Efficient and collaborative physical training management system

Abstract

The invention discloses an efficient and collaborative physical training management system, and relates to the field of data processing systems. The efficient and collaborative physical training management system comprises the steps that multi-modal original data are synchronously collected through various sensors and monitoring equipment; preprocessing and fusing the collected data to generate a consistent fused data packet; generating and distributing an atomization training task based on the instruction of the coach, the team analysis input and the medical evaluation information; dynamically adjusting a training plan by using a deep reinforcement learning optimization model; and continuously monitoring key indexes and automatically triggering an intervention process. The system comprises a multi-source heterogeneous data acquisition and fusion module, a training task collaborative scheduling and distribution module, an adaptive training plan dynamic optimization engine and an active abnormal state monitoring and intervention module. According to the technical scheme, full-dimension training state evaluation, efficient and lossless information transmission and collaboration, personalized and optimized training plans and early recognition and active intervention of exercise risks can be achieved.

Description

Technical Field

[0001] This invention belongs to the field of data processing systems, and specifically relates to a highly efficient and collaborative sports training management system. Background Technology

[0002] With the continuous improvement of the scientific and intelligent level of sports training, efficient and collaborative training management has become a core support for improving athletes' competitive performance and the overall effectiveness of the team. Modern sports training involves multi-dimensional data collection, multi-role collaborative decision-making, and dynamic adjustment of training plans, covering multiple aspects such as physiological indicator monitoring, movement and posture analysis, tactical simulation and deduction, and psychological state assessment. The organic integration of these elements directly determines the accuracy and execution efficiency of the training program.

[0003] Among them, the sports training management system, as a key platform connecting coaches, athletes, medical support, and data analysis teams, aims to achieve full-process digitalization of training, real-time collaborative task allocation, and closed-loop automation of feedback mechanisms. However, existing technologies still face significant bottlenecks in practical applications.

[0004] Existing technologies generally suffer from data silos, with heterogeneous data from various sensors, video analytics systems, and health monitoring devices struggling to be effectively integrated, leading to one-sided training status assessments. Collaboration mechanisms are poorly designed, lacking a unified scheduling interface between coach instructions, athlete feedback, and logistical support, resulting in high information transmission delays and distortion rates. Training plan adjustments rely on human experience, failing to dynamically optimize based on real-time physical fitness data and environmental variables. Furthermore, the system lacks proactive response capabilities to emergencies (such as sports injury warnings), often initiating intervention only after the problem has occurred. These shortcomings severely restrict the collaborative efficiency and decision-making quality of high-level sports teams in high-intensity, fast-paced training scenarios, necessitating a sports training management system capable of deeply integrating multi-source data, intelligently driving task collaboration, and possessing adaptive optimization capabilities. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the prior art and provide a highly efficient and collaborative sports training management system that can effectively solve the problems in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: In one aspect, a highly efficient and collaborative sports training management system, comprising the following components:

[0007] The multi-source heterogeneous data acquisition and fusion module is used to collect multimodal training data in real time, including physiological index data, movement posture data, environmental parameter data, and video image data, through various sensors and monitoring devices deployed on the athlete's body, training equipment, and training environment. It performs timestamp alignment, data cleaning, and format standardization on the multimodal training data to construct a training data stream with a unified spatiotemporal reference. The training task collaborative scheduling and distribution module receives training plan instructions from the coach's terminal, data analysis results from the data analysis team, and physiological health assessment reports from the medical support team. Based on preset collaborative rules and task priorities, it generates atomized training tasks with clearly defined execution subjects, execution time windows, and resource dependencies, and pushes these atomized training tasks to the corresponding athlete terminals in real time. The system includes a logistics support terminal and an adaptive training plan dynamic optimization engine. This engine, based on the real-time training data stream output by the multi-source heterogeneous data acquisition and fusion module, combines athlete individual ability models and team tactical models from the historical training database. Using a built-in optimization algorithm, it performs online evaluation and dynamic adjustment of the intensity, content, and pace of the current training plan, generating an optimized version and submitting it to the training task collaborative scheduling and distribution module. An active abnormal state monitoring and intervention module continuously monitors key performance indicators and physiological safety thresholds in the real-time training data stream. When a data pattern deviates from a preset normal fluctuation range or triggers a preset warning rule, it automatically generates warning information of different levels and initiates corresponding intervention plans. These intervention plans include suspending training tasks, medical personnel intervention, or emergency adjustments to the training plan.

