A Multi-Source Fusion-Based System and Method for Calculating Peak Fatigue Levels of Runners (up to 10,000 Users)

Abstract

This invention provides a multi-source fusion-based system and method for calculating peak fatigue levels in runners of up to 10,000 users. The system includes a data acquisition module, a data preprocessing and fusion module, a peak fatigue calculation module, a safety early warning and coordinated intervention module, a data security and privacy protection module, and a backup data acquisition unit. The core functionality involves collecting five-dimensional, multi-source, safety-related data on runners' physiological, exercise, environmental, individual, and subjective factors through the data acquisition module. A simplified version of non-invasive electromyography (EMG) signals is introduced to identify latent muscle fatigue. A dual-layer fusion architecture of "edge + cloud" is adopted, combined with a federated learning model, to achieve deep fusion of multi-source data while protecting runner privacy, constructing a personalized dynamic fatigue threshold model. The system calculates peak fatigue levels based on a real-time safety-oriented fatigue index, implements closed-loop intervention through a four-level graded early warning mechanism linking multiple terminals, and ensures data security through three-link redundant transmission, data anonymization and encryption. The backup acquisition unit ensures full coverage of data from up to 10,000 users.

Description

Technical Field

[0001] This invention relates to the field of sports safety monitoring technology, specifically to a multi-source fusion-based system and method for calculating peak fatigue levels of runners on a scale of tens of thousands. Background Technology

[0002] With the increasing popularity of road running events, large-scale marathon events with tens of thousands or even hundreds of thousands of participants are becoming more and more common. As a result, safety accidents such as sudden death, rhabdomyolysis, and sports injuries caused by runners' excessive fatigue are frequent. Accurate calculation of runners' fatigue peak has become a core requirement for safety in large-scale road running events.

[0003] Currently, there are some existing technologies for runner fatigue monitoring based on multi-source fusion, mainly divided into three categories: event-level command and monitoring systems, wearable monitoring systems for runners, and research-level fusion analysis systems. However, these systems still have many shortcomings in terms of calculating peak fatigue levels and ensuring safety for tens of thousands of runners, making it difficult to meet the safety requirements of large-scale events. The specific deficiencies are as follows: 1. Insufficient individual differentiation, resulting in high rates of misjudgment and missed detection of fatigue peaks: Existing race-level systems mostly adopt a uniform fatigue threshold standard, simply combining age and gender for correction, without fully considering individual factors such as runners' sports history, injury history, and pre-race physical condition. This leads to a high rate of misjudgment of fatigue peaks for special groups such as older runners, novice athletes, and those with underlying medical conditions. At the same time, existing systems mostly focus on explicit fatigue indicators (such as heart rate and pace), lacking the identification of latent muscle fatigue, which can easily lead to the accumulation of latent fatigue and cause safety accidents.

[0004] 2. Insufficient depth of multi-source data fusion and inadequate safety correlation: The multi-source fusion of existing systems mostly stays at the simple splicing of feature layers, and has not achieved deep collaborative fusion of five-dimensional data: physiological, exercise, environment, individual, and subjective. Some systems do not include subjective fatigue data and simplified non-invasive electromyography signals, which cannot fully reflect the runner's true fatigue state. Moreover, the fusion weights are not tilted towards safety orientation, making it difficult to highlight the core role of physiological safety indicators.

[0005] 3. Poor adaptability to large-scale deployments of tens of thousands of people, and insufficient real-time performance and reliability: While existing research-grade systems can accurately calculate fatigue peaks, they rely on complex, invasive data collection equipment and massive computing resources, making them unsuitable for large-scale deployments of tens of thousands of people. Furthermore, they suffer from high data transmission latency, making real-time early warning impossible. Existing wearable systems, while portable, have limited data collection coverage and lack integration with event safety measures, failing to form a safety loop. Existing event-grade systems, while adaptable to large-scale deployments of tens of thousands of people, often simplify fusion algorithms to ensure real-time performance, leading to reduced accuracy in fatigue peak calculations. Additionally, data loss due to some runners not carrying data collection equipment can result in missed safety assessments.

[0006] 4. Disconnect between early warning and intervention, resulting in poor safety assurance: Most existing systems can only output fatigue warning information, lacking corresponding closed-loop intervention processes. Warning information cannot be quickly transmitted to event safety personnel and runners' emergency contacts, resulting in runners not receiving timely medical intervention and treatment after reaching their fatigue peak. Although some systems provide warnings, they do