An adaptive physical education course intelligent arrangement system

Abstract

The application discloses a kind of self-adapting sports teaching course intelligent arrangement system, belong to the cross technical field of education information and artificial intelligence. System uses cloud-edge-end collaborative distributed architecture, builds the whole cycle closed loop system of data perception, modeling decision, simulation verification, execution control, iterative optimization. System generates student dynamic digital portrait by multi-source heterogeneous data fusion, realizes course global optimal arrangement by CSP constraint solving and multi-agent reinforcement learning dual-engine, relies on digital twin preplay to complete conflict resolution, realizes whole-process management and control in combination with real-time closed-loop control and motion safety warning. The application greatly improves sports scheduling efficiency and individual adaptation ability, reduces course conflict and motion injury risk, improves teaching resource utilization, adapts to all-scenario sports teaching needs, has the value of large-scale promotion.

Description

Technical Field

[0001] This invention belongs to the technical field of educational informatization, artificial intelligence, Internet of Things, digital twins and multi-agent reinforcement learning, and specifically relates to an adaptive intelligent scheduling system for physical education courses that is based on the deep integration of multi-source heterogeneous data fusion, digital twin pre-simulation, multi-agent reinforcement learning optimization and traditional academic scheduling system. Background Technology

[0002] Physical education is a core component of my country's quality education system. Its core objectives are to comprehensively improve students' physical health, master specific sports skills, cultivate lifelong exercise habits, and foster teamwork and resilience. It is a crucial link in promoting students' all-round development in morality, intelligence, physical fitness, aesthetics, and labor. Curriculum design, as a core prerequisite for physical education, directly determines the efficiency of teaching resource utilization, the smoothness of teaching implementation, the effectiveness of personalized teaching, and the level of sports safety. With the comprehensive advancement of the digital and intelligent transformation of education, the state has successively introduced a number of policies to promote the deep integration of physical education and information technology, requiring physical education to upgrade from the traditional "uniform, experience-based, and static" model to a "personalized, precise, dynamic, and safe" model. Traditional physical education curriculum design models and existing intelligent technology solutions are no longer suitable for the core needs of physical education in the new era, specifically exhibiting two major core technological deficiencies:

[0003] I. Inherent defects of traditional manual scheduling and basic electronic academic affairs systems

[0004] Currently, the vast majority of primary and secondary schools and some universities in China still use a manual, experience-based curriculum design model. The basic physical education curriculum management systems introduced by a few institutions have only achieved electronic assistance for manual design and have not formed a truly intelligent design capability. Their core deficiencies are mainly reflected in the following five points:

[0005] 1. Insufficient Adaptation to Individual Student Differences: Traditional models employ a uniform "one-size-fits-all" approach to curriculum content, training intensity, and teaching pace, failing to adequately consider individual differences among students in areas such as physical fitness baseline, athletic ability, injury contraindications, interests, and learning abilities. Students with weaker physical constitutions are prone to frustration due to excessively difficult courses, potentially leading to sports injuries; while physically gifted students may struggle to improve effectively due to the simplicity of the curriculum, completely failing to achieve the teaching goal of "individualized instruction."

[0006] 2. The course scheduling is extremely inefficient and prone to conflicts: The scheduling of physical education courses requires the simultaneous coordination of multiple factors such as student schedules, teacher teaching time, venue usage rights, equipment allocation, and teaching objectives. Manual scheduling requires one-by-one coordination and repeated adjustments. Primary and secondary school teaching staff spend as much as 1-2 weeks each semester scheduling physical education courses. Even so, more than 30% of schools still experience problems such as course time conflicts, venue usage conflicts, teacher time conflicts, and insufficient equipment allocation. The subsequent adjustment costs are extremely high, which seriously disrupts the normal teaching order.

[0007] 3. Lack of dynamic adjustment capability and rigid curriculum plans: Once a traditional curriculum plan is determined, the teaching content, pace, and intensity remain fixed throughout the semester, making it impossible to adjust in real time according to students' learning progress, physical changes, shifts in interest, weather changes, damage to venues and equipment, or other unforeseen circumstances. For example, when outdoor teaching cannot be conducted due to rain, classes can only be temporarily suspended or unplanned content can be replaced; when students' overall skill mastery lags behind, the curriculum still needs to proceed according to a fixed schedule, resulting in a serious waste of teaching resources.

[0008] 4. Delayed teaching feedback, unable to form a closed-loop optimization: The traditional teaching effectiveness feedback relies solely on teachers' subjective evaluations and students' verbal feedback after class. Data collection is one-sided, delayed, and inaccurate, failing to provide data support for curriculum optimization and unable to build a closed-loop system of "curriculum design-teaching-feedback-optimization". The quality of curriculum design remains at the experience level for a long time and cannot be continuously improved through iteration.

[0009] 5. Low resource utilization efficiency and serious cost waste: Manual scheduling cannot achieve the optimal global allocation of venues, equipment and teachers, which easily leads to a polarization of resources being crowded during peak periods and idle during off-peak periods. Equipment maintenance and venue use lack data support, which increases the hardware investment and management costs of physical education teaching.

[0010] II. Limitations of Existing Intelligent Physical Education Teaching Arrangement Technology

[0011] In recent years, scholars and enterprises both at home and abroad have attempted to apply artificial intelligence, the Internet of Things, and big data technologies to the field of physical education, and have launched some intelligent curriculum design solutions. However, these solutions all have obvious technical shortcomings and cannot achieve large-scale implementation or full-scenario adaptation.

