AI large model-based exercise injury prediction system for students at school

Abstract

The invention relates to the technical field of artificial intelligence, and particularly provides a school student exercise injury prediction system based on an AI large model. The system comprises a data collection module, a data integration and preprocessing module, an AI prediction model module, a real-time monitoring and alarm module, a user interaction module and a feedback and optimization module. Original data flows into the data collection module from multiple sources, passes through the data integration and preprocessing module and then is input into the AI prediction model module, and an injury risk prediction result is generated; the real-time monitoring and alarm module gives an alarm according to a risk prediction result, the user interaction module provides feedback, the feedback and optimization module is used for updating an AI prediction model, and the system improves the accuracy of exercise injury prediction of students at school.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence (AI) technology, and in particular to a sports injury prediction system for school students based on an AI large model. Background Art

[0002] The physical health of students is an important part of school education, and sports injuries have a significant impact on students' physical and mental health. Traditional injury prevention relies on teachers' experience and students' self-reporting, which lacks systematicness and foresight. The rapid development of artificial intelligence (AI) technology, especially large language models (LLMs) and machine learning algorithms, has provided new possibilities for sports injury prediction.

[0003] For students in school, the school's physical health monitoring platform usually records physical fitness test data (such as vital capacity, sit-and-reach), while wearable devices (such as smart watches) can collect data such as heart rate, number of steps and sleep quality in real time. These data provide rich material for AI models, but existing technologies have not yet fully integrated these resources for injury prediction. The defects of existing technologies: rule-based systems rely on medical guidelines to set thresholds and lack personalization; simple machine learning models, such as logistic regression, are suitable for a single data source and have low prediction accuracy; commercial fitness applications provide injury assessments, but are not designed for student groups and have limited data integration capabilities. Summary of the invention

[0004] In view of this, the present invention provides a sports injury prediction system for school students based on an AI big model, so as to improve the accuracy of sports injury prediction for school students.

[0005] In a first aspect, the present invention provides a sports injury prediction system for school students based on an AI big model, the system comprising: a data collection module, a data integration and preprocessing module, an AI prediction model module, a real-time monitoring and alarm module, a user interaction module, and a feedback and optimization module; The raw data flows into the data collection module from multiple sources, and is input into the AI ​​prediction model module after passing through the data integration and preprocessing module to generate injury risk prediction results; the real-time monitoring and alarm module issues alarms based on the risk prediction results, the user interaction module provides feedback, and the feedback and optimization module is used to update the AI ​​prediction model.

[0006] Optionally, the data integration and preprocessing module is used to integrate the raw data collected from different sources into a unified format and perform preprocessing; Data integration: Combine wearable device data, school platform data, health records, and questionnaire data to form a dataset; Data cleaning: During the data cleaning process, abnormal data such as duplicate data, error records, and missing values need to be processed; Data normalization: Convert data with different units into a unified dimension; Feature engineering: Feature engineering transforms the original data into input data for AI prediction models to make predictions by extracting or generating new features.

[0007] Optionally, the AI prediction model module is used to analyze the integrated data using AI large model technology to predict the risk of sports injuries; a. Data reception and routing: The system receives the preprocessed data; The data router assigns tasks according to the data type; b. Run the LLM inference model to process structured data: Use the LLM inference model to process structured data and generate a preliminary injury risk score; c. Run the LLM inference model to process unstructured data: The LLM inference model extracts semantic information related to injury risk by processing unstructured text data, supplements the analysis of structured data; and combines the semantic analysis it generates with the injury risk score; d. Fusion prediction: The prediction fusion unit integrates the output of the LLM inference model, uses the attention mechanism to calculate the final risk score; generates an interpretive report, including a quantitative score and a semantic description; e. Result output: The system integrates the outputs of both through the prediction fusion unit to generate the final injury risk prediction result; The final injury risk prediction result is output in a structured format, including a risk score, risk factors, and an interpretive description, for the real-time monitoring and warning module to perform threshold judgment and warning through the final injury risk prediction result, and for the user interaction module to present the final injury risk prediction result.

