Health big data-oriented delivery and education integrated intelligent prediction method and implementation method thereof

Abstract

The invention provides a health big data-oriented production and education integrated intelligent prediction and implementation method thereof, and the method comprises the steps: collecting the internship work data of internship students through the connection with a hospital information system, carrying out the correlation matching of the collected internship work data and a student health record data set, and constructing a production and education integrated health big data platform; according to the birth and education fusion health big data platform, individual dietary habit factors, work and rest rule factors and internship work intensity key features of the students are extracted, and the probability of gastrointestinal diseases of the students in the internship period is predicted by adopting a decision tree model; and obtaining an internship performance score and an internship unit evaluation in the internship period of the student, performing correlation analysis with the health condition data in the internship period of the student, and determining a correlation rule between the internship performance and the health level of the student.

Description

Technical Field

[0001] The present invention relates to the field of information technology, and in particular to an industry-education integration intelligent prediction for health big data and an implementation method thereof. Background Art

[0002] In the school environment, the incidence of gastrointestinal diseases in the student population is high. In order to study this phenomenon in depth, it is necessary to collect life data related to gastrointestinal diseases, such as school schedules and diet. However, it is difficult to fully reflect the problem by relying solely on school data. When students enter the hospital internship stage, they face changes in work content and schedule, which may have new impacts on gastrointestinal health. How to effectively integrate the work data during hospital internship with the previously collected school gastrointestinal disease data has become a technical problem. This requires the design of a complex data fusion model that takes into account the heterogeneity of data sources and ensures the temporal continuity and spatial consistency of data. At the same time, how to obtain sufficiently detailed hospital work data while protecting patient privacy is also a thorny issue. In addition, the prediction and intervention of gastrointestinal diseases based on the fused data involves multi-dimensional factor analysis and complex algorithm design. How to achieve efficient and accurate predictions under limited computing resources and give personalized intervention suggestions puts high demands on the system. Finally, dynamically adjusting the internship time and work content according to the prediction and intervention results requires a flexible scheduling system, which is in conflict with the existing work arrangements in the hospital. Summary of the invention

[0003] The present invention provides an industry-education integration intelligent prediction for health big data and its implementation method, which mainly includes:

[0004] Obtain the student's identification code, extract the student's gastrointestinal disease history, work and rest schedule, and eating habits information from the school database, clean and standardize the extracted data, and construct the student health record data set;

[0005] By connecting to the hospital information system, the internship work data of the interns are collected, and the collected internship work data are associated and matched with the student health record data set to build a health big data platform integrating industry and education;

[0006] Based on the big data platform of health integrated with industry and education, we extracted the key characteristics of students’ personal eating habits, work and rest schedules, and internship work intensity, and used a decision tree model to predict the probability of students developing gastrointestinal diseases during their internship.

[0007] According to the prediction results of the decision tree model, the key influencing factors leading to the high incidence of gastrointestinal diseases among students are determined. If students experience gastrointestinal discomfort during the internship, the health risk level of students is determined through the multi-layer perceptron neural network model based on their symptom characteristics, medical diagnosis results of the internship unit, medication treatment and key influencing factors.

[0008] For students whose health risk level exceeds the preset threshold, an optimized internship schedule and internship position arrangement plan will be generated through the industry-education integration health big data platform according to the adjustment range of internship duration and the classification rules of internship positions. After being reviewed and approved by the internship adjustment approval process, the optimized plan will be pushed to students and internship units;

[0009] Obtain students’ internship performance scores and internship unit evaluations during their internships, and conduct correlation analysis with students’ health status data during their internships to determine the correlation between internship performance and students’ health levels;

[0010] For students with abnormal internship performance scores and health risk levels, personalized internship improvement suggestions are generated based on correlation rules, and pushed to students and internship units through the industry-education integration health big data platform to continuously track changes in students' internship status and health conditions.

[0011] The technical solution provided by the embodiment of the present invention may have the following beneficial effects:

[0012] The present invention discloses an industry-education integration intelligent prediction for health big data and its implementation method, aiming to solve the problem of high incidence of gastrointestinal diseases during student internships by building an industry-education integration health big data platform. First, by obtaining the student identification code, the gastrointestinal disease history record, work and rest schedule and dietary habit information in the school database are extracted, and the information is cleaned and standardized. Then, the internship position type, internship duration and internship duty period work data of the interns are collected, and the data is associated with the student health file data set. Next, the key features of personal dietary habits, work and rest rules and internship work intensity are extracted, and the probability of gastrointestinal diseases during the internship is predicted using a decision tree model. On this basis, the health risk level of the students is judged by a multi-layer perceptron neural network model in combination with symptom characteristics, medical diagnosis and medication treatment of the internship unit. For students whose health risk level exceeds the preset threshold, their internship duration and position type are adjusted to generate an optimized internship arrangement plan. Finally, by associating the internship performance score with the health status data, personalized internship improvement suggestions are generated for students with poor internship performance and high health risks, and pushed to students and internship units through the platform. The present invention realizes comprehensive monitoring and management of the health status of students during internship, effectively reduces the risk of gastrointestinal diseases, and improves the health level and work efficiency of students during internship. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 This is a flow chart of an industry-education integration intelligent prediction for health big data and its implementation method of the present invention.

[0014] Figure 2 It is a schematic diagram of an industry-education integration intelligent prediction for health big data and its implementation method according to the present invention.

[0015] Figure 3 It is another schematic diagram of the industry-education integration intelligent prediction for health big data and its implementation method according to the present invention. DETAILED DESCRIPTION

[0016] The technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0017] like Figure 1 -3. In this embodiment, an industry-education integration intelligent prediction for health big data and its implementation method may specifically include:

[0018] S101. Obtain the student's identification code, extract the student's gastrointestinal disease history, work and rest schedule, and eating habits information from the school database, clean and standardize the extracted data, and construct a student health record data set.

[0019] Obtain the student identification code, and extract the student health-related data from the campus database according to the student identification code; use the data cleaning method to process the student health-related data to obtain the cleaned student health data; construct a student health record data set based on the cleaned student health data; encrypt the student health record data set through the AES-256 encryption algorithm to obtain an encrypted student health record data set; store the encrypted student health record data set in a secure data storage system, and determine whether the user identity is legal according to the preset access permission control rules. If the user identity is legal, the user is allowed to access the decrypted data.