[0008] Preferably, the multi-source heterogeneous data acquisition and fusion module specifically includes a data interface adaptation submodule, a spatiotemporal alignment submodule, and a feature extraction submodule. The data interface adaptation submodule is configured to support multiple communication protocols such as Bluetooth 5.0, Wi-Fi 6, and ZigBee 3.0, and is used to receive raw data from heart rate monitors, electromyography sensors, inertial measurement units, GPS receivers, and high-speed cameras. The spatiotemporal alignment submodule adopts a clock synchronization mechanism based on the Network Time Protocol (NTP) to apply a unified timestamp to all input data streams, and uses a Kalman filter algorithm to correct the position coordinates of sensor data with spatial drift. The feature extraction submodule performs feature calculations for data of different modalities, extracts real-time estimates of heart rate variability (HRV) and maximum oxygen uptake (VO2max) from physiological index data, extracts 3D coordinate sequences of joint angles, angular velocities, and motion trajectories from motion posture data, and extracts the skeletal key point sequences of athletes from video image data using a convolutional neural network.

[0009] Furthermore, the training task collaborative scheduling and distribution module integrates a task dependency graph and a resource conflict detector. The task dependency graph stores the temporal dependencies and logical prerequisites between atomic training tasks in a directed acyclic graph structure, ensuring that the task distribution order meets the training process constraints. Before distributing a task, the resource conflict detector verifies the availability of the venue, equipment, or medical support resources requested by the task within the current time window. If a conflict is detected, it automatically triggers the rescheduling of the task execution time or sends a resource coordination request to the coach terminal.

[0010] Furthermore, the core optimization model of the adaptive training plan dynamic optimization engine is a deep reinforcement learning network based on the actor-critic framework. This network takes the athlete's real-time physiological load index, technical movement completion score, and environmental temperature and humidity as state inputs, and takes the adjustments to training intensity, training content combination, and rest interval as action outputs. Its objective function is... The aim is to maximize long-term training benefits, where R(s) t ,a t ) is an immediate reward function that comprehensively considers the improvement of training effect and the risk of fatigue accumulation, and γ is a discount factor with a value of 0.95.

[0011] Preferably, the proactive abnormal state monitoring and intervention module adopts a dual early warning mechanism that combines an unsupervised anomaly detection algorithm based on isolated forest with logical judgment based on rule reasoning. The isolated forest algorithm performs rapid anomaly score calculation on the high-dimensional training data stream to identify data points that deviate significantly from historical normal patterns. The rule reasoning logic predefines a set of judgment conditions for sports injuries, over-fatigue, and environmental risks. When real-time data simultaneously meets the condition that the isolated forest anomaly score exceeds the threshold of 0.65 and triggers at least one rule judgment condition, the module determines it as a high-risk event and initiates the highest-level intervention plan.

[0012] On the other hand, a highly efficient and collaborative sports training management method, the specific steps of which are as follows:

[0013] Step S110: Simultaneously collect multimodal raw data of athletes during training by deploying multiple sensors and monitoring devices. The multimodal raw data includes at least physiological electrical signals, three-dimensional kinematic data, environmental parameters, and high-definition video streams.

[0014] Step S120: The collected multimodal raw data is preprocessed and fused in real time. The preprocessing includes data noise reduction, outlier removal and missing value imputation. The fusion includes data stream splicing based on a unified time reference and multi-source feature association based on motion scene to generate a consistent fused data package.