[0012] 1. Lack of global resource collaborative optimization capability: Existing technologies mostly use a single rule engine or basic optimization algorithm, which can only solve resource conflicts in a single dimension. They cannot achieve global optimal configuration under multi-dimensional constraints such as teaching syllabus requirements, teacher expertise matching, equipment turnover cycle, site maintenance window, and individual student needs, resulting in limited improvement in resource utilization.

[0013] 2. Lagging environmental dynamic response capability: Physical education teaching is highly dependent on outdoor meteorological conditions (temperature, humidity, precipitation, air quality, ultraviolet intensity) and indoor venue microenvironment. The existing system only supports manual temporary class rescheduling and lacks real-time environmental data collection, risk prediction and automatic rescheduling mechanism, which can easily lead to class interruption, idle resources or safety hazards.

[0014] 3. Disconnect between sports safety early warning and load regulation: Although existing technologies can collect students' physiological data through wearable devices, they are only used for post-class statistical analysis and have not built a safety management closed loop of "real-time monitoring - risk assessment - dynamic intervention - feedback optimization". The intensity of the course cannot be dynamically adjusted according to the students' real-time fatigue status and injury risk, and sports injury prevention still relies on human experience judgment.

[0015] 4. Poor practicality and insufficient compatibility of high-end algorithms: Some solutions adopt cutting-edge algorithms such as deep reinforcement learning and posture recognition, but they focus too much on single-skill training and optimization, and are not compatible with the core needs of traditional academic scheduling. They are complex to operate, have high hardware requirements, and high deployment costs, making them unsuitable for the implementation scenarios of basic physical education teaching in primary and secondary schools. At the same time, they lack structured teaching knowledge connections, the logic of skill teaching is chaotic, and they are prone to skipping steps in teaching, which affects the learning effect.

[0016] 5. Disconnect between teaching evaluation and scheduling iteration: The existing system has not established a quantitative feedback channel based on full data of the teaching process. Once the scheduling plan is generated, it is fixed and cannot be continuously optimized by driving the algorithm and plan through real-time teaching data. The level of intelligence remains at a primary stage.

[0017] In summary, existing technologies suffer from the dual contradiction of traditional models being unintelligent and intelligent models being impractical. They cannot solve the efficiency, conflict, and adaptability issues of traditional academic scheduling, nor can they achieve the personalized, secure, and global optimization goals of cutting-edge intelligent technologies. There is an urgent need for a full-cycle adaptive intelligent scheduling system for physical education courses that integrates the practicality of traditional academic scheduling with the innovation of cutting-edge intelligent technologies to fill the technological gap in the field of intelligent scheduling of physical education courses in China. Summary of the Invention

[0018] To address the shortcomings of existing traditional physical education curriculum design models and intelligent technology solutions, such as insufficient individual adaptation, low design efficiency, unbalanced resource utilization, lack of dynamic adjustment, weak safety control, broken closed-loop optimization, and poor implementation, the core objective of this invention is to provide a full-cycle adaptive intelligent curriculum design system for physical education based on digital twin pre-visualization and multi-agent reinforcement learning. Supported by digital twins, multi-agent reinforcement learning, and multi-source data fusion, this system comprehensively integrates the efficiency, ease of use, and practicality advantages of traditional physical education curriculum design, achieving intelligent, personalized, safe, efficient, and all-scenario adaptability in physical education curriculum design. Specific objectives are as follows:

[0019] 1. Construct multi-dimensional dynamic digital profiles of students to accurately depict individual characteristics such as physical fitness, skills, interests, injuries, and learning abilities, thereby completely solving the shortcomings of the traditional "one-size-fits-all" model and truly implementing the goal of "teaching according to aptitude" in physical education.

[0020] 2. By integrating constraint satisfaction problem solving with multi-agent reinforcement learning algorithms and combining them with genetic algorithm-assisted optimization, we can achieve rapid generation of course plans and global optimal scheduling, reducing the course scheduling time from 1-2 weeks to 1-2 days, reducing the course conflict rate to below 1.2%, and significantly improving scheduling efficiency.

[0021] 3. Build a digital twin virtual teaching environment to resolve potential conflicts in terms of venue, equipment, teachers, and physiological safety through pre-rehearsal simulation, so as to achieve conflict-free verification of the course plan and ensure the smooth implementation of teaching.

[0022] 4. Establish a real-time feedback and closed-loop control mechanism, and combine it with sports safety early warning and load management modules to realize three-level dynamic adjustment of course intensity, content and venue during class, and build a four-level sports safety protection system of "prediction-prevention-intervention-recovery" to reduce the incidence of sports injuries among students by more than 60%.

[0023] 5. Achieve overall collaborative optimization of teaching resources such as venues, equipment, teachers, and environment, improve resource utilization by more than 20%, and reduce the cost of physical education teaching management and hardware investment.

[0024] 6. It is compatible with primary and secondary schools, universities, and sports training institutions in all scenarios. The user interface is simple and easy to use, and the hardware configuration is universal, which lowers the threshold for deployment and use and has the value for large-scale industrial promotion.

[0025] 7. Construct a full-cycle optimization system of "data collection - preprocessing - profile construction - intelligent arrangement - pre-rehearsal verification - teaching implementation - real-time feedback - closed-loop iteration", and continuously improve the quality of physical education teaching by continuously driving the self-evolution of system algorithms and curriculum plans through data.