[0020] Specifically, after obtaining the student identification code, the student's gastrointestinal disease history, work and rest schedule, and dietary habits information are extracted from the school database using SQL query statements. The extracted data is cleaned, duplicates and invalid items are removed, and the data format is unified. The mean filling method is used to handle missing values, and the quartile method is used to identify and delete abnormal data points. The cleaned data is Z-score standardized to convert data of different dimensions to the same scale to obtain standardized student health data. Based on the standardized student health data, a student health record data set is constructed, and the main fields are defined to include student basic information, gastrointestinal disease history, work and rest schedule, and dietary habits. The corresponding data types such as integer, floating point, and string are set. A data consistency check is performed to ensure that the data in each field conforms to the predefined format and value range. For the constructed student health record data set, the AES-256 encryption algorithm is used for encryption, and the encryption key is securely stored and distributed through the key management server. The integrity of the encrypted data set is verified, and the hash value of the data set is calculated and compared with the original data set to ensure that the data has not been tampered with during the encryption process. The encrypted student health record data set is stored in a secure data storage system, and access control is set to allow only authorized users to access the decrypted data after identity verification. When extracting student information from the school database, SQL query statements are used to obtain relevant data for specific students. During the data cleaning process, the regular expression "^\d{4}-

[0021] \d{2}-\d{2}￥"Verify the date format and mark dates that do not conform to the format as missing values. For missing values, the mean filling method is used, such as filling the missing sleep time with the average value of all records of the student, 7.5 hours. The quartile method is used to handle outliers. The Q1 of the sleep time is calculated as 6 hours and the Q3 is 9 hours. Data below Q1-1.5IQR (3 hours) or above Q3+1.5IQR (12 hours) are deleted. In the Z-score standardization process, the formula Z=(X-μ) / σ is applied to the sleep time data, where μ is the average sleep time of 7.5 hours and σ is the standard deviation of 1.2 hours. When constructing the health record data set, define the word The segment "student_id" is of integer type, "disease_history" is of string type, and "sleep_time" is of floating point type. Data consistency check ensures that the value of the "sleep_time" field is between 0 and 24. When using the AES-256 encryption algorithm, select CBC mode, the initialization vector is a 16-byte random number, and the key length is 256 bits. After encryption, calculate the SHA-256 hash value of the data set and compare it with the hash value of the original data set to ensure data integrity. Finally, the encrypted data is stored in a secure storage system with an access control list (ACL), and read permissions are set to be open only to specific user groups.

[0022] S102. Collect the internship work data of interns by connecting to the hospital information system, associate and match the collected internship work data with the student health record data set, and build a health big data platform integrating industry and education.

[0023] When obtaining the work data of internship students, the HL7 protocol request is sent according to the hospital information system application program interface to obtain the original data including the internship position type, internship duration and internship duty period; the XML parser is used to perform structured processing on the original data to generate a standardized internship work data set. The health record data is extracted from the student health record database, and a consistency check is performed on the health record data to determine whether the data format and content meet the predefined standards; if it meets the predefined standards, it is determined to be valid health record data. The standardized internship work data set and the valid health record data are matched through the SQL INNER JOIN operation, and the association relationship between the two data sets is established according to the student unique identifier to obtain the associated comprehensive data set. Apache Griffin is used to perform quality inspection on the associated comprehensive data set to determine whether the data is abnormal or inconsistent; if there is anomalies or inconsistencies, the abnormal or inconsistent data is marked. Based on the associated comprehensive data set, an industry-education integration health big data platform is constructed, and the comprehensive data set is stored in the Hadoop distributed file system; Apache Hive is used to perform data warehouse management and query operations on the stored comprehensive data set to obtain a multidimensional data cube.

[0024] Specifically, the HL7 protocol is used to obtain the internship work data of the internship students' internship position type, internship duration and internship duty period through the application program interface of the hospital information system. The XML parser is used to perform structured processing on the obtained raw data to generate a standardized internship work data set. The student health record data is extracted from the student health record database, and the data consistency check is performed on the extracted health record data to ensure that the data format and content meet the predefined standards. The SQL INNER JOIN operation is used to match the internship work data set and the student health record data set. The association relationship between the two data sets is established through the student unique identifier to generate a comprehensive data set after association. Apache Griffin is used for data quality detection to identify and mark potential data anomalies or inconsistencies. Based on the comprehensive data set after association, an industry-education integration health big data platform is constructed, and the data is stored in the Hadoop distributed file system. Apache Hive is used for data warehouse management and query. A multidimensional data cube is created, including internship positions, internship duration, and health status. Apache Kylin is used for online analysis and processing to support multidimensional data analysis and visualization. Data indexes and partitions are established to optimize query performance and achieve fast data retrieval and complex analysis. In the hospital information system, the ADT^A08 message type of the HL7 protocol is used to obtain the internship student information, and the internship position type (such as internal medicine, surgery), internship duration (unit: hours) and internship duty period (such as day shift, night shift) are extracted through the XML parser. The extracted data is structured to generate a standardized internship work data set, which contains the fields: student ID, internship position, cumulative internship duration, and duty period distribution. When extracting data from the student health record database, SQL queries are executed to obtain student health status information. Data consistency check algorithms are used, such as checking whether the date format conforms to the "YYYY-MM-DD" standard and whether the numerical field is within a reasonable range (such as BMI value between 15 and 40). The SQL INNER JOIN operation is used to associate the internship work data and health record data. Apache Griffin is used for data quality detection, and rules such as "internship duration>0AND internship duration<1000" are set to identify abnormal data. The associated data set is stored in Hadoop HDFS, and Apache Hive is used to create an external table. Apache Kylin is used to build a multidimensional data cube. The selected dimensions include internship positions, internship duration ranges (0-100 hours, 101-200 hours), health status (good, average, need attention), and the measurement values ​​include the number of students and the average internship duration.

[0025] S103. Based on the industry-education integration health big data platform, extract the key features of students' personal eating habits, work and rest schedules, and internship work intensity, and use a decision tree model to predict the probability of students developing gastrointestinal diseases during their internship.

[0026] Obtain the original data of students' personal eating habits, work and rest regularity factors, and internship work intensity in the industry-education integration health big data platform; clean and standardize the original data, remove outliers and missing values, and obtain the preprocessed feature data set. Calculate the internship work intensity index according to working hours and work types; quantify the internship work intensity index into a standard score of 0-100. Use feature engineering methods to process feature data sets; calculate dietary regularity indicators according to dietary time rules; calculate sleep quality index according to sleep duration and sleep quality; calculate work intensity coefficient according to internship work intensity. Use CART decision tree algorithm to train the model of training set data; select the optimal tree structure and pruning parameters through grid search combined with cross-validation to obtain a trained decision tree model. Use the decision tree model to predict the test set data; calculate the probability value of each student developing gastrointestinal diseases during the internship; and judge whether the student is susceptible to gastrointestinal diseases based on the probability value.