[0015] Step S130: Based on the received coach instructions, team analysis input, and medical assessment information, and combined with the real-time training status reflected by the fusion data packet, a series of atomic training tasks with execution sequences and resource constraints are generated through a task decomposition algorithm, and the atomic training tasks are distributed to the designated execution terminals.

[0016] Step S140: Using the embedded deep reinforcement learning optimization model, perform online performance evaluation on the currently executing training plan, and dynamically calculate the adjustment parameters of training intensity, training content ratio or rest period based on the evaluation results and preset optimization objectives, and generate adjustment instructions.

[0017] Step S150: Continuously monitor the key indicators in the fused data packet. When the anomaly detection algorithm identifies data characteristics that conform to the preset risk mode, automatically trigger and execute the corresponding intervention process. The intervention process includes sending an alarm to the relevant terminal, pausing the current training task, or calling the backup training scheme.

[0018] The beneficial effects of this invention are:

[0019] 1. Through the multi-source heterogeneous data acquisition and fusion module, the deep integration and unified processing of physiological, action, environmental and video data were realized, which completely broke down data silos and provided a complete and consistent data foundation for all-dimensional training status evaluation.

[0020] 2. The training task collaborative scheduling and distribution module, through atomic task management and resource conflict detection, enables efficient and lossless transmission and collaboration of instructions and feedback among coaches, athletes, medical and logistics teams, significantly reducing information delays and decision distortion.

[0021] 3. The adaptive training plan dynamic optimization engine can automatically and accurately adjust training parameters based on real-time data and historical models, freeing the training plan from excessive reliance on human experience and achieving personalization and optimization of the training process.

[0022] 4. The proactive abnormal state monitoring and intervention module combines unsupervised abnormality detection with rule-based reasoning to achieve early identification and proactive intervention of risks such as sports injuries and over-fatigue, greatly improving the safety and reliability of the training process. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of the overall technical architecture of the efficient and collaborative sports training management system proposed in this invention;

[0024] Figure 2 This is a schematic diagram of the core principle framework of the adaptive training plan dynamic optimization engine in this invention. Detailed Implementation

[0025] The present invention will be further described below with reference to the accompanying drawings and specific embodiments. The illustrative embodiments and descriptions herein are used to explain the present invention, but are not intended to limit the present invention.

[0026] Example 1

[0027] At the provincial-level professional track and field training center, the multi-source heterogeneous data acquisition and fusion module conducts comprehensive data collection and processing for sprinters' daily training. (See also...) Figure 1 The system captures real-time three-dimensional kinematic data during the 100-meter sprint using inertial measurement units deployed on the athlete's torso and limbs, including hip flexion and extension angles, knee swing angular velocities, and body center of gravity displacement trajectory. A medical-grade heart rate monitor worn on the athlete's chest collects electrocardiogram signals once per second, accurately calculating heart rate variability. GPS receivers installed on the starting blocks and track edges record the athlete's coordinates throughout the race at a 50 Hz sampling rate. Eight high-speed cameras deployed around the training ground simultaneously capture side, front, and overhead video streams of the athlete at 120 frames per second. An environmental monitoring station continuously collects track surface temperature, air humidity, and wind speed and direction parameters. The data interface adapter submodule receives data from the inertial measurement units and heart rate monitors via Bluetooth 5.0, receives video streams from the high-speed cameras via Wi-Fi 6, and collects environmental sensor readings via ZigBee 3.0, establishing a unified device identification mapping table for all input data. The spatiotemporal alignment submodule uses the Network Time Protocol (NTP) to control the clock deviation of each device to within 5 milliseconds. It applies a Kalman filter algorithm to the latitude and longitude coordinates output by the GPS receiver to eliminate position drift errors caused by signal reflection, ensuring strict spatiotemporal alignment of all data streams. The feature extraction submodule performs frequency domain analysis on the ECG signal, extracting the heart rate variability power spectral density value in the 0.04 to 0.15 Hz frequency band. It uses a quaternion solving algorithm on the inertial measurement unit data to output real-time 3D Euler angle sequences for the athlete's shoulder, elbow, hip, knee, and ankle joints. It applies an OpenPose skeleton recognition model based on a convolutional neural network to the high-speed camera video stream, extracting the pixel coordinate sequences and motion velocity vectors of 25 key body points of the athlete.