[0026] The adaptive physical education curriculum intelligent scheduling system disclosed in this invention adopts a distributed architecture of cloud-edge-device collaboration. It achieves asynchronous communication and state synchronization among modules through a standardized data bus and message queue, constructing a full-cycle closed-loop control architecture of "perception-modeling-decision-execution-feedback-iteration". The system as a whole includes fourteen core functional modules: multi-source data acquisition and fusion module, data preprocessing module, student digital profile construction module, teaching constraint knowledge graph module, curriculum resource library module, teaching objective input module, adaptive scheduling engine module, digital twin pre-playback and conflict resolution module, real-time feedback and closed-loop control module, safety early warning and load management module, curriculum output module, data storage module, system management module, and system interactive terminal. These modules work collaboratively to achieve full-process adaptive intelligent scheduling of physical education courses.

[0027] I. Multi-source data acquisition and fusion module

[0028] This module forms the foundation for the system's data input, enabling comprehensive collection of data across five major categories: students, teachers, facilities and equipment, environment, and teaching process. This provides high-quality data support for subsequent intelligent scheduling. The module is internally divided into student data collection units, teacher data collection units, facility and equipment data collection units, environmental and meteorological data collection units, and teaching process data collection units. Each unit collects data independently yet integrates it uniformly. Data collection equipment includes physical fitness testing devices, wearable smart bracelets, AI vision cameras, IoT sensors, environmental monitoring terminals, and teacher / student operating terminals.

[0029] 1. Student Data Collection Unit: Collects static and dynamic data from students. Static data includes basic information such as name, gender, age, grade, class, and student ID; physical fitness test data such as height, weight, vital capacity, 50-meter sprint, standing long jump, sit-and-reach, and pull-ups / sit-ups; and basic sports data such as sports-specific foundation, past sports experience, injuries, contraindications, and rehabilitation status. Dynamic data includes physiological data such as heart rate, heart rate variability, blood oxygen saturation, and body surface temperature; behavioral data such as movement standardization, movement trajectory, exercise intensity, and training duration; and learning feedback data such as course satisfaction, learning difficulties, changes in interests and preferences, and training performance.

[0030] 2. Teacher Data Collection Unit: Collects data such as teachers' basic information, teaching experience, professional title, special teaching qualifications, teaching schedule, teaching style, classroom teaching feedback, and student evaluations to achieve precise matching between teachers' expertise and course content.

[0031] 3. Venue and Equipment Data Acquisition Unit: Collects basic information such as venue type, area, location, projects that can be carried out, and number of people that can be accommodated; equipment information such as equipment name, specifications, quantity, venue ownership, service life, and maintenance cycle; and real-time status data such as venue vacancy / occupancy / maintenance and equipment availability / damage / idleness.

[0032] 4. Environmental meteorological data collection unit: Real-time collection of meteorological data such as temperature, humidity, precipitation probability, wind speed, PM2.5, air quality index, and ultraviolet intensity around the teaching site, providing a basis for adaptive adjustment of the course environment.

[0033] 5. Teaching Process Data Collection Unit: This unit collects data on the teaching process, including course start time, completion rate of teaching content, student participation, and unexpected events (student injuries, venue malfunctions, sudden weather changes), comprehensively recording the entire teaching implementation process. The module employs a time-series data alignment engine, using a dynamic time warping algorithm to eliminate clock drift across multiple devices, achieving timestamp consistency for physiological, behavioral, and environmental data, ensuring the synchronization and comparability of multi-source data.

[0034] II. Data Preprocessing Module

[0035] This module employs a layered preprocessing mechanism to standardize the collected raw data, removing invalid data, correcting abnormal data, and completing missing data to ensure the accuracy, integrity, and usability of the data. The specific processing flow consists of five steps:

[0036] 1. Data cleaning: Remove null values, garbled characters, and logically incorrect data from the original data. Use the isolated forest algorithm and the 3σ criterion to identify abnormal data. For data such as heart rate, physical fitness test scores, and exercise intensity that are outside the normal range, correct them by interpolation and mean replacement. Abnormal data that cannot be corrected is removed.

[0037] 2. Data deduplication: Deduplicatizes repeatedly collected data of the same type, retains the latest valid data, reduces data redundancy, and improves processing efficiency.

[0038] 3. Data Standardization: Standardize data formats and units of measurement, convert data such as height, weight, and exercise intensity into standardized values, and transform text data into computable feature vectors to meet the needs of subsequent algorithm operations.

[0039] 4. Data Completion: For a small amount of missing data, the STGCN spatiotemporal graph convolutional network is used to accurately complete the data by combining students' historical data and the average data of students of the same age in the same class, ensuring data integrity.

[0040] 5. Data Validation: The system employs a triple validation mechanism of range validation, format validation, and logical validation to verify the compliance of processed data. Data that fails validation is returned for reprocessing until it meets system requirements.

[0041] III. Student Digital Profile Construction Module

[0042] This module generates a five-dimensional dynamic digital profile of students—comprising physiology, skills, interests, risks, and learning abilities—through multi-level clustering, dynamic weight evaluation, and temporal modeling algorithms. This accurately depicts individual student characteristics and provides a core basis for personalized curriculum design. The specific implementation process is as follows:

[0043] 1. Feature Extraction: Extract six core feature parameters from the preprocessed student data: physical fitness features, skill mastery features, interest and preference features, learning ability features, injury and illness risk features, and physical fitness baseline features.