[0008] Optionally, the rules for injury risk scoring include: The LLM inference model processes the input features through a multi-layer neural network and finally outputs a risk score between 0 and 100; The calculation of the injury risk score is based on the output layer of the LLM inference model. The activation function Sigmoid is used to map the original output of the LLM inference model to between 0 and 1, and then multiplied by 100 to obtain the final score. Its expression is: Injury risk score = Sigmoid(z) × 100; where z represents the linear output of the last layer of the LLM inference model, that is, the original prediction value of the LLM inference model for the injury risk; Sigmoid(z) = , which means mapping z to between 0 and 1, that is, the probability value of injury occurrence; ×100 means converting the probability value to a score between 0 and 100.

[0009] Optionally, the reasoning process of the LLM reasoning model includes: Text input preprocessing: segment the original text and remove stop words; Prompt word design: Design specific prompt words to guide the LLM reasoning model to extract semantic information related to injury risk; Model reasoning: The LLM reasoning model performs reasoning based on the input text and prompt words and generates a semantic report; Model compression: The system uses knowledge distillation technology to compress the model.

[0010] Optionally, the real-time monitoring and alarm module is used to monitor the injury risk prediction results output by the AI ​​prediction model module in real time, and send an early warning when the risk score is greater than the risk threshold.

[0011] Optionally, the user interaction module is used to provide an interface to present the final injury risk prediction results and health recommendations; the feedback and optimization module is used to collect user feedback for optimizing the AI ​​prediction model.

[0012] In the technical solution provided by the present invention, the system includes a data collection module, a data integration and preprocessing module, an AI prediction model module, a real-time monitoring and alarm module, a user interaction module, and a feedback and optimization module; the original data flows into the data collection module from multiple sources, and is input into the AI ​​prediction model module after passing through the data integration and preprocessing module to generate an injury risk prediction result; the real-time monitoring and alarm module issues an alarm based on the risk prediction result, the user interaction module provides feedback, and the feedback and optimization module is used to update the AI ​​prediction model. The system improves the accuracy of sports injury prediction for students in school. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0014] Figure 1 A schematic diagram of a sports injury prediction system for school students based on an AI big model provided by an embodiment of the present invention; Figure 2 A workflow diagram of the AI ​​prediction model module provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0015] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0016] It should be clear that the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0017] The terms used in the embodiments of the present invention are only for the purpose of describing specific embodiments, and are not intended to limit the present invention. The singular forms "a", "said" and "the" used in the embodiments of the present invention are also intended to include plural forms, unless the context clearly indicates other meanings.

[0018] It should be understood that the term "and / or" used in this article is only a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship.

[0019] The word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining" or "in response to detecting", depending on the context. Similarly, the phrases "if it is determined" or "if (stated condition or event) is detected" may be interpreted as "when it is determined" or "in response to determining" or "when detecting (stated condition or event)" or "in response to detecting (stated condition or event)", depending on the context.

[0020] Figure 1 A schematic diagram of a sports injury prediction system for school students based on an AI big model provided in an embodiment of the present invention is shown in FIG. Figure 1 As shown, the system includes: Data collection module, data integration and preprocessing module, AI prediction model module, real-time monitoring and alarm module, user interaction module, feedback and optimization module; The raw data flows into the data collection module from multiple sources, and is input into the AI ​​prediction model module after passing through the data integration and preprocessing module to generate injury risk prediction results; the real-time monitoring and alarm module issues alarms based on the risk prediction results, the user interaction module provides feedback, and the feedback and optimization module is used to update the AI ​​prediction model.

[0021] In the embodiment of the present invention, the data collection module is used to collect students' exercise and health data from multiple sources, providing a basis for subsequent analysis. The data sources include: wearable devices (such as smart watches): real-time collection of physiological indicators such as heart rate, steps, and sleep quality. School physical health monitoring platform: including physical fitness test results (such as weight, height, flexibility), physical education training logs. Student health records: including past medical history, past injury records. Questionnaires: questionnaires filled out by students and parents about lifestyle, nutritional status, and mental health. Students can also manually input activity logs or symptom records through the system's user interaction module. Input: No direct input (collected directly from external systems or users). Output: Raw data, including various formats and types of data. Function: Ensure that the system obtains comprehensive and real-time data support for subsequent injury risk prediction.