[0028] The training task coordination and distribution module receives today's training plan from the coach via the terminal. This plan includes 6 sets of 100-meter interval runs, 4 sets of start reaction drills, and 3 sets of core strength training. It also receives yesterday's training data analysis report from the data analysis team, which points out a technical defect in the athlete's insufficient forward lean angle in the second half of the sprint. Additionally, it receives a physiological health assessment report from the medical support team, confirming that the athlete's blood lactate clearance rate is within the normal range, but the Achilles tendon load requires close monitoring. The module's internal task dependency graph establishes the logical relationships between training tasks using a directed acyclic graph structure: start reaction drills must be performed before the 100-meter interval runs, and core strength training must be scheduled after all running training. When the resource conflict detector detects that lane three is already occupied by the long jump team during the planned time slot, it automatically adjusts the 100-meter interval run task to lane four and sends a venue change notification to the coach's terminal. The final generated atomized training tasks include: starting reaction training in the starting training area from 9:00 to 9:20, requiring the use of two sets of starting blocks; 100-meter interval running in lane four from 9:30 to 10:30, requiring heart rate monitoring equipment; and core strength training in the strength training area from 11:00 to 11:40, requiring a yoga mat and resistance bands. These tasks are pushed to the designated athletes' smart wristbands, starting block control terminals, and equipment management terminals via encrypted data channels.

[0029] The adaptive training plan dynamic optimization engine, based on real-time collected athlete physiological and movement data, combined with the athlete's personal ability model in the historical training database, initiates online optimization calculations using a deep reinforcement learning network. The network input state vector includes the average heart rate, VO2 max utilization, trunk lean angle deviation, stride frequency variation coefficient, and environmental temperature and humidity factors for the current 100-meter sprint. The output movement vector includes the intensity adjustment coefficient, rest interval correction value, and key technical reminders for the next training session. The immediate reward function in the objective function comprehensively considers three indicators: technical improvement, cumulative physiological load, and training goal completion rate. When the network detects that the athlete's heart rate recovery rate decreases by 15% and the trunk lean angle deviation remains greater than 3 degrees for two consecutive sprints, it automatically generates training plan adjustment instructions: reducing the number of remaining 100-meter interval sprint sets from 4 to 2, extending the rest time between sets from 3 minutes to 5 minutes, and highlighting the technique of maintaining trunk lean on the athlete's terminal. The optimized training plan version is submitted to the training task collaborative scheduling and distribution module for redistribution via the application programming interface.

[0030] The proactive abnormal state monitoring and intervention module continuously monitors key indicators in the fused data stream. When the athlete performs the fifth set of the 100-meter run, the isolated forest algorithm detects an abnormal score of 0.72 in the knee varus angle sequence during the landing phase. Simultaneously, the rule-based reasoning logic determines that the athlete's historical Achilles tendon load data has exceeded the warning line three times consecutively, and the current track temperature has reached 32 degrees Celsius, meeting the trigger conditions for a high-risk event. The module immediately activates the highest-level intervention plan: sending a vibration alarm command to the athlete's wristband, pushing a knee joint protection warning message to the coach's terminal, sending an emergency dispatch request to the medical station, and forcibly suspending the current training task through the training task coordination scheduling and distribution module. After the medical personnel arrive, the system automatically retrieves a comparison chart of the athlete's historical Achilles tendon ultrasound images and current biomechanical parameters to assist in on-site diagnosis.