[0044] 2. Feature weight allocation: The Analytic Hierarchy Process (AHP) is used to dynamically allocate feature weights according to teaching objectives and teaching scenarios. In basic physical education, physical fitness and basic features have the highest weight, followed by interest features, while learning ability and injury risk features are adjusted according to the actual situation. In specialized sports training, the weights of skill features and interest features are increased to ensure that the profile matches the teaching needs.

[0045] 3. Profile Modeling: The K-means clustering algorithm is used to divide students into groups with excellent physical fitness, average physical fitness, and weak physical fitness according to their physical fitness level and athletic foundation. Then, the LSTM long short-term memory network is used to perform time-series modeling of students' historical learning trajectories, extracting personalized parameters such as skill progress speed, fatigue recovery rate, interest decay cycle, and skill forgetting curve. Combined with Bayesian inference, a student skill mastery map is constructed to generate a unique dynamic profile for each student.

[0046] 4. Profile Update: Adopting an incremental learning architecture, the profile is updated incrementally every week by automatically synchronizing students' real-time learning data, physiological data, and training results. When students experience injuries, illnesses, or significant changes in physical condition, an immediate update is triggered to ensure that the profile always accurately reflects the student's latest status.

[0047] IV. Knowledge Graph Module for Instructional Constraints

[0048] This module transforms structured and unstructured knowledge in the field of physical education, including expert experience, teaching syllabus, venue usage regulations, teacher qualification requirements, safety red line standards, and sports skill association rules, into a structured knowledge graph based on a graph database. This enables the visualization and reasoning of teaching constraints. The entity types in the knowledge graph include course type, venue type, equipment type, teacher qualifications, time period, weather conditions, and student group characteristics. Relationship types include prerequisite relationships, facilitating relationships, conflict relationships, compatibility relationships, and priority-weighted relationships. Each relationship edge has a dynamic weight that can be updated in real time based on group teaching data. The module supports SPARQL queries and graph neural network (GNN) inference, enabling rapid retrieval of feasible scheduling paths, detection of course conflicts, and verification of the rationality of teaching logic, providing constraint rule support for the adaptive scheduling engine.

[0049] V. Course Resource Database Module

[0050] This module adopts a hierarchical and categorized storage mechanism to construct a standardized and systematic physical education curriculum resource system, providing rich resource support for curriculum design. Resources are stored hierarchically and categorized according to four dimensions: project type, difficulty level, teaching scenario, and teaching objectives, specifically including five categories of resources:

[0051] 1. Course content resources: Covering all categories of sports such as basketball, football, track and field, martial arts, gymnastics, and swimming, the courses are divided into basic, intermediate, and advanced levels to suit students with different physical conditions and skill levels.

[0052] 2. Teaching plan resources: Includes detailed teaching plans for each project and difficulty level, clearly defining teaching objectives, key points and difficulties, teaching steps, teaching methods, time allocation, and assessment requirements.

[0053] 3. Teaching video resources: Stores action demonstrations, skill explanations, and training guidance videos to support in-class teaching and after-class review.

[0054] 4. Training plan resources: Develop personalized physical training, skills training, and rehabilitation training plans based on different physical fitness levels and teaching objectives.

[0055] 5. Assessment Standards Resources: Quantitative assessment standards are established for each project and difficulty level, clearly defining the assessment content, methods, and scoring rules. The module supports resource uploading, updating, searching, and automatic recommendation functions, accurately matching suitable resources based on student profiles and teaching objectives, eliminating the need for manual selection by teachers.

[0056] VI. Learning Objective Input Module

[0057] This module adopts a visual hierarchical input design, providing a convenient interface for academic administrators and teachers to enter teaching objectives. It supports three levels of objective input: overall teaching objectives, stage teaching objectives, and unit teaching objectives. It also allows the input of corresponding assessment requirements and achievement standards for each level of objective. The module includes standardized teaching objective templates adapted to primary and secondary school physical education syllabi, university physical education course requirements, and specific training objectives. Users can directly access and edit these templates, eliminating the need for starting from scratch and significantly improving objective input efficiency. Teaching objectives can be modified, deleted, and queried at any time, adapting to dynamic teaching needs.

[0058] VII. Adaptive Orchestration Engine Module

[0059] This module is the core decision-making unit of the system, constructing a dual-engine collaborative optimization architecture to achieve the globally optimal generation of the course plan:

[0060] 1. CSP Constraint Solver: Responsible for hard constraint verification, including class time mutual exclusion, site capacity limit, teacher qualification matching, sufficient equipment quantity, safe weather threshold, etc. It quickly filters out the initial feasible solution space and eliminates invalid solutions that do not conform to the basic teaching rules.

[0061] 2. MARL Multi-Agent Optimizer: Using the MADDPG algorithm, the course, venue, and teacher are modeled as independent agents. Based on a multi-objective reward function, collaborative strategy learning is carried out. The reward function covers five dimensions: skill improvement benefits, sports safety margin, resource turnover efficiency, teaching fairness, and environmental adaptability. The optimal candidate solution is output through iterative optimization.

[0062] 3. Genetic Algorithm-Assisted Optimization: Crossover, mutation, and selection operations are performed on feasible solutions to minimize course conflict rate, maximize individual student fit, and maximize resource utilization, further improving the rationality and adaptability of the solution. The final course solution output by the engine simultaneously meets teaching objectives, individual student needs, resource constraints, safety requirements, and environmental adaptability, achieving multi-dimensional collaborative optimization.