[0027] Specifically, the original data of students' personal dietary habits, work and rest regularity factors, and internship work intensity were extracted from the industry-education integration health big data platform. The extracted data were cleaned and standardized, and outliers and missing values ​​were removed to obtain the preprocessed feature data set. The internship work intensity index was calculated, and the working hours and work type (physical / mental) factors were comprehensively considered and quantified into a standard score of 0-100. The SMOTE algorithm was used to balance the data and generate a new balanced data set. Feature engineering was performed on the balanced feature data set to calculate the dietary regularity index, sleep quality index, and work intensity coefficient. The Pearson correlation coefficient was used to calculate the correlation between features, and a threshold of 0.8 was set to remove highly correlated features. Then, the principal component analysis method was used for dimensionality reduction to construct the input feature vector for the decision tree model. The constructed feature vector and the corresponding student gastrointestinal disease history data were divided into a training set and a test set, and the CART decision tree algorithm was used to train the model on the training set data. The optimal tree structure and pruning parameters were selected by grid search combined with 5-fold cross validation. The trained decision tree model was used to predict the test set data and calculate the probability of each student developing gastrointestinal diseases during the internship. The accuracy, precision, recall and F1 score of the model were calculated, and the ROC curve was drawn to evaluate the model performance. The trained decision tree was visualized, the feature importance was analyzed, and the interpretability analysis of the model prediction results was provided. According to the feature importance sorting, the key factors affecting the occurrence of gastrointestinal diseases in students were identified. When extracting student data from the industry-education integration health big data platform, SQL query statements were used to obtain the original data. During the data cleaning process, the dietary regularity index was standardized to the range of 0-1, and the sleep duration was normalized to a deviation value based on 8 hours per day. The internship work intensity index calculation formula is: intensity index = (daily working hours / 8) * 0.6 + (work type coefficient) * 0.4, where the physical work coefficient is 1 and the mental work coefficient is 0.8. The SMOTE algorithm was used to generate a new balanced data set, and the minority class samples (sick students) were synthesized to be equivalent to the majority class samples. In feature engineering, the dietary regularity index was calculated as the inverse of the standard deviation of meal times over 7 consecutive days, and the sleep quality index was the weighted sum of the average sleep duration and the number of sleep interruptions. The Pearson correlation coefficient was applied to screen features, and feature pairs with correlation coefficients greater than 0.8 were removed. Principal component analysis retained principal components with a cumulative explained variance rate of 95%. CART decision tree training used the Gini index as the splitting criterion, and optimized the maximum depth (3-10) and the minimum number of leaf node samples (5-20) through grid search. In the 5-fold cross-validation, the training set of each fold contained 80% of the data and the validation set 20%. In the model evaluation stage, the AUC value of the ROC curve reached 0.85, indicating that the model has good discrimination ability. The decision tree visualization shows that the internship work intensity index is at the top level of the tree and is the most important predictor, followed by the sleep quality index and dietary regularity index.The results of feature importance analysis showed that internship work intensity accounted for 40%, sleep quality accounted for 30%, dietary regularity accounted for 20%, and other features accounted for 10%.

[0028] S104. Based on the prediction results of the decision tree model, determine the key influencing factors that lead to the high incidence of gastrointestinal diseases among students. If students experience gastrointestinal discomfort during the internship, the health risk level of the students is determined through a multi-layer perceptron neural network model based on their symptom characteristics, medical diagnosis results of the internship unit, medication treatment and key influencing factors.

[0029] Obtain the key influencing factors that lead to the high incidence of gastrointestinal diseases among students, quantify the key influencing factors, and obtain the key influencing factor feature vectors. According to the industry-education integration health big data platform, obtain the health-related data of students during their internship, including gastrointestinal discomfort symptom data, medical diagnosis results, and medication details. Standardize and encode the health-related data to generate symptom feature vectors, diagnosis result vectors, and treatment status vectors. Concatenate the key influencing factor feature vectors, symptom feature vectors, diagnosis result vectors, and treatment status vectors to construct a complete input feature matrix. Use a multilayer perceptron neural network model to train the input feature matrix and health risk level labels to obtain a training model for judging the student's health risk level. If a student is detected to have gastrointestinal discomfort symptoms, obtain their symptoms, diagnosis, and treatment data, combine them with key influencing factors, and input them into the training model to judge the student's health risk level.

[0030] Specifically, according to the prediction results of the decision tree model, the key influencing factors that lead to the high incidence of gastrointestinal diseases among students are extracted, and these factors are quantified to convert qualitative descriptions into numerical features. For example, the degree of irregular diet is converted into a score of 0-10, where 0 means very regular and 10 means extremely irregular; the sleep quality is converted into a score of 0-5, where 0 means extremely poor sleep quality and 5 means excellent sleep quality. Construct a feature vector of key influencing factors, which contains all quantified factor values. Obtain the data on gastrointestinal discomfort symptoms of students during the internship, the medical diagnosis results of the internship unit, and the medication treatment status from the industry-education integration health big data platform. Standardize and encode these data, use the Z-score method to standardize the numerical features, and use one-hot encoding for the categorical features such as symptom type and diagnosis results. Generate symptom feature vectors, diagnosis result vectors, and treatment status vectors. Concatenate the key influencing factor feature vectors, symptom feature vectors, diagnosis result vectors, and treatment status vectors to construct a complete input feature matrix. At the same time, the corresponding health risk level is annotated as the output label according to the historical data, and the risk level is divided into low risk (0-0.3), medium risk (0.3-0.7), and high risk (0.7-1). The input data is verified and anomaly processed, and outliers are removed or corrected. The constructed feature matrix and labels are trained using a multilayer perceptron neural network model. The network structure includes an input layer (the number of nodes is equal to the number of features), two hidden layers (64 nodes each), and an output layer (3 nodes, corresponding to three risk levels). The ReLU activation function, Adam optimizer, and cross entropy loss function are used. The network parameters are optimized by the back propagation algorithm to obtain a training model that can judge the health risk level of students. When students are detected to have gastrointestinal discomfort symptoms, their symptoms, diagnosis and treatment data are collected, combined with the previous key influencing factors, input into the trained model, and the health risk level is output. A feedback mechanism is established to regularly update the model according to the actual situation to improve the prediction accuracy. When extracting key influencing factors from the decision tree model, it is found that irregular diet, sleep quality, and work intensity are the top three factors. The degree of irregular eating is quantified into 0-10 points, for example, not eating on time for three consecutive days is recorded as 8 points; sleep quality is converted into 0-5 points, and the average sleep time per night is less than 6 hours is recorded as 1 point; work intensity is calculated based on daily working hours, 8 hours is recorded as 5 points, and 12 hours is recorded as 10 points. The data related to students' gastrointestinal discomfort is obtained through database query statements. The Z-score method is used to standardize numerical features, such as standardizing "abdominal pain duration": Z = (X-μ) / σ, where μ is the average duration of 4 hours and σ is the standard deviation of 1.5 hours. One-hot encoding is performed on categorical features, such as encoding "nausea" in the "symptom type" feature as [1, 0, 0], "abdominal pain"

[0031] Encoded as [0, 1, 0]. All feature vectors are concatenated to obtain the input matrix X, with the dimension of [number of samples, number of features]. The health risk level Y is encoded as [low risk, medium risk, high risk], such as [1, 0, 0] represents low risk. Construct a multi-layer perceptron, with the number of nodes in the input layer being the number of features (such as 100), 64 nodes in each of the two hidden layers, and 3 nodes in the output layer. Use the ReLU activation function: f(x) = max(0, x), the initial learning rate of the Adam optimizer is set to 0.001, and the cross entropy loss function calculates the difference between the predicted value and the true value. After 1000 rounds of training, the model achieved an accuracy of 85% on the test set. When new students have symptoms, such as abdominal pain lasting for 2 hours and working 12 hours / day for 5 consecutive days, the model outputs a risk level of [0.1, 0.3, 0.6], which is judged as high risk. Update the model with newly collected data every month to continuously improve the prediction accuracy.