[0022] In the embodiment of the present invention, the data integration and preprocessing module is used to integrate the raw data collected from different sources into a unified format and perform preprocessing; Data integration: Merge wearable device data, school platform data, health records, and questionnaire data to form a data set; Data cleaning: In the process of data cleaning, in order to ensure the accuracy, consistency, and integrity of the data set, it is necessary to process abnormal data such as duplicate data, error records, and missing values; Data normalization: Convert data with different units into a unified dimension; Feature engineering: Feature engineering transforms the raw data into input data for prediction by an AI prediction model by extracting or generating new features.

[0023] In the embodiment of the present invention, the specific steps of data cleaning include: Duplicate data processing: Delete exactly the same records to ensure data uniqueness; Error record processing: Identify outliers and delete or correct them to ensure data rationality; Missing value processing: Delete records or fill in missing fields to ensure data integrity; As shown in Table 1, the processing methods and bases are summarized.

[0024] Table 1 Processing methods and bases 。

[0025] In the embodiment of the present invention, based on the characteristics of students' exercise and health data, the system designs several new features, such as training load ratio (training volume / recovery time), exercise intensity change rate, etc. These features are extracted from the raw data through specific screening formulas and analysis processes, ensuring a high correlation with exercise injury risk. The specific process and examples are as follows: Training load ratio (training volume / recovery time), the training load ratio is an indicator to measure the balance between students' training intensity and recovery time, and the calculation formula is: ; Training volume: Estimated by the daily exercise time (in minutes) or the number of steps (in steps) recorded by a wearable device. Recovery time: Calculated comprehensively based on sleep quality and daily rest time (in hours).

[0026] Sample data: Student ID = 001, Date = 2025-3-01: Training volume: Exercise time = 60 minutes, Number of steps = 10,000 steps (equivalent to a training volume of approximately 120 units); Recovery time: Sleep time = 7 hours, Rest time = 2 hours (Total recovery time = 9 hours). Calculate the training load ratio: ; Screening process and basis: The reason the training load ratio was selected as a feature is that its correlation with the risk of sports injuries has been confirmed through statistical analysis and medical research.

[0027] Rate of change of exercise intensity, which measures the short-term fluctuations in a student's exercise intensity, is calculated as: ; Exercise intensity: Estimated by heart rate (in bpm) or step frequency (in steps per minute).

[0028] Screening process and basis: The reason the rate of change of exercise intensity was selected as a feature is by analyzing its correlation with acute injuries (such as sprains). The rate of change of exercise intensity can reflect the short-term fluctuations in a student's exercise load, especially when the intensity suddenly increases after physical education classes or extracurricular activities, and is highly correlated with acute injuries. Compared with the original heart rate data, the rate of change can capture dynamic risk factors more effectively.

[0029] The training load ratio and the rate of change of exercise intensity were selected as key features because they capture the core risk factors of sports injuries in a quantitative way: Training load ratio: By integrating training volume and recovery time, it reflects the degree of fatigue of a student's body. Correlation analysis (r = 0.65) and medical research show that a high training load ratio is significantly correlated with chronic injuries such as muscle strains. Rate of change of exercise intensity: By capturing the short-term fluctuations in exercise intensity, it reflects the potential risk of acute injuries (such as sprains). Correlation analysis (r = 0.58) and feature importance scoring (0.15) confirm its predictive value.

[0030] The screening of these features is based on strict statistical analysis and medical evidence, ensuring a strong correlation and high interpretability between the features and the injury risk. For example, the calculation of the training load ratio not only considers exercise time but also incorporates recovery time, providing a more comprehensive assessment of injury risk compared to a single metric (such as the number of steps).

[0031] Input: The raw data comes from the data acquisition module. Output: Preprocessed data, including a cleaned, standardized, and feature-rich dataset. Function: Ensure data quality and consistency, and provide high-quality input for the AI model.