[0063] VIII. Digital Twin Rehearsal and Conflict Resolution Module

[0064] This module serves as the system's solution verification unit. Based on the Agent-Based Modeling simulation framework, it constructs a high-fidelity virtual teaching digital twin environment, materializing students, teachers, venues, and equipment as independent simulation agents. It injects student physiological response models, kinematic parameters, and teaching behavior rules, and uses Monte Carlo random simulation to perform a full-process preview of the candidate solutions output by the choreography engine.

[0065] 1. Conflict detection: Automatically identifies potential problems during simulation, such as site occupancy conflicts, equipment queuing timeouts, student heart rate exceeding limits, cross-interference in teaching, and logical errors in skills teaching.

[0066] 2. Conflict resolution: Simulated annealing and tabu search algorithms are used to perform local optimizations on high-conflict nodes, such as time period swapping, content replacement, class splitting, and equipment adjustment, until the conflict rate is reduced to the system's preset threshold (hardware conflict is 0, physiological conflict is ≤3%).

[0067] 3. Solution Verification: Solutions that pass the pre-run verification are marked as "executable solutions". Solutions that fail are returned to the orchestration engine for re-optimization to ensure that the officially released course solutions have no execution obstacles.

[0068] IX. Real-time Feedback and Closed-Loop Control Module

[0069] This module is deployed during the teaching execution phase to collect comprehensive data from the class at a high frequency of 10Hz. It then uses the Flink streaming computing engine to calculate the deviation index between the actual load and the planned load, triggering a three-level dynamic control strategy.

[0070] 1. Mild Deviation Adjustment: When the deviation index is within a reasonable range, initiate intensity reduction by decreasing the number of training sessions, lowering the intensity of exercise, and extending the rest time.

[0071] 2. Moderate Deviation Control: When the deviation index exceeds the reasonable range, content replacement is initiated, replacing high-intensity training with low-intensity skill practice, tactical explanation, or stretching recovery.

[0072] 3. Severe Deviation Control: In the event of safety risks or extreme environmental changes, an emergency training halt will be initiated, and alternative recovery plans will be pushed out to ensure student safety. The control results are fed back to the adaptive orchestration engine in real time for algorithm optimization and plan adjustment in the next cycle, forming a closed-loop control.

[0073] 10. Safety Early Warning and Load Management Module

[0074] This module constructs an LSTM-Attention bimodal fatigue prediction model, integrating heart rate variability, gait symmetry, electromyography fatigue index, and subjective fatigue score (RPE) data to accurately predict the probability of sports injury risk for students in the next 15-30 minutes.

[0075] 1. Risk warning: When the probability of damage risk is ≥0.65, the system will automatically issue a warning, generate a load reduction instruction, and link the control module to adjust the course intensity.

[0076] 2. Individualized recovery: Based on students' physiological data, personalized recovery suggestions such as cold compress duration, nutritional supplementation, and sleep intervention are generated and pushed to student terminals and the school doctor system simultaneously.

[0077] 3. Long-term risk tracking: Record students' historical fatigue data and injury status to build health risk profiles and provide a safety basis for long-term curriculum design.

[0078] XI. Course Output Module

[0079] This module will output the final validated course plan in various formats to suit different user needs:

[0080] 1. Output formats: including six categories: class timetable, teacher teaching plan, student individual training tasks, course suitability report, resource utilization report, and learning effect report.

[0081] 2. Output channels: Supports viewing on multiple terminals such as the system web page, mobile APP, and tablet. It can export Excel and Word format files, supports printing output, and pushes course adjustment information in real time through system messages and SMS.

[0082] 3. Access Control: Content is displayed hierarchically according to user roles. Students can only view their personal courses and training tasks, teachers can only view teaching-related plans, and academic affairs staff can view global data, ensuring data security.

[0083] XII. Data Storage Module

[0084] This module uses a combination of local storage and cloud storage:

[0085] 1. Local storage: Deployed at the school's edge nodes to store frequently accessed data such as real-time teaching data, current course plans, and user login information, ensuring rapid data response.

[0086] 2. Cloud Storage: Stores large volumes of data such as historical course plans, long-term student profiles, course resources, and system logs, enabling data backup and off-site access. The module uses AES encryption to encrypt student privacy and physiological data, employs a strict access control mechanism, and combines scheduled full backups with real-time incremental backups to prevent data loss and leakage, complying with educational data security standards.

[0087] XIII. System Management Module

[0088] This module enables full-process management and maintenance of the system, ensuring its stable, secure, and efficient operation.

[0089] 1. User Management: Supports registration, login, information modification, and password reset functions for four types of users: teachers, students, academic affairs staff, and administrators.

[0090] 2. Access Control: Assign corresponding operation permissions to different roles to prevent unauthorized operations. Permissions can be flexibly adjusted.

[0091] 3. System parameter settings: Supports custom settings for parameters such as data acquisition frequency, anomaly threshold, number of algorithm iterations, and backup cycle.

[0092] 4. Log Management: Records all user operation logs and system operation logs, supporting querying, exporting, and auditing, facilitating troubleshooting.

[0093] 5. System Maintenance: Real-time monitoring of system operation status, automatic fault alarm, and support for version updates and vulnerability patching.