[0032] S105. For students whose health risk level exceeds the preset threshold, an optimized internship schedule and internship position arrangement plan will be generated through the industry-education integration health big data platform based on the adjustment range of the internship duration and the internship position type classification rules. After being reviewed and approved by the internship adjustment approval process, the optimized plan will be pushed to students and internship units.

[0033] Obtain information about students whose health risk level exceeds the preset threshold in the industry-education integration health big data platform, and generate a list of students to be optimized. According to the internship duration adjustment range rule, calculate the new internship duration for each student in the list of students to be optimized. Use the internship position type classification rule to match new suitable positions for students in the list of students to be optimized. Generate a preliminary optimization plan based on the new internship duration and new position matching results. Input the preliminary optimization plan into the genetic algorithm, and set the internship unit position capacity and student course schedule as constraints. Evaluate the advantages and disadvantages of the plan through the fitness function, and obtain the optimized internship schedule and internship position arrangement plan that meets the constraints. Submit the optimized internship schedule and internship position arrangement plan to the internship adjustment approval process. Automatically review through the rule engine, which contains the review rule library for internship duration compliance and position matching. If the review is passed, send the optimization plan to the relevant students and internship units; if the review is not passed, re-optimize.

[0034] Specifically, the information of students whose health risk level exceeds the preset threshold of 0.7 is extracted from the industry-education integration health big data platform, including student number, current internship duration, internship position type and health risk level value, to generate a list of students to be optimized. According to the internship duration adjustment range rule, the new internship duration is calculated for each student in the list of students to be optimized. The internship duration of high-risk students (risk level 0.7-1) is reduced by 20%, and the internship duration of medium-risk students (risk level 0.3-0.7) is reduced by 10%. At the same time, according to the internship position type classification rule, based on the work intensity classification, the original high-intensity positions are adjusted to medium or low-intensity positions, and new suitable positions are matched for students to generate a preliminary optimization plan. The preliminary optimization plan is input into the genetic algorithm, considering the position capacity limit of the internship unit and the course arrangement of the students as constraints, setting the fitness function to evaluate the advantages and disadvantages of the plan, and generating the optimized internship schedule and internship position arrangement plan that meet the constraints. The generated optimization plan is submitted to the internship adjustment approval process, automatically reviewed by the rule engine, and the review rule library is set, including the compliance of the internship duration and the position matching degree. If the review is passed, the optimization plan will be sent to the relevant students and internship units through the message push service. If the review is not passed, the rule engine will give the specific reasons and return to the second step for re-optimization. Collect feedback from students and internship units on the optimization plan and store it in the database for subsequent adjustment and improvement of the optimization algorithm. For student A with a risk level of 0.85, the original internship duration is 400 hours, and the position is a high-intensity position. The adjustment rule is applied: new internship duration = 400 * (1-20%) = 320 hours, and the position is adjusted to medium intensity. Genetic algorithm is used to generate the optimization plan, the chromosome encoding is [student ID, new internship duration, new position type], the population size is set to 100, and the number of iterations is 1000. The fitness function takes into account the capacity limit of the internship unit (such as a maximum of 10 people per position) and the conflict of student courses (such as a maximum of 8 hours of internship per day). The crossover probability is set to 0.8 and the mutation probability is 0.1. After iteration, the optimal solution is obtained: student A is assigned to position B, and the internship time is Monday to Thursday, 6 hours a day. The optimization plan is input into the rule engine for review. The rules include: the internship duration is not less than 300 hours as required by the school, and the job type matches the student's major by more than 80%. If Student A's plan passes the review, the system will push it to the student and the internship unit via SMS and email. If it fails, such as the internship duration is 280 hours, the system will automatically adjust it to 300 hours and resubmit it for review. When collecting feedback, an online questionnaire is used, with a score of 1-5 points. The feedback results are stored in the MongoDB database for subsequent optimization algorithm improvements, such as adjusting the mutation probability of the genetic algorithm or adding new constraints.

[0035] Adjust internship duration based on health risk score.

[0036] Obtain the student's health risk score and benchmark internship duration. The health risk score is generated by the health risk assessment module, and the benchmark internship duration is extracted from the internship management database; judge the student's health risk level according to the preset warning threshold and safety threshold. If the health risk score is higher than the warning threshold, calculate the new internship duration according to the shortening ratio corresponding to the risk level; if the health risk score is lower than the safety threshold, calculate the new internship duration according to the corresponding extension ratio; compare the new internship duration with the shortest and longest time limits in the internship management regulations. If it exceeds the limit range, adjust the new internship duration to the nearest boundary value to obtain the adjusted internship duration; calculate the utility value corresponding to each duration by the following formula;

[0037] ∪=w1·(1-R)+w2·L+w3·S

[0038] U represents utility, w1, w2 and w3 are weights, R represents risk score, L represents learning progress, and S represents satisfaction with the internship unit. The maximum value is selected from the utility value, and the corresponding internship duration is output as the optimal internship duration recommendation.

[0039] Specifically, the health risk score of the student is obtained from the health risk assessment module, and the benchmark internship duration of the student is extracted from the internship management database. These two values ​​are input into the internship duration adjustment algorithm. According to the preset warning threshold of 0.7 and the safety threshold of 0.3, the health risk level of the student is judged. If the risk score is higher than the warning threshold of 0.7, the new internship duration is calculated according to the shortening ratio corresponding to the risk level. For every 0.1 that the risk score exceeds the warning threshold, the internship duration is shortened by 5%; if the risk score is lower than the safety threshold of 0.3, the new internship duration is calculated according to the corresponding extension ratio. For every 0.1 that the risk score is lower than the safety threshold, the internship duration is extended by 3%, but not more than 20% of the original duration. The calculated new internship duration is compared with the shortest and longest time limits in the internship management regulations. If it exceeds the limit range, the duration is adjusted to the nearest boundary value to obtain the adjusted internship duration. Under the constraints of meeting the shortest and longest time limits, the duration is gradually increased from the minimum duration through recursive or iterative methods, the utility value corresponding to each duration is calculated, and the optimal solution is recorded. The internship duration when the utility is the maximum is finally output as the optimal internship duration recommendation. The health risk score of student A is 0.75 obtained from the health risk assessment module, and the benchmark internship duration of the student is 400 hours extracted from the internship management database. According to the preset warning threshold of 0.7 and the safety threshold of 0.3, the health risk level of student A is judged to be high risk. Since the risk score exceeds the warning threshold of 0.05, the internship duration needs to be shortened by 2.5%, that is, the new internship duration is 390 hours. The minimum time limit stipulated by the internship management is 300 hours and the maximum time limit is 600 hours. 390 hours is within this range and does not need to be adjusted. Utility = 0.5 * (1-0.75) + 0.3 * 0.8 + 0.2 * 0.9, where 0.8 is the learning progress and 0.9 is the satisfaction of the internship unit. Starting from 300 hours, increase 10 hours each time to calculate the utility value. When the internship duration is 350 hours, the maximum utility value is 0.56. After traversing all possible durations, 350 hours is determined to be the optimal recommended internship duration. The system stores this recommended value in the database and sends it to relevant personnel through the message push interface.