[0031] Example 2

[0032] At professional basketball club training bases, multi-source heterogeneous data acquisition and fusion modules conduct multimodal data collection for team tactical training. (See also...) Figure 1 The system uses 12 optical tracking cameras installed on the stadium roof to capture the movement trajectory of each athlete and the ball's path in real time. Athletes wear smart protective gear with built-in electromyography (EMG) sensors to collect data on the activation levels of major muscle groups. Microphone arrays deployed around the court record audio of coaching instructions and player communication. An environmental monitoring system collects real-time data on illuminance, air quality index, and ground friction coefficient within the venue. The data interface adaptation submodule receives optical tracking data via a dedicated fiber optic network, collects EMG sensor data via a near-field communication protocol, and transmits audio streams via a wireless LAN. The spatiotemporal alignment submodule uses a hardware synchronization signal generator to ensure that the frame synchronization error of all optical cameras is less than 1 millisecond, and applies an adaptive filtering algorithm to the EMG signals to eliminate motion artifacts. The feature extraction submodule extracts athletes' instantaneous velocity, acceleration, change of direction angle, and team formation dispersion indicators from the optical tracking data; calculates the root mean square value and median frequency of eight major muscle groups, including the quadriceps and gastrocnemius, from the EMG signals; and separates the frequency of coaching tactical instructions and player response delays from the audio stream.

[0033] The training task coordination and distribution module receives the head coach's input via tablet, indicating that today's training focus is on zonal defense tactical drills. It also receives reports from the data analysis team on the opponent's tactical preferences and player fatigue index assessments from the team doctor. The task dependency graph ensures that tactical explanations must precede field drills, and individual technical training must alternate with team training. A resource conflict detector automatically relocates training to a backup venue and reallocates locker room usage time when a conflict occurs at the originally scheduled training venue. The generated atomized training tasks include: 14:00-14:30 zonal defense theory explanation in the tactical analysis room (requiring multimedia projection equipment); 14:40-16:10 live-fire drills in Training Hall 2 (requiring the use of halves 5-8); and 16:30-17:00 individual player relaxation training in the rehabilitation room (requiring fascia guns and ice packs). These tasks are distributed to player mobile terminals, the stadium control center, and physiotherapist workstations via the club's internal management system.

[0034] The adaptive training plan dynamic optimization engine uses real-time collected team tactical execution data, combined with the team's tactical success rate model from a historical database, to dynamically adjust the system using a deep reinforcement learning network. The network input state vector includes defensive formation maintenance, help defense response time, number of successful tackles, and muscle fatigue index of key players; the output action vector includes the number of tactical drill rounds, offensive / defensive transition frequency, and individual player substitution plans. When the system detects that the team has three consecutive defensive lapses and the quadriceps activation level of key players has decreased by 20%, it automatically generates training plan optimization instructions: change the zone defense drill from full-court to half-court, reduce the offensive / defensive transition pace from 15 seconds to 22 seconds per possession, and rotate two substitute players onto the field. These optimization instructions are transmitted to the coach's headset in real-time via a digital voice system.

[0035] The proactive abnormal condition monitoring and intervention module calculates the knee joint torque during sudden stops and changes of direction in real time using optical tracking data. When it detects a sudden increase in the force on the lateral side of a player's left knee joint, triggering an isolated forest anomaly score of 0.68, and the rule base determines that the player has a history of anterior cruciate ligament injury and the field friction coefficient is below the standard value of 0.1, the module immediately activates the intermediate intervention plan: sending a tactile alert to the player's smart brace, sending a biomechanical parameter abnormality report to the team doctor's terminal, and automatically inserting a 5-minute mandatory rest period through the training task coordination scheduling and distribution module. The system simultaneously retrieves the player's historical injury imaging data and rehabilitation training plan to provide decision support for the medical team.

[0036] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

[0037] The technical solutions of the present invention are not limited to the specific embodiments described above. Any technical modifications made in accordance with the technical solutions of the present invention fall within the protection scope of the present invention.