[0094] XIV. System Interactive Terminal

[0095] This module serves as the user interface, including a teacher scheduling dashboard, a student personal app, an administrator resource scheduling dashboard, and an API interface for the academic affairs system. The interface is simple and easy to use, and the operation process conforms to the usage habits of traditional academic affairs personnel. At the same time, it supports seamless integration with the school's existing academic affairs system and education cloud platform, reducing system adaptation costs.

[0096] Compared with traditional physical education curriculum design models and existing intelligent technology solutions, this invention has outstanding technical advantages and beneficial effects, specifically reflected in the following eight aspects:

[0097] 1. Enhanced Personalized Adaptation Capabilities: Through a five-dimensional dynamic digital profile of students, the system accurately matches students' physical condition, skills, interests, injuries, and learning abilities, completely abandoning the "one-size-fits-all" approach. Students with weak foundations can steadily improve, while outstanding students can fully develop, truly achieving "teaching according to aptitude." Student course satisfaction has increased from 70% to over 88%.

[0098] 2. Significantly improved scheduling efficiency and conflict control: By integrating CSP solution, multi-agent reinforcement learning and genetic algorithms, the traditional 1-2 week scheduling time is shortened to 1-2 days, and the course conflict rate is reduced from more than 30% to less than 1.2%, which greatly reduces the workload of teaching staff and ensures the stability of teaching order.

[0099] 3. Significantly improved overall resource utilization: Enables coordinated scheduling of venues, equipment, and teachers, increasing resource utilization by 20%-24.6%, avoiding resource idleness and congestion, reducing hardware investment and maintenance costs, and lowering the cost of sports teaching and management.

[0100] 4. A closed-loop system for sports safety management throughout the entire lifecycle: Digital twin simulations help avoid risks in advance, real-time fatigue prediction and three-level dynamic intervention build a four-level safety protection system, reducing the incidence of sports injuries by 60%-68.3% and providing comprehensive protection for students' sports safety.

[0101] 5. Extremely strong environmental adaptability: Real-time access to meteorological and environmental data, and automatic rescheduling of courses within minutes when an early warning is triggered, without manual intervention, ensuring the continuous conduct of teaching activities in rainy, hot, and polluted weather.

[0102] 6. Full-cycle closed-loop iterative optimization: Construct a complete closed loop of "collection-arrangement-rehearsal-teaching-feedback-optimization". The system algorithm and curriculum plan continue to evolve with the teaching data, the level of intelligence is constantly improved, and the teaching quality is continuously improved.

[0103] 7. Highly practical for all scenarios: It is compatible with primary and secondary schools, universities, and sports training institutions. It has universal hardware configuration, simple operation interface, no need for professional technicians to maintain it, low deployment cost, and has value for large-scale industrial promotion.

[0104] 8. High data security compliance: Multiple security mechanisms such as encrypted storage, access control, and data backup are adopted to strictly protect student privacy and teaching data, and comply with relevant national laws and regulations on education data security and personal information protection. Attached Figure Description

[0105] Figure 1 Overall architecture block diagram of the adaptive physical education teaching curriculum intelligent scheduling system of this invention;

[0106] Figure 2 This invention provides a flowchart of the multi-source data acquisition, preprocessing, and student digital profile construction process.

[0107] Figure 3 Workflow diagram of conflict resolution between the adaptive orchestration engine and digital twin pre-simulation in this invention;

[0108] Figure 4 The present invention provides a flowchart of the real-time feedback closed-loop control and safety early warning load management process. Detailed Implementation

[0109] The technical solution, module operation process, and implementation effect of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. This embodiment takes the integration of primary and secondary schools and universities as an example, covering both basic education and higher education scenarios. It aims to fully illustrate the technical implementation and application effect of the present invention and is not intended to limit the scope of protection of the present invention.

[0110] I. System Deployment Architecture and Hardware / Software Configuration

[0111] This invention adopts a cloud-edge-device collaborative distributed deployment architecture, which is compatible with the school's existing network and hardware environment, requires no large-scale modification, and has low deployment costs.

[0112] 1. Cloud Server: Deployed on the education cloud platform, configured with a 32-core Intel Xeon CPU, 128GB of memory, and an NVIDIA A100 GPU accelerator card, running the Windows Server 2022 operating system. It is responsible for large-scale data fusion, maintenance of teaching constraint knowledge graph, training of multi-agent reinforcement learning algorithms, and digital twin simulation computing. It adopts Kubernetes to implement containerized microservice orchestration to ensure high-concurrency operation of the system.

[0113] 2. Edge computing nodes: Deployed in school computer rooms or gymnasium control rooms, equipped with industrial-grade IoT gateways and TensorRT lightweight AI inference engine, responsible for local data aggregation, real-time stream computing, and low-latency control command issuance, ensuring that the data processing and command response latency during class does not exceed 1 second.

[0114] 3. Terminal devices: These include student wearable smart bracelets (integrating heart rate, blood oxygen, and 6-axis IMU sensors), AI visual depth cameras, site IoT sensors (temperature, humidity, PM2.5, and pressure sensors), equipment RFID tags, teacher tablets, student mobile apps, and administrator scheduling screens. All terminals are networked via a hybrid network of 5G, gigabit WiFi, and BLE 5.0, using the MQTT protocol to ensure real-time data transmission, and critical commands are encrypted and authenticated using TLS.