[0040] Extract job feature vectors from the internship job database, calculate similarity with students’ personal feature vectors, and select the internship positions with the highest matching degree.

[0041] Obtain the feature vectors of the positions in the internship position database; generate a standardized position feature matrix based on professional requirements and work intensity factors. Receive data carrying the student's health risk level, professional background, and personal characteristics; preprocess and digitize the data to obtain the student's personal feature vector. Use the weighted cosine similarity algorithm to calculate the similarity score between the student's personal feature vector and the position feature vector; sort the similarity scores in descending order; if the similarity score is greater than the preset threshold, the corresponding position is used as the matching result. Determine whether the work intensity of the position exceeds the tolerance corresponding to the student's health risk level; if the work intensity does not exceed the tolerance, save the student's position information to the internship recommendation list.

[0042] Specifically, the feature vectors of all positions, including professional requirements and work intensity factors, were extracted from the internship position database, and these features were numerically processed. The professional requirements were converted into one-hot codes, and the work intensity factors were represented by values ​​of 1-5 to generate a standardized position feature matrix. The health risk level, professional background and personal feature data of the students were obtained, and these data were preprocessed and numerically converted. The health risk level was quantified into low, medium and high risks, represented by 1, 2 and 3 respectively. The professional background was processed in the same one-hot coding method as the position requirements to construct the student's personal feature vector. The weighted cosine similarity algorithm was used to calculate the similarity score between the student's personal feature vector and the feature vector of each position. The weight factor was introduced, giving the health risk level a weight of 0.4, the professional matching weight of 0.3, the work intensity matching weight of 0.2, and the weight of other personal features of 0.1 to generate a similarity score list. According to the similarity score list, the positions were sorted in descending order, and the positions with a similarity score greater than 0.7 or the first 5 positions were fixed as the matching results. Add a screening step to exclude positions whose work intensity exceeds the health tolerance of students, and save the screened position information to the student's internship recommendation list. Establish a feedback mechanism to dynamically adjust the weight factor based on the student's choice and internship performance to improve the accuracy of future matching. Extract position characteristics from the internship position database, such as the professional requirements of the software development position are converted to [1, 0, 0, 0, 0] (representing computer science majors), and the work intensity factor is set to 4 (levels 1-5). Student A's health risk level is 2 (medium risk), his professional background is [1, 0, 0, 0, 0] (computer science), and his personal characteristics include a programming skill score of 8 (full score 10). Use the weighted cosine similarity algorithm to calculate the matching degree, setting the weights as: health risk 0.4, professional match 0.3, work intensity 0.2, and personal characteristics 0.1. For the software development position, the calculated similarity score is 0.85. After calculating the similarity of all positions, sort them in descending order and select positions with a score greater than 0.7. Assume that there are 3 positions with scores exceeding 0.7: software development (0.85), data analysis (0.78), and IT support (0.72). During screening, it was found that the work intensity of the IT support position was 5, which exceeded the student's health tolerance and was eliminated. Finally, the software development and data analysis positions were recommended to student A. The system records that student A chose the software development position, performed well during the internship, and had a stable health. Based on this feedback, the system fine-tuned the weight factor, slightly increasing the professional matching weight to 0.35 and reducing the work intensity weight to 0.15 to optimize future matching accuracy.

[0043] S106. Obtain the internship performance score and internship unit evaluation of the students during the internship, and conduct correlation analysis with the health status data of the students during the internship to determine the correlation pattern between the internship performance and the health level of the students.

[0044] Obtain the student internship performance score data in the internship management system and the student health status data in the health monitoring system; integrate the internship performance score data with the health status data according to the student ID and timestamp to obtain a complete student internship-health data set. Perform natural language processing on the internship unit evaluation text data in the student internship-health data set; extract keywords and sentiment tendencies in the text data, and convert the keywords and sentiment tendencies into numerical form. Use the Z-score method to detect outliers in the student internship-health data set and remove outliers; standardize the data set to obtain a standardized data set. Perform principal component analysis on the standardized data set; extract the main features in the data set, reduce the dimension of the data set, and obtain a feature matrix after dimensionality reduction. Use the Pearson correlation coefficient method to calculate the correlation coefficient between each indicator in the feature matrix; construct a multivariate linear regression model based on the correlation coefficient; and quantify the influence of each health indicator on internship performance through the multivariate linear regression model.

[0045] Specifically, the internship performance scores and internship unit evaluation data of students during the internship were obtained from the internship management system, and the health status data of students during the internship was extracted from the health monitoring system. These data were integrated according to student ID and timestamp to form a complete student internship-health data set. Natural language processing technology was used to extract keywords and sentiment tendencies from the text data of the internship unit evaluation and converted into numerical form. The integrated data set was preprocessed, including missing value processing, outlier detection and removal, and data standardization. The Z-score method was used to detect outliers, and data of different dimensions were converted into a unified scale to generate a standardized data set. The principal component analysis method was used to extract features from the standardized data set to reduce the data dimension while retaining key information to obtain the feature matrix after dimensionality reduction. The extracted features were preliminarily analyzed, and the univariate correlation between each feature and the internship performance was calculated. The features with correlation higher than the preset threshold were selected for subsequent analysis. The Pearson correlation coefficient method was used to perform correlation analysis on the internship performance indicators and health status indicators in the feature matrix, and the correlation coefficients between the indicators were calculated. Considering the time lag effect, the correlations in different time windows (such as 1 week later and 2 weeks later) were analyzed. A heat map was drawn to identify the significant correlation between internship performance and students' health level. Based on the results of correlation analysis, a prediction model was constructed using a multivariate linear regression algorithm to quantify the degree of influence of each health indicator on internship performance, so as to more accurately determine the correlation pattern. The internship performance score of student A was extracted from the internship management system, which was 85 points, and the evaluation text of the internship unit was "work seriously and have a sense of responsibility". The keywords "serious" and "responsibility" were extracted using the Jieba word segmentation and TF-IDF algorithm, and the sentiment analysis score was 0.8 (positive). The health monitoring system recorded that the average sleep time of student A was 7 hours and the average exercise time was 30 minutes / day. The data was standardized by Z-score, and the sleep time was converted to 0.5 and the exercise time was converted to -0.2. The principal component analysis retained 95% of the information and reduced the 10 original features to 4 principal components. The univariate correlation was calculated, and the correlation coefficient between sleep time and internship performance was 0.6, and this feature was selected for in-depth analysis. The Pearson correlation coefficient method was used to analyze the association of different time windows, and it was found that the correlation coefficient between sleep time and internship performance one week later reached 0.75, showing a significant lag effect. The heat map intuitively shows the correlation strength between health indicators and internship performance. Sleep quality, exercise habits, and dietary regularity have correlation coefficients of 0.75, 0.6, and 0.5, respectively. The results of the multiple linear regression model show that for every 1 standard deviation increase in sleep quality, internship performance increases by 0.4 standard deviations, which is the most influential factor.

