ,x j )+w2·K Linear (x i ,x j )+w3·K Poly (x i ,x j ) K Linear (x i ,x j )=x i ·x j K Poly (x i ,x j )=(x i ·x j +c)d Among them, K RBF (x i ,x j ) is the Gaussian kernel function, K Linear (x i ,x j ) is a linear kernel function, K Poly (x i ,x j ) is the polynomial kernel function, w k is the weight, Var k is the variance of the specific feature distribution, σ is the bandwidth parameter used to control the similarity between data points, c is the kernel constant, and d is the order of the polynomial kernel, which jointly handle the nonlinear distribution of data.

8. The artificial intelligence-based psychological assessment method for high school students according to claim 3, characterized in that: The artificial intelligence model is a deep neural network model, and the network structure includes an input layer, a feature interaction layer, a hidden layer, and an output layer. The feature interaction layer fuses multimodal features through a feature interaction module.

9. The artificial intelligence-based psychological assessment method for high school students according to claim 3, characterized in that: The multi-objective loss function for training the artificial intelligence model is: L=λ1·L Classification +λ2·L TemporalConsistency +λ3·L FeaturePenalty L TemporalConsistency =λ·||W f ||1 Among them, L Classification is the cross entropy loss based on mental health labels; L TemporalConsistency is the smoothing loss of time series data; L FeaturePenalty is the sparsity constraint missing of the input feature; λ1, λ2, λ3 are the weight parameters of each loss function, y i is the true label of the i-th sample, is the health score predicted by the model for the i-th sample.

10. The artificial intelligence-based psychological assessment method for high school students according to claim 3, characterized in that: The output of the psychological assessment model for high school students is: Anomaly Score=α·f physio (x)+(1-α)·f score (x) f physio (x)=w physio ·φ(x physio )-r t f score (x)=w score ·φ(x score )-r t Among them, w physio is the physiological data weight, w score is the weight of the performance data, φ(x physio ) is the representation of physiological data mapped to high-dimensional space through kernel function, φ(x score ) is the representation of the performance data mapped to the high-dimensional space through the kernel function, ρ t is the dynamic threshold of the current time step t, α is the proportional coefficient, f physio (x) is the test value of physiological data, f score (x) is the test value of the performance data.

11. The artificial intelligence-based psychological assessment method for high school students according to claim 8, characterized in that: The deep neural network model uses a Sigmoid activation function, converts the score into a percentage, and outputs a mental health score.

12. A high school student psychological assessment system implementing the method according to any one of claims 1 to 11, characterized in that: include: Data collection unit: the test subject wears a smart bracelet in daily study life, and the bracelet collects the test subject's daily physiological data and obtains the test subject's ranking in all exams from the school; The data processing unit performs data statistics, feature extraction and analysis on the acquired physiological data and performance ranking data to obtain the physiological and performance characteristic data of the test subjects; The mental health assessment unit imports physiological and performance characteristic data into the high school student psychological assessment model to infer the subject's psychological assessment results and generate a mental health score. The high school student psychological assessment model is an artificial intelligence model obtained through machine learning and deep learning training, and is updated using reinforcement learning.

Images