[0115] 4. Software Configuration: The database uses MySQL 8.0 and Neo4j graph database, the algorithm framework is based on TensorFlow and PyTorch, the programming languages ​​are Python and Java, and the front-end interface is developed using the Vue framework to ensure system stability and ease of use.

[0116] II. Detailed Operation Flow of the System Core Modules

[0117] 1. Multi-source data acquisition and preprocessing workflow

[0118] At the beginning of the semester, the system initiated comprehensive data collection: the student data collection unit collected static physical fitness data from all students through physical fitness testing equipment, and collected information on interests, sports experience, and injuries through questionnaires; teachers entered their personal qualifications and teaching times through terminals; administrative staff entered basic information on venues and equipment; and environmental sensors collected meteorological data in real time. During the teaching process, wearable wristbands collected students' physiological data every second, AI cameras collected motion data every 5 seconds, and venue sensors collected resource status data every minute. All data was time-aligned and then transmitted to the data preprocessing module. The preprocessing module cleaned the raw data, removing abnormal heart rates and incorrect physical fitness data; deduplicated duplicate data and retained the latest valid data; standardized height, weight, and exercise intensity; and completed missing data such as sit-and-reach and equipment maintenance using a spatiotemporal graph convolutional network. Finally, after triple verification of range, format, and logic, qualified data was transmitted to subsequent modules.

[0119] 2. Student Digital Profile Construction Process

[0120] The system extracts five key characteristics of students: physical fitness, skills, interests, learning ability, and injuries. It then uses the analytic hierarchy process (AHP) to assign weights: physical fitness 0.3, basic skills 0.25, interests 0.2, learning ability 0.15, and injuries 0.1. Students are divided into three groups—excellent, average, and weak—through K-means clustering. An LSTM network is then used to model student learning trajectories, calculating skill progress rates and fatigue recovery speeds. Bayesian inference is combined to update the skill mastery graph, generating a five-dimensional dynamic digital profile. The system automatically updates the profile weekly, and the data is refreshed instantly when students experience injuries or significant improvements in physical fitness.

[0121] 3. Process of matching teaching objectives with course resources

[0122] Academic staff can use the teaching objective input module to access the system's built-in primary and secondary school physical education teaching syllabus templates, entering overall semester goals, monthly stage goals, and unit goals for individual lessons, and specifying assessment criteria. The course resource library module automatically matches corresponding project and difficulty level course content, teaching plans, training schedules, and assessment criteria based on teaching objectives and student profiles, pushing these to the adaptive scheduling engine without requiring teachers to manually select resources.

[0123] 4. Adaptive orchestration and digital twin rehearsal process

[0124] The adaptive orchestration engine first uses the CSP constraint solver to verify hard constraints such as class time, venue capacity, and teacher qualifications, generating 100,000 initial feasible solutions. Then, the MARL optimizer employs the MADDPG algorithm, iteratively optimizing based on five reward functions: skill improvement, safety, resources, fairness, and environment. A genetic algorithm assists in resolving remaining conflicts, outputting the top three candidate solutions. The digital twin pre-simulation module constructs a virtual teaching environment, injects candidate solutions into a simulation agent, runs 1000 Monte Carlo simulations, detects issues such as venue conflicts, equipment queuing, and excessive heart rate, and optimizes and adjusts through simulated annealing. Finally, it outputs a conflict-free, safe, and feasible optimal course solution, which is then transmitted to the course output module.

[0125] 5. Teaching Implementation and Real-time Closed-Loop Control Process

[0126] During the course execution, the real-time feedback module collects students' physiological, movement, and environmental data at a frequency of 10Hz and calculates the load deviation index. When a student's average heart rate exceeds the safe threshold or the fatigue index is too high, a mild adjustment is triggered to reduce the training intensity; when there is a sudden mild pollution weather event, a moderate adjustment is triggered, replacing outdoor track and field classes with indoor physical training; when an emergency occurs, such as a student injury, a severe adjustment is triggered, immediately suspending training and pushing out a recovery plan. All adjustment data is fed back to the scheduling engine in real time to optimize subsequent course plans.

[0127] 6. Safety Early Warning and Load Management Process

[0128] The safety warning module uses an LSTM-Attention model to fuse heart rate variability, gait, and subjective fatigue data to predict injury risk. When the risk probability is ≥0.65, the system automatically issues a warning, reduces the intensity of the course, and generates personalized recovery suggestions for students, pushing them to the student app and the school medical system to track students' health status over the long term and avoid the risk of chronic injuries.

[0129] 7. Course Output and System Management Process

[0130] The course output module generates class timetables, teacher teaching plans, student individual training tasks, and data reports, supporting viewing, exporting, and printing on multiple devices. The system management module assigns corresponding permissions to academic affairs staff, teachers, and students, records all operation logs, and allows administrators to monitor the system's operating status in real time, adjust parameters, and update versions to ensure stable system operation.

[0131] III. Implementation Results

[0132] After being deployed and run simultaneously in primary and secondary schools and universities for one semester, this system has achieved remarkable results:

[0133] 1. Primary and secondary school scenarios: The time for teaching staff to schedule classes has been shortened from 10 days to 1 day, the course conflict rate has dropped from 32% to 0.8%, the average physical health level of students has increased by 10%, the skill mastery rate has reached more than 90%, the incidence of sports injuries has decreased by 65%, and student course satisfaction has increased to 88%.