Claims

1. A highly efficient and collaborative sports training management system, characterized in that, The system includes the following components: a multi-source heterogeneous data acquisition and fusion module, which is used to collect multimodal training data in real time, including physiological index data, movement posture data, environmental parameter data and video image data, through various sensors and monitoring devices deployed on the athlete's body, training equipment and training environment, and to perform timestamp alignment, data cleaning and format standardization processing on the multimodal training data to construct a training data stream with a unified spatiotemporal reference. The training task collaborative scheduling and distribution module is used to receive training plan instructions from the coach terminal, data analysis results from the data analysis team, and physiological health assessment reports from the medical support team. Based on preset collaborative rules and task priorities, it generates atomized training tasks with clear execution subjects, execution time windows, and resource dependencies, and pushes the atomized training tasks to the corresponding athlete terminals and logistics support terminals in real time. An adaptive training plan dynamic optimization engine is used to evaluate and dynamically adjust the intensity, content, and pace of the current training plan online based on the real-time training data stream output by the multi-source heterogeneous data acquisition and fusion module, combined with the athlete's individual ability model and team tactical model in the historical training database, using a built-in optimization algorithm. It then generates an optimized training plan version and submits it to the training task collaborative scheduling and distribution module. An active abnormal state monitoring and intervention module continuously monitors key performance indicators and physiological safety thresholds in the real-time training data stream. When a data pattern deviates from a preset normal fluctuation range or a preset warning rule is triggered, it automatically generates warning information of different levels and initiates corresponding intervention plans. These intervention plans include suspending training tasks, medical personnel intervention, or emergency adjustments to the training plan.

2. The efficient and collaborative sports training management system according to claim 1, characterized in that, The multi-source heterogeneous data acquisition and fusion module specifically includes a data interface adaptation submodule, a spatiotemporal alignment submodule, and a feature extraction submodule. The data interface adaptation submodule is configured to support multiple communication protocols, including Bluetooth 5.0, Wi-Fi 6, and ZigBee 3.0, for receiving raw data from heart rate monitors, electromyography sensors, inertial measurement units, GPS receivers, and high-speed cameras. The spatiotemporal alignment submodule employs a clock synchronization mechanism based on the Network Time Protocol (NTP) to apply a unified timestamp to all input data streams and uses a Kalman filter algorithm to correct the position coordinates of sensor data exhibiting spatial drift. The feature extraction submodule performs feature calculations for data of different modalities, extracting real-time estimates of heart rate variability and maximum oxygen uptake from physiological index data, extracting 3D coordinate sequences of joint angles, angular velocities, and motion trajectories from motion posture data, and extracting key skeletal point sequences of athletes from video image data using a convolutional neural network.

3. The efficient and collaborative sports training management system according to claim 1, characterized in that, The training task collaborative scheduling and distribution module integrates a task dependency graph and a resource conflict detector. The task dependency graph stores the temporal dependencies and logical prerequisites between atomic training tasks in a directed acyclic graph structure, ensuring that the task distribution order meets the training process constraints. Before distributing a task, the resource conflict detector checks the availability of the venue, equipment, or medical support resources requested by the task within the current time window. If a conflict is detected, it automatically triggers the rescheduling of the task execution time or sends a resource coordination request to the coach terminal.

4. The efficient and collaborative sports training management system according to claim 1, characterized in that, The core optimization model of the adaptive training plan dynamic optimization engine is a deep reinforcement learning network based on the actor-critic framework; The network takes the athlete's real-time physiological load index, technical movement completion score, and ambient temperature and humidity as state inputs, and takes adjustments to training intensity, training content combination, and rest interval as movement outputs. Its objective function is... The aim is to maximize long-term training benefits, where R(s) t ,a t ) is an immediate reward function that comprehensively considers the improvement of training effect and the risk of fatigue accumulation, and γ is a discount factor with a value of 0.

95.