[0032] In the embodiments of the present invention, as Figure 2 shown, the AI prediction model module is used to analyze and integrate the data using AI large model technology to predict the risk of sports injuries; a. Data reception and routing: The system receives preprocessed data (structured + unstructured); the data router assigns tasks according to the data type (numeric / text); b. Running the LLM inference model to process structured data: Use the LLM inference model to process structured data (such as heart rate, number of steps, training load ratio, etc.) and generate a preliminary injury risk score; the LLM inference model combines a convolutional neural network (CNN) and a long short-term memory network (LSTM) and can effectively capture patterns and trends in time series data; c. Running the LLM inference model to process unstructured data: The LLM inference model processes unstructured text data to extract semantic information related to the injury risk, supplement the analysis of structured data; and combine the semantic analysis it generates with the injury risk score to provide more comprehensive support and explanatory suggestions for injury prediction; d. Fusion prediction: The prediction fusion unit integrates the output of the LLM inference model, uses the attention mechanism to calculate the final risk score; generates an explanatory report, including a quantitative score and a semantic description; the format is: {"risk_score": 30, "risk_factors": ["High training load", "Knee inflammation"], "explanation": "A heart rate of 180 bpm and self-reported pain indicate a risk of patellar inflammation. It is recommended to reduce running and apply heat compresses."} e. Result output: Provide an accurate injury risk prediction, providing a basis for subsequent monitoring and warning. The system integrates the outputs of both through the prediction fusion unit to generate the final injury risk prediction result, that is, output the risk prediction and generate a risk report; the final injury risk prediction result is output in a structured format, including a risk score, risk factors, and an explanatory description, for the real-time monitoring and warning module to perform threshold judgment and warning through the final injury risk prediction result, and for the user interaction module to present the final injury risk prediction result.

[0033] The final risk prediction result is in JSON format and contains the following fields: risk_score: the risk score (from 0 to 100), representing the probability of injury or illness; risk_factors: a list of the main factors leading to high risk; explanation: a semantic explanation of the risk score, including possible types of injuries or illnesses and suggestions.

[0034] Example of prediction result: The following is a specific example of a prediction result, based on the data of student ID = 002: Input data (structured): Heart rate: 120 bpm; Steps: 12,000 steps; Training load ratio: 15.0; Rate of change of exercise intensity: 0.2 (20%); Input data (unstructured): Symptom report: "Recently, my knees have been a bit painful after running, especially when going up and down stairs. The sleep quality is okay, but sometimes I wake up due to knee discomfort."

[0035] Inference result of the LLM inference model for structured data: Risk score: 62 (Sigmoid output based on structured data) LLM inference result: Symptom: Knee pain; Severity: Moderate; Possible injury or illness: Patellar inflammation; Final prediction result (in JSON format): json { "risk_score": 65, "risk_factors": ["High training load", "Knee pain", "Fluctuation of exercise intensity"], "explanation": "The student's heart rate (120 bpm), high training load ratio (15.0), and self-reported knee pain indicate a risk of patellar inflammation (moderate). It is recommended to reduce the running intensity, increase knee hot compress, and consult a doctor." }。

[0036] Example explanation: Risk score (65): Calculated comprehensively by fusing the score (62) of the LLM inference model and semantic analysis through the attention mechanism. A score of 65 indicates that the student has a 65% probability of suffering from sports injuries, belonging to a moderately high risk. Risk factors: Include "High training load" (the training load ratio of 15.0 exceeds the safety threshold), "Knee pain" (extracted from the symptom report by the LLM inference model), and "Fluctuation of exercise intensity" (the rate of change of exercise intensity of 20% is relatively high).

[0037] Explanatory description: Combining structured and unstructured data, clearly indicating the possible type of injury or illness (patellar inflammation) and providing specific suggestions for users to understand and take actions.

[0038] In the embodiments of the present invention, the rules for injury risk scoring include: The LLM inference model processes the input features through multiple layers of neural networks and finally outputs a risk score between 0 and 100; the calculation of the injury risk score is based on the output layer of the LLM inference model. The activation function Sigmoid is used to map the original output of the LLM inference model to between 0 and 1, and then multiplied by 100 to obtain the final score. Its expression is: Injury risk score = Sigmoid(z) × 100; where z represents the linear output of the last layer of the LLM inference model, that is, the original predicted value of the LLM inference model for the injury risk; Sigmoid(z) = , which means mapping z to between 0 and 1, that is, the probability value of the injury occurring; ×100 means converting the probability value to a score between 0 and 100 for easy intuitive understanding by the user.

[0039] The Sigmoid function is used to smoothly map any real number to between 0 and 1 and is often used for probability prediction in binary classification problems. In this system, the injury risk can be regarded as a binary classification problem (injury occurs or does not occur), and the Sigmoid function converts the original output of the model into the probability of the injury occurring. Risk score: By multiplying the probability value by 100, an intuitive score is generated. For example, a score of 30 means the injury risk is 30%, and a score of 70 means the injury risk is 70%. This conversion enables users to quickly judge the level of risk.