[0134] 2. In university settings: the utilization rate of venues and equipment increased by 24.6%, teachers' scheduling hours decreased by 72%, the incidence of sports injuries decreased by 68.3%, resource allocation costs decreased by 30%, and teaching quality assessment scores increased by 18.7%.

[0135] 3. Universal effectiveness across all scenarios: The system's ease of operation has been recognized by over 95% of teachers and academic affairs staff, and deployment and maintenance costs have been reduced by 40%. It has achieved the core goals of "personalized teaching, efficient scheduling, safe implementation, optimal resource utilization, and iterative improvement," and is fully adapted to the intelligent transformation needs of physical education teaching in the new era.

[0136] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention, and are not actually limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the present invention, such designs should fall within the protection scope of the present invention.

Claims

1. An adaptive physical education teaching curriculum intelligent scheduling system, characterized in that: It includes a distributed hardware architecture with cloud-edge-device collaboration and a closed-loop software functional system throughout its entire lifecycle. The software functional system includes, in sequence, a data perception layer, a modeling and decision-making layer, a simulation and verification layer, an execution and control layer, and an iterative optimization layer. The data perception layer completes the collection and standardization processing of multi-source heterogeneous teaching data. The modeling and decision-making layer completes the intelligent generation of course plans based on dynamic digital profiles of students and teaching constraint rules. The simulation and verification layer completes the resolution of plan conflicts and feasibility verification through digital twin pre-drills. The execution and control layer realizes real-time load control and safety early warning during class. The iterative optimization layer drives the continuous iteration of system plans and algorithms through the feedback of data from the entire teaching process.

2. The adaptive physical education teaching curriculum intelligent scheduling system according to claim 1, characterized in that: The data perception layer is equipped with a multi-source data acquisition and fusion module and a data preprocessing module. The multi-source data acquisition and fusion module simultaneously collects five categories of data: students, teachers, venue equipment, environmental meteorology, and teaching process, and achieves unified timestamps for multi-source data through a time-series data alignment engine. The data preprocessing module completes data cleaning, deduplication, standardization, completion, and triple compliance verification through a layered preprocessing mechanism.

3. The adaptive physical education teaching curriculum intelligent scheduling system according to claim 1, characterized in that: The modeling decision layer includes a student digital profile construction module. This module extracts six core features of students: physical fitness, skills, interests, learning ability, and injury risk. It dynamically allocates feature weights through the analytic hierarchy process and generates a five-dimensional dynamic student digital profile by combining K-means clustering and LSTM time-series modeling. It also supports weekly incremental updates and real-time updates for sudden situations.

4. The adaptive physical education teaching curriculum intelligent scheduling system according to claim 1, characterized in that: The modeling decision layer is equipped with an adaptive orchestration engine. The engine adopts a dual architecture of CSP constraint solver and MARL multi-agent optimizer, supplemented by genetic algorithm to complete the deep optimization of the solution. The CSP constraint solver completes hard constraint verification and generates an initial feasible solution space, and the MARL multi-agent optimizer completes the global optimal solution iteration based on multi-objective reward function.

5. The adaptive physical education teaching curriculum intelligent scheduling system according to claim 4, characterized in that: The MARL multi-agent optimizer uses the MADDPG algorithm to model courses, venues, and teachers as independent agents. The multi-objective reward function covers five core dimensions: skill improvement benefits, sports safety margin, resource turnover efficiency, teaching equity, and environmental adaptability.

6. The adaptive physical education teaching curriculum intelligent scheduling system according to claim 1, characterized in that: The simulation verification layer is equipped with a digital twin pre-play and conflict resolution module. The module constructs a high-fidelity virtual teaching environment based on the Agent-Based Modeling simulation framework. It completes the pre-play of the entire course process and potential conflict detection through Monte Carlo random simulation, and uses simulated annealing and tabu search algorithms to complete conflict resolution and local optimization of the solution.

7. The adaptive physical education teaching curriculum intelligent scheduling system according to claim 1, characterized in that: The execution control layer is equipped with a real-time feedback and closed-loop control module. The module collects full-dimensional data at a high frequency of 10Hz, calculates the deviation index between the actual and planned load through the Flink streaming computing engine, and triggers three-level dynamic control strategies of mild, moderate and severe based on the deviation level.

8. The adaptive physical education teaching curriculum intelligent scheduling system according to claim 1, characterized in that: The execution control layer is equipped with a safety early warning and load management module. The module constructs an LSTM-Attention bimodal fatigue prediction model, integrates multi-dimensional physiological and behavioral data to predict the risk of sports injuries in students, triggers early warning and load reduction based on risk level, and generates individualized recovery plans.

9. The adaptive physical education teaching curriculum intelligent scheduling system according to claim 1, characterized in that: The modeling decision layer also includes a teaching constraint knowledge graph module and a course resource library module. The teaching constraint knowledge graph module constructs a reasonable system of teaching rules and constraints based on a graph database. The course resource library module stores standardized physical education teaching resources in a multi-dimensional hierarchical classification, and supports automatic matching and recommendation based on student profiles and teaching objectives.

10. The adaptive physical education teaching curriculum intelligent scheduling system according to claim 1, characterized in that: The system also includes a course output module with hierarchical access control, an encrypted data storage module that combines local and cloud storage, a multi-role system management module, and a multi-terminal interactive entry point that is compatible with existing academic affairs systems, enabling adaptation to all teaching scenarios and ensuring the security and compliance of educational data.

Images