[0046] In view of the impact of specific health factors on the performance of different types of internship positions, internship positions are classified, and a prediction model of health factors and internship performance is constructed for each type of position. Through feature importance analysis, health factors that have a significant impact on this type of position are identified, and the correlation between internship performance and student health level is determined.

[0047] Hierarchical clustering algorithm was used to classify internship positions and obtain different types of internship position groups. The student health factor data and internship performance data of each type of position were preprocessed to generate a standardized data set for modeling. The Spearman correlation coefficient between each health factor and internship performance was calculated to preliminarily screen out health factors with significant correlation. For the standardized data set of each type of position, the random forest algorithm was used to build a prediction model of health factors and internship performance, and the parameters of the model were optimized by grid search method. Feature importance analysis was performed on the prediction model of each type of position, the influence of each health factor on internship performance was calculated, and the health factors were ranked according to the importance score. Internship position information and student internship performance data were obtained from the internship management database; based on the internship position information and student internship performance data, the hierarchical clustering algorithm was used to classify the internship positions and obtain different types of internship position groups. For different types of internship position groups, the student health factor data and internship performance data of each type of position were preprocessed, including using multiple interpolation method to handle missing values, outlier processing and feature standardization to generate a standardized data set for modeling. The Spearman correlation coefficient was calculated for each health factor and internship performance in the standardized data set, and health factors with significant correlation were preliminarily screened out. The random forest algorithm was used to construct a prediction model of health factors with significant correlation and internship performance. The parameters of the model were optimized by the grid search method to obtain the optimal prediction model for each type of position. The feature importance analysis was performed on the optimal prediction model to calculate the influence of each health factor on internship performance. The health factors were ranked according to the importance score to identify the key health factors that have a significant impact on internship performance in this type of position.

[0048] Specifically, the internship position information and student internship performance data were extracted from the internship management database. Combined with the position description and work content, the hierarchical clustering algorithm was used to classify the internship positions and obtain different types of internship position groups. The student health factor data and internship performance data of each type of position were preprocessed, including the use of multiple interpolation methods to handle missing values, outlier processing and feature standardization to generate a standardized data set for modeling. For each type of position, the Spearman correlation coefficient between each health factor and internship performance was calculated, and health factors with significant correlation were preliminarily screened out. For the standardized data set of each type of position, the random forest algorithm was used to construct a prediction model of health factors and internship performance. The number and depth of trees were optimized by the grid search method to obtain the optimal prediction model for each type of position. The feature importance analysis was performed on the prediction model of each type of position, the influence of each health factor on internship performance was calculated, the health factors were ranked according to the importance score, and the key health factors that had a significant impact on the internship performance of this type of position were identified. The partial dependence graph analysis method was used to visualize the nonlinear effect of health factors on internship performance. Based on the results of feature importance and partial dependence analysis, health factor thresholds were set for each type of position to clearly define the association between internship performance and student health level. 1000 job records were extracted from the internship management database and divided into five categories, such as technology, management, service, creativity and research, using a hierarchical clustering algorithm. The data of 200 students in service positions were preprocessed, and 15% of missing values ​​were filled using multiple interpolation, and 3% of outliers were detected and processed using the Z-score method. The Spearman correlation coefficient was calculated, and it was found that the correlation coefficient between sleep quality and internship performance was 0.72, and the significant p value was less than 0.01. The random forest model used the grid search method to optimize the parameters, and finally selected 100 trees with a maximum depth of 8. Feature importance analysis showed that sleep quality had an importance score of 0.35, ranking first; exercise frequency had a score of 0.25, ranking second. Partial dependence plot analysis showed that when the average daily sleep time increased from 6 hours to 8 hours, the internship performance showed a significant upward trend, but it tended to be flat after more than 9 hours. According to the analysis results, health factor thresholds were set for service positions: it is recommended that the average daily sleep time be 7-9 hours and the exercise frequency be no less than 3 times a week. These thresholds define the relationship between internship performance in service positions and students' health levels.

[0049] S107. For students with abnormal internship performance scores and health risk levels, personalized internship improvement suggestions are generated based on the correlation rules, and pushed to students and internship units through the industry-education integration health big data platform to continuously track students' internship status and health changes.

[0050] Obtain the student internship performance score and health risk level data in the industry-education integration health big data platform; filter out the student group data whose internship performance score is lower than the threshold and whose health risk level is higher than the threshold according to the preset threshold. Generate personalized internship improvement suggestions using a rule-based natural language generation method; extract student personal data from the student group data; match the student personal data with the improvement suggestion template; and obtain personalized internship improvement suggestions for each student. Send personalized internship improvement suggestions through the message push interface of the industry-education integration health big data platform; establish a student internship improvement tracking file; perform regular data collection tasks; obtain the latest health data of students from the health monitoring system; obtain the latest internship performance data of students from the internship management system; and analyze the changing trends of health data and internship performance data by the moving average method. Determine the student's progress and feedback; if the progress does not meet the expected goal, adjust the personalized internship improvement suggestion generation algorithm; if the feedback indicates that the improvement suggestion is not applicable, update the preset improvement suggestion template; and obtain an optimized personalized internship improvement suggestion generation mechanism.

[0051] Specifically, the student internship performance score and health risk level data were extracted from the industry-education integration health big data platform, and the screening threshold was set as the internship performance score below 60 points and the health risk level above 0.7. The eligible student groups were screened out through the data filtering algorithm. According to the established association rules, the rule-based natural language generation method was used for the screened student groups, combined with the preset improvement suggestion template and student personal data to generate personalized internship improvement suggestions. A student internship improvement tracking file was established, and the information stored in the file included: student basic information, internship performance score, health risk level, personalized internship improvement suggestions, and feedback. The generated personalized internship improvement suggestions were sent to the relevant students and internship units through the message push interface of the industry-education integration health big data platform. At the same time, a student internship improvement tracking file was established on the platform, and a feedback form was set up to collect the opinions of students and internship units. The implementation of the improvement suggestions was regularly evaluated, and an implementation report was generated every week to record the progress and feedback of students. Regular data collection tasks were set up to continuously obtain the latest health data and internship performance data of students from the health monitoring system and internship management system, and the changing trends of students' internship status and health conditions were analyzed by the moving average method. According to the students' progress and feedback, the improvement suggestions are dynamically adjusted to form a closed-loop optimization mechanism, and the algorithm for generating personalized internship improvement suggestions is continuously optimized. Data of 1,000 students were extracted from the industry-education integration health big data platform, and 50 students with internship performance scores below 60 and health risk levels above 0.7 were screened out. For student A, his internship performance score was 55 points, his health risk level was 0.8, and his average sleep time was 5 hours per day. The rule-based natural language generation system combined the template "It is recommended that you {improve the project} to improve {related performance}" and personal data to generate the suggestion: "It is recommended that you ensure 7-8 hours of sleep every day to improve work concentration and efficiency." The suggestion was sent to student A and the internship unit through the platform push interface, and a feedback form was created at the same time. One week later, the execution report showed that student A's sleep time increased to an average of 6.5 hours / day, and the internship performance score increased to 58 points. The 7-day moving average method was used to analyze the health data, and it was found that student A's health risk level was reduced to 0.75. Based on these data, the system dynamically adjusts the recommendations and generates new suggestions for improvement: "It is recommended that you schedule 30 minutes of aerobic exercise every day on the basis of your improved sleep to further improve your physical fitness and work performance.