[0040] Example derivation process: Suppose the structured data input of student ID = 001 on a certain day is: heart rate: 80 bpm; number of steps: 10,000 steps; training load ratio: 13.33; exercise intensity change rate: 0.1; after multiple layers of processing by the LLM inference model (CNN extracts local features, LSTM captures time dependence), the linear output z of the last layer of the model is 0.5.

[0041] Derivation steps: Calculate the activation function Sigmoid: Sigmoid(0.5) = ; ≈0.6065, based on the exponential approximation of the natural logarithm; Sigmoid(0.5) = ≈0.6225.

[0042] Calculate the risk score: Risk score = 0.6225 × 100 = 62.25; After rounding, the risk score is 62.

[0043] In this example, the LLM inference model predicts that the injury risk score of the student on that day is 62 based on features such as the student's heart rate, number of steps, training load ratio, and rate of change of exercise intensity; this means that the student has a 62% probability of suffering from sports injuries, belonging to a moderately high risk level. Users (such as students or teachers) can take corresponding preventive measures according to this score, such as adjusting the training intensity or increasing the rest time.

[0044] Model training: The LLM inference model is trained using historical data, and the parameters are optimized through the binary cross-entropy loss function to ensure the accuracy of the predicted probability. Feature weights: The model automatically learns the importance of each feature. For example, the training load ratio may contribute more to the risk score. Application scenario: The risk score can be used for real-time monitoring. When the score exceeds a threshold (such as 50), the system will trigger an alarm.

[0045] In the embodiment of the present invention, the inference process of the LLM inference model includes: Preprocessing of text input: Perform operations such as word segmentation and stop word removal on the original text; Prompt design: Design specific prompts to guide the LLM inference model to extract semantic information related to injury risk; Prompts used: In order to extract injury-related information from student surveys and symptom reports, the system designs the following prompts: Prompt example 1: "Please analyze the following text, identify any physical discomfort or injury symptoms mentioned by the student, and evaluate their severity".

[0046] Prompt example 2: "Extract keywords related to sports injuries from the student's description, such as 'pain','swelling', 'fatigue', etc.".

[0047] Model inference: The LLM inference model makes inferences based on the input text and prompts, and generates a semantic report; Model compression: Since the LLM inference model has a large number of parameters, in order to run efficiently on the school server, the system uses knowledge distillation technology for model compression.

[0048] Algorithm principle: By training a smaller student model to imitate the output of the original LLM (teacher model), retaining the core capabilities while reducing the model size and computational requirements. Compression effect: The original LLM inference model has approximately 150 million parameters, and the compressed student model has approximately 30 million parameters. The inference speed is increased by 3 times, and the memory occupancy is reduced by 70%.

[0049] Example illustration: Suppose the symptom report submitted by student ID = 002 is: Text input: "My knees hurt a bit after running recently, especially when going up and down stairs. My sleep quality is okay, but sometimes I wake up because of knee discomfort."

[0050] Detailed processing process: Text preprocessing: Word segmentation results: ["recently", "running", "after", "knee", "a little", "pain", "especially", "is", "up and down", "stairs", "when", "feel", "obvious", ".", "sleep", "quality", "ok", "but", "sometimes", "will", "because", "knee", "discomfort", "wake up", "."]; After removing stop words: ["running", "after", "knee", "pain", "up and down", "stairs", "feel", "obvious", "sleep", "quality", "ok", "sometimes", "knee", "discomfort", "wake up"]; Prompt word application: The system combines the preprocessed text with the prompt word to generate the input: "Please analyze the following text, identify any physical discomfort or injury symptoms mentioned by the student, and assess their severity: 'Knee pain after running, especially when going up and down stairs. Sleep quality is okay, but sometimes I wake up because of knee pain.'

[0051] The reasoning process of the LLM reasoning model: Extract keywords: ["knee", "pain", "going up and down stairs", "discomfort"]; Severity Assessment: Based on "some pain" and "sometimes wakes up", it is judged to be a mild to moderate injury risk. Generate a report: json { "symptom": "knee pain", "severity": "moderate", "possible_condition": "Patellar inflammation" }.