5. The efficient and collaborative sports training management system according to claim 1, characterized in that, The proactive anomaly monitoring and intervention module employs a dual early warning mechanism combining an unsupervised anomaly detection algorithm based on isolated forests with rule-based logical judgment. The isolated forest algorithm rapidly calculates anomaly scores on high-dimensional training data streams, identifying data points that significantly deviate from historical normal patterns. The rule-based logical judgment predefines a set of judgment conditions for sports injuries, over-fatigue, and environmental risks. When real-time data simultaneously meets the condition that the isolated forest anomaly score exceeds the threshold of 0.65 and triggers at least one rule judgment condition, the module determines it as a high-risk event and initiates the highest-level intervention plan.

6. The efficient and collaborative sports training management system according to claim 2, characterized in that, The feature extraction submodule performs frequency domain analysis on the electrocardiogram signal to extract the heart rate variability power spectral density value in the 0.04 to 0.15 Hz frequency band; it uses a quaternion solution algorithm on the inertial measurement unit data to output the real-time three-dimensional Euler angle sequence of the athlete's shoulder, elbow, hip, knee and ankle joints; and it applies a skeleton recognition model based on convolutional neural networks to the high-speed camera video stream to extract the pixel coordinate sequence and motion velocity vector of 25 key body points of the athlete.

7. The efficient and collaborative sports training management system according to claim 3, characterized in that, When the resource conflict detector detects a venue resource conflict, it automatically adjusts the training task to an available venue and sends a venue change notification to the coach's terminal; when it detects an equipment resource conflict, it automatically triggers a rescheduling of the task execution time or sends a resource coordination request to the equipment management terminal.

8. The efficient and collaborative sports training management system according to claim 4, characterized in that, The input state vector of the deep reinforcement learning network includes the average heart rate, maximum oxygen uptake utilization, technical movement deviation value, step frequency variation coefficient, and environmental temperature and humidity comprehensive influence factors of the current training group; the output movement vector is the intensity adjustment coefficient, interval time correction value, and key technical movement reminders for the next training group.

9. A highly efficient and collaborative sports training management method, characterized in that, The method includes the following steps: Step S110: Simultaneously collect multimodal raw data of athletes during training by deploying multiple sensors and monitoring devices. The multimodal raw data includes at least physiological electrical signals, three-dimensional kinematic data, environmental parameters, and high-definition video streams. Step S120: The collected multimodal raw data is preprocessed and fused in real time. The preprocessing includes data noise reduction, outlier removal and missing value imputation. The fusion includes data stream splicing based on a unified time reference and multi-source feature association based on motion scene to generate a consistent fused data package. Step S130: Based on the received coach instructions, team analysis input, and medical assessment information, and combined with the real-time training status reflected by the fusion data packet, a series of atomic training tasks with execution sequences and resource constraints are generated through a task decomposition algorithm, and the atomic training tasks are distributed to the designated execution terminals. Step S140: Using the embedded deep reinforcement learning optimization model, perform online performance evaluation on the currently executing training plan, and dynamically calculate the adjustment parameters of training intensity, training content ratio or rest period based on the evaluation results and preset optimization objectives, and generate adjustment instructions. Step S150: Continuously monitor the key indicators in the fused data packet. When the anomaly detection algorithm identifies data characteristics that conform to the preset risk mode, automatically trigger and execute the corresponding intervention process. The intervention process includes sending an alarm to the relevant terminal, pausing the current training task, or calling the backup training scheme.

10. The efficient and collaborative sports training management method according to claim 9, characterized in that, In step S140, the deep reinforcement learning optimization model adopts a network structure based on the actor-critic framework, and its objective function is: Where R(s) t ,a t ) is the immediate reward function, γ is the discount factor with a value of 0.95; when the network detects that the athlete's heart rate recovery rate drops by 15% and the technical movement deviation is greater than 3 degrees for two consecutive training sessions, it automatically generates a training plan adjustment instruction to reduce the number of remaining training sessions and extend the rest time between sessions.

Images