[0052] Although the present invention has been described in detail above by general description and specific embodiments, it is obvious to those skilled in the art that some modifications or improvements can be made to the present invention. Therefore, these modifications or improvements made without departing from the spirit of the present invention all belong to the scope of protection claimed by the present invention.

Claims

1. An industry-education integration intelligent prediction for health big data and its implementation method, characterized in that: The method comprises: Obtain the student's identification code, extract the student's gastrointestinal disease history, work and rest schedule, and eating habits information from the school database, clean and standardize the extracted data, and construct the student health record data set; By connecting to the hospital information system, the internship work data of the interns are collected, and the collected internship work data are associated and matched with the student health record data set to build a health big data platform integrating industry and education; Based on the big data platform of health integrated with industry and education, we extracted the key characteristics of students’ personal eating habits, work and rest schedules, and internship work intensity, and used a decision tree model to predict the probability of students developing gastrointestinal diseases during their internship. According to the prediction results of the decision tree model, the key influencing factors leading to the high incidence of gastrointestinal diseases among students are determined. If students experience gastrointestinal discomfort during the internship, the health risk level of students is determined through the multi-layer perceptron neural network model based on their symptom characteristics, medical diagnosis results of the internship unit, medication treatment and key influencing factors. For students whose health risk level exceeds the preset threshold, an optimized internship schedule and internship position arrangement plan will be generated through the industry-education integration health big data platform according to the adjustment range of internship duration and the classification rules of internship positions. After being reviewed and approved by the internship adjustment approval process, the optimized plan will be pushed to students and internship units; Obtain students’ internship performance scores and internship unit evaluations during their internships, and conduct correlation analysis with students’ health status data during their internships to determine the correlation between internship performance and students’ health levels; For students with abnormal internship performance scores and health risk levels, personalized internship improvement suggestions are generated based on correlation rules, and pushed to students and internship units through the industry-education integration health big data platform to continuously track changes in students' internship status and health conditions.

2. The method according to claim 1, wherein: The student's identity code is obtained, the student's gastrointestinal disease history record, work and rest schedule and eating habit information are extracted from the school database, the extracted data is cleaned and standardized, and the student health record data set is constructed, including: Obtaining a student identification code, and extracting student health-related data from a school database based on the student identification code; Using data cleaning methods to process student health-related data to obtain cleaned student health data; Construct the student health record dataset based on the cleaned student health data; The student health record data set is encrypted by using the AES-256 encryption algorithm to obtain the encrypted student health record data set; The encrypted student health record data set is stored in a secure data storage system, and the user's identity is judged to be legitimate based on the preset access control rules. If the user's identity is legitimate, the user is allowed to access the decrypted data.

3. The method according to claim 1, wherein: The method is to connect to the hospital information system, collect the internship work data of the interns, associate and match the collected internship work data with the student health record data set, and build a health big data platform integrating industry and education, including: When obtaining the work data of internship students, an HL7 protocol request is sent according to the hospital information system application program interface to obtain the original data including internship position type, internship duration and internship duty period; XML parser is used to structure the raw data and generate a standardized internship work data set; Extract health record data from the student health record database and perform consistency checks on the health record data to determine whether the data format and content meet predefined standards; If it meets the predefined criteria, it is determined to be valid health record data; The standardized internship data set and the valid health record data are matched through the SQL INNER JOIN operation, and the association relationship between the two data sets is established based on the student's unique identifier to obtain the associated comprehensive data set; Apache Griffin is used to perform quality checks on the associated comprehensive data set to determine whether there are anomalies or inconsistencies in the data; If there is anomaly or inconsistency, mark the abnormal or inconsistent data; Build an industry-education integration health big data platform based on the associated comprehensive data set, and store the comprehensive data set in the Hadoop distributed file system; Apache Hive is used to perform data warehouse management and query operations on the stored comprehensive data sets to obtain multidimensional data cubes.

4. The method according to claim 1, wherein: According to the industry-education integration health big data platform, the key features of students' personal eating habits, work and rest schedules, and internship work intensity are extracted, and a decision tree model is used to predict the probability of students developing gastrointestinal diseases during their internship, including: Obtain the original data of students’ personal eating habits, work and rest schedules, and internship work intensity from the industry-education integration health big data platform; Clean and standardize the original data, remove outliers and missing values, and obtain the preprocessed feature data set; Calculate the internship work intensity index based on working hours and work type; Quantify the internship work intensity index into a standard score of 0-100; Use feature engineering methods to process feature data sets; Calculate dietary regularity index based on dietary time patterns; Calculate the sleep quality index based on sleep duration and sleep quality; Calculate the work intensity coefficient based on the internship work intensity; The CART decision tree algorithm is used to train the model on the training set data; The optimal tree structure and pruning parameters are selected through grid search combined with cross-validation to obtain a trained decision tree model; Use the decision tree model to predict the test set data; Calculate the probability of each student developing gastrointestinal diseases during internship; Determine whether students are susceptible to gastrointestinal diseases based on the probability value.

5. The method according to claim 1, wherein: According to the prediction results of the decision tree model, the key influencing factors leading to the high incidence of gastrointestinal diseases among students are determined. If students experience gastrointestinal discomfort during the internship, the health risk level of the students is determined through the multi-layer perceptron neural network model based on their symptom characteristics, medical diagnosis results of the internship unit, medication treatment and key influencing factors, including: Obtain the key influencing factors that lead to the high incidence of gastrointestinal diseases among students, quantify the key influencing factors, and obtain the key influencing factor feature vectors; According to the industry-education integration health big data platform, students' health-related data during internships were obtained, including gastrointestinal discomfort symptom data, medical diagnosis results, and medication details; Standardize and convert health-related data to generate symptom feature vectors, diagnosis result vectors, and treatment status vectors; The key influencing factor feature vector, symptom feature vector, diagnosis result vector and treatment status vector are concatenated to construct a complete input feature matrix; A multi-layer perceptron neural network model is used to train the input feature matrix and health risk level labels to obtain a training model for judging students' health risk levels; If a student is detected to have symptoms of gastrointestinal discomfort, their symptom, diagnosis and treatment data will be obtained, combined with key influencing factors, and input into the training model to determine the student's health risk level.