[0052] Model compression application: The compressed student model runs on the school server, and the inference time is shortened from 2 seconds of the original model to 0.6 seconds, meeting real-time monitoring needs.

[0053] In an embodiment of the present invention, the real-time monitoring and alarm module is used to monitor the injury risk prediction results output by the AI ​​prediction model module in real time, and send an early warning when the risk score is greater than the risk threshold.

[0054] Set risk thresholds: Define risk thresholds for different types of injuries and illnesses. For example, the risk threshold for muscle strains is 25%, and for joint sprains is 30%. These risk thresholds are determined based on historical injury data and expert medical opinions. For example, a risk score of 25% for muscle strains is derived from analyzing the injury occurrence distribution in a similar student population, and a score above this level is associated with a significant increase in actual injuries. These thresholds are regularly reviewed and adjusted to be optimized as more data becomes available.

[0055] Early warning notifications: When a student's risk score exceeds the threshold, send early warnings to students, parents, and teachers via APP push, text message, or email.

[0056] Suggestion generation: Provide personalized preventive suggestions according to the risk type, such as reducing high-intensity training, increasing rest time, or consulting a doctor.

[0057] Input: Risk Predictions come from the AI prediction model module.

[0058] Output: Alerts, including risk scores, specific risk types, and preventive suggestions.

[0059] Function: Timely detect potential risks, help students and teachers take preventive measures, and reduce the occurrence of actual injuries.

[0060] In an embodiment of the present invention, the user interaction module is used to provide an interface to present the final injury risk prediction results and health suggestions; the feedback and optimization module is used to collect user feedback for optimizing the AI prediction model.

[0061] User interaction module: Student side: View personal risk scores, health suggestions, and training logs, and manually input activity logs or symptom records; Teacher side: Monitor the overall injury risk of the class and adjust the physical education class plan; Parent side: View the health status and risk alerts of their children; Users can provide feedback (such as the occurrence of actual injuries and the effectiveness of suggestions) for system optimization.

[0062] Input: Risk Predictions come from the AI prediction model module, and Alerts come from the real-time monitoring and early warning module.

[0063] Output: User Feedback, including evaluations of the system's predictions and suggestions; User-input Data, such as activity logs or symptom records.

[0064] Function: Provide personalized health management services, and at the same time collect user feedback and new data to optimize the system.

[0065] Feedback and Optimization Module: Feedback Collection: Collect feedback from students and teachers through the user interaction module, such as actual injuries and illnesses, and the effectiveness of suggestions; Model Iteration: Regularly update the AI ​​model with new data (including user feedback and new injury records) to improve prediction accuracy. Model updates are performed once a month to incorporate new data and feedback; Evaluation Method: Use A / B testing methods to evaluate the effect of model improvements, in which some users use the updated model and others use the previous version to compare prediction accuracy and user satisfaction. Automate model deployment through continuous integration (CI / CD) processes.

[0066] Input: User feedback comes from the user interaction module.

[0067] Output: Model Updates, used to improve AI prediction models.

[0068] Function: Through continuous optimization, ensure that the system's predictive capabilities keep pace with changes in actual demand.

[0069] The system in the present invention integrates multi-source data for student groups (such as wearable devices and school health platforms); uses the LLM reasoning model to significantly improve prediction accuracy, with an F1 score of 0.82; currently, the prediction of student sports injuries mainly relies on teacher experience and student self-reporting, which lacks systematicity and real-time performance. Traditional methods cannot integrate multi-source data (such as smart watch data and school health platform data), have low prediction accuracy, and cannot provide personalized prevention advice for each student. The system in the present invention uses the LLM reasoning model, which is suitable for school environments, significantly improves the efficiency of student health management, solves the above problems, realizes real-time monitoring and personalized advice, and fills the gap in existing technologies.

[0070] Through AI big model technology, this system significantly improves the accuracy and real-time performance of sports injury prediction, provides personalized health management solutions, improves school health management efficiency, and reduces long-term health costs through continuous optimization. These effects not only solve the shortcomings of traditional methods, but also bring positive social value to student health and school management.

[0071] In the technical solution provided by the present invention, the system includes a data collection module, a data integration and preprocessing module, an AI prediction model module, a real-time monitoring and alarm module, a user interaction module, and a feedback and optimization module; the original data flows into the data collection module from multiple sources, and is input into the AI ​​prediction model module after passing through the data integration and preprocessing module to generate an injury risk prediction result; the real-time monitoring and alarm module issues an alarm based on the risk prediction result, the user interaction module provides feedback, and the feedback and optimization module is used to update the AI ​​prediction model. The system improves the accuracy of sports injury prediction for students in school.

[0072] Those skilled in the art can clearly understand that for the convenience and conciseness of description, the specific working processes of the systems, devices, and units described above can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated herein.

[0073] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the scope of protection of the present invention.

Claims

1. A sports injury prediction system for school students based on AI big model, characterized by: The system includes: a data collection module, a data integration and preprocessing module, an AI prediction model module, a real-time monitoring and alarm module, a user interaction module, and a feedback and optimization module; The raw data flows into the data collection module from multiple sources, and is input into the AI ​​prediction model module after passing through the data integration and preprocessing module to generate injury risk prediction results; the real-time monitoring and alarm module issues alarms based on the risk prediction results, the user interaction module provides feedback, and the feedback and optimization module is used to update the AI ​​prediction model.

2. The system according to claim 1, characterized in that The data integration and preprocessing module is used to integrate the raw data collected from different sources into a unified format and perform preprocessing; Data integration: merge wearable device data, school platform data, health records and questionnaire data to form a data set; data cleaning: in the data cleaning process, it is necessary to deal with abnormal data such as duplicate data, erroneous records and missing values; data normalization: convert data in different units into a unified dimension; Feature Engineering: Feature engineering converts raw data into input data for AI predictive models to make predictions by extracting or generating new features.

3. The system according to claim 1, characterized in that The AI ​​prediction model module is used to analyze the integrated data using AI big model technology to predict sports injury risks; a. Data reception and routing: The system receives pre-processed data; The data router assigns tasks based on data type; b. Run the LLM reasoning model to process structured data: Use the LLM reasoning model to process structured data and generate a preliminary injury risk score; c. Run the LLM reasoning model to process unstructured data: The LLM reasoning model processes unstructured text data to extract semantic information related to injury risk, supplementing the analysis of structured data; and combines the semantic analysis it generates with the injury risk score; d. Fusion prediction: The prediction fusion unit integrates the output of the LLM inference model and uses the attention mechanism to calculate the final risk score; Generate an explanatory report containing quantitative scores and semantic descriptions; e. Result output: The system integrates the outputs of the two through the prediction fusion unit to generate the final injury risk prediction result; the final injury risk prediction result is output in a structured format, including risk score, risk factors and explanatory description, so that the real-time monitoring and alarm module can make threshold judgments and warnings through the final injury risk prediction result, and the user interaction module can present the final injury risk prediction result.

4. The system according to claim 3, characterized in that The rules for injury risk scoring include: The LLM inference model processes input features through a multi-layer neural network and finally outputs a risk score between 0 and 100. The calculation of the injury risk score is based on the output layer of the LLM inference model. The activation function Sigmoid is used to map the original output of the LLM inference model to between 0 and 1, and then multiplied by 100 to obtain the final score. Its expression is: Injury risk score = Sigmoid (z) × 100; Where z represents the linear output of the last layer of the LLM inference model, that is, the original prediction value of the injury risk by the LLM inference model; Sigmoid (z) = , which means mapping z to between 0 and 1, i.e. the probability value of injury; ×100 means converting the probability value into a score between 0 and 100.

5. The system according to claim 3, characterized in that The reasoning process of the LLM reasoning model includes: Text input preprocessing: segment the original text and remove stop words; Prompt word design: Design specific prompt words to guide the LLM reasoning model to extract semantic information related to injury risk; Model reasoning: The LLM reasoning model performs reasoning based on the input text and prompt words and generates a semantic report; Model compression: The system uses knowledge distillation technology to compress the model.

6. The system according to claim 1, characterized in that The real-time monitoring and alarm module is used to monitor the injury risk prediction results output by the AI ​​prediction model module in real time, and send an early warning when the risk score is greater than the risk threshold.

7. The system according to claim 1, characterized in that The user interaction module is used to provide an interface to present the final injury risk prediction results and health recommendations; the feedback and optimization module is used to collect user feedback for optimizing the AI ​​prediction model.

Images