6. The method according to claim 1, wherein: For students whose health risk level exceeds the preset threshold, an optimized internship schedule and internship position arrangement plan will be generated through the industry-education integration health big data platform according to the adjustment range of the internship duration and the internship position type classification rules. After being reviewed and approved by the internship adjustment approval process, the optimized plan will be pushed to students and internship units, including: Obtain information on students whose health risk levels exceed the preset threshold in the industry-education integration health big data platform, and generate a list of students to be optimized; According to the internship duration adjustment rules, calculate the new internship duration for each student in the list of students to be optimized; Adopt internship position classification rules to match new suitable positions for students on the list of students to be optimized; Generate a preliminary optimization plan based on the new internship duration and new job matching results; The preliminary optimization scheme was input into the genetic algorithm, and the internship unit position capacity and student course schedule were set as constraints; The fitness function is used to evaluate the pros and cons of the plan, and the optimized internship schedule and internship position arrangement plan that meets the constraints is obtained; Submit the optimized internship schedule and internship position arrangement plan to the internship adjustment approval process; Automatic review through the rule engine, which includes a review rule library for internship duration compliance and job matching; If approved, the optimization plan will be sent to the relevant students and internship units; If the review fails, re-optimize; It also includes: adjusting the internship duration according to the health risk score; extracting job feature vectors from the internship job database, calculating the similarity with the student's personal feature vector, and selecting the internship job with the highest matching degree; The adjusting of the internship duration according to the health risk score specifically includes: obtaining the student's health risk score and the benchmark internship duration, the health risk score is generated by the health risk assessment module, and the benchmark internship duration is extracted from the internship management database; judging the student's health risk level according to the preset warning threshold and safety threshold, if the health risk score is higher than the warning threshold, calculating the new internship duration according to the shortening ratio corresponding to the risk level, if the health risk score is lower than the safety threshold, calculating the new internship duration according to the corresponding extension ratio; comparing the new internship duration with the shortest and longest time limits in the internship management regulations, if it exceeds the limit range, adjusting the new internship duration to the nearest boundary value to obtain the adjusted internship duration; calculating the utility value corresponding to each duration by the following formula; ∪=w1·(1-R)+w2·L+w3·S, where U represents utility, w1, w2 and w3 are weights, R represents risk score, L represents learning progress, and S represents satisfaction with the internship unit; select the maximum value from the utility value and output the corresponding internship duration as the optimal internship duration recommendation; The method extracts a job feature vector from an internship job database, performs similarity calculation with a student's personal feature vector, and selects the internship job with the highest matching degree, specifically comprising: obtaining a feature vector of a job in the internship job database; generating a standardized job feature matrix according to professional requirements and work intensity factors; receiving data carrying a student's health risk level, professional background, and personal characteristics; preprocessing and numerically converting the data to obtain a student's personal feature vector; using a weighted cosine similarity algorithm to calculate a similarity score between the student's personal feature vector and the job feature vector; sorting the similarity scores in descending order; if the similarity score is greater than a preset threshold, taking the corresponding job as a matching result; judging whether the work intensity of the job exceeds the tolerance corresponding to the student's health risk level; and if the work intensity does not exceed the tolerance, saving the student's job information to an internship recommendation list.

7. The method according to claim 1, wherein: The obtaining of the internship performance scores and internship unit evaluations of the students during the internship, and the correlation analysis with the health status data of the students during the internship, to determine the correlation between the internship performance and the health level of the students, includes: Obtain student internship performance rating data from the internship management system and student health status data from the health monitoring system; Integrate the internship performance score data with the health status data based on the student ID and timestamp to obtain a complete student internship-health data set; Perform natural language processing on the internship unit evaluation text data in the Student Internship-Health Dataset; Extracting keywords and sentiment trends from the text data, and converting the keywords and sentiment trends into numerical values; The Z-score method was used to detect outliers in the student internship-health data set and remove outliers; Standardize the data set to obtain a standardized data set; Perform principal component analysis on the standardized data set; Extracting the main features of the data set, reducing the dimension of the data set, and obtaining a feature matrix after dimensionality reduction; The Pearson correlation coefficient method was used to calculate the correlation coefficient between the indicators in the characteristic matrix; Construct a multiple linear regression model based on the correlation coefficients; The influence of each health indicator on internship performance was quantified through a multiple linear regression model; It also includes: classifying internship positions according to the impact of specific health factors on the performance of different types of internship positions, and constructing a prediction model of health factors and internship performance for each type of position; identifying health factors that have a significant impact on this type of position through feature importance analysis, and determining the correlation between internship performance and students' health levels, specifically including: using a hierarchical clustering algorithm to classify internship positions and obtain different types of internship position groupings; preprocessing student health factor data and internship performance data for each type of position to generate a standardized data set for modeling; calculating the Spearman correlation coefficient between each health factor and internship performance, and preliminarily screening out health factors with significant correlation; using a random forest algorithm to construct a prediction model of health factors and internship performance for the standardized data set of each type of position, and optimizing the parameters of the model through a grid search method; performing feature importance analysis on the prediction model for each type of position, calculating the degree of influence of each health factor on internship performance, and ranking the health factors according to importance scores; Internship position information and student internship performance data were obtained from the database; based on the internship position information and student internship performance data, the hierarchical clustering algorithm was used to classify the internship positions to obtain different types of internship position groups; for different types of internship position groups, the student health factor data and internship performance data of each type of position were preprocessed, including using multiple interpolation methods to deal with missing values, outlier processing and feature standardization to generate a standardized data set for modeling; the Spearman correlation coefficient was calculated for each health factor and internship performance in the standardized data set, and health factors with significant correlation were preliminarily screened out; the random forest algorithm was used to construct a prediction model for health factors with significant correlation and internship performance, and the parameters of the model were optimized through the grid search method to obtain the optimal prediction model for each type of position; the feature importance analysis was performed on the optimal prediction model, the influence of each health factor on the internship performance was calculated, the health factors were ranked according to the importance score, and the key health factors that significantly affected the internship performance of this type of position were identified.

8. The method according to claim 1, wherein: For students with abnormal internship performance scores and health risk levels, personalized internship improvement suggestions are generated based on the correlation rules, and pushed to students and internship units through the industry-education integration health big data platform to continuously track students' internship status and health changes, including: Obtain student internship performance scores and health risk level data from the industry-education integration health big data platform; According to the preset threshold, the data of the student group with internship performance scores below the threshold and health risk levels above the threshold are screened; Use rule-based natural language generation methods to generate personalized internship improvement suggestions; extracting individual student data from the student group data; matching the student personal data with the improvement suggestion template; Get personalized internship improvement advice for each student; Send personalized internship improvement suggestions through the message push interface of the industry-education integration health big data platform; Establish a student internship improvement tracking file; Perform regular data collection tasks; Obtain students’ latest health data from the health monitoring system; Get the latest student internship performance data from the internship management system; Analyze the changing trends of health data and internship performance data using the moving average method; judging students’ progress and providing feedback; If the described progress does not meet the expected goal, the algorithm for generating personalized internship improvement suggestions is adjusted; If the feedback indicates that the improvement suggestion is not applicable, then update the preset improvement suggestion template; The optimized mechanism for generating personalized internship improvement suggestions is obtained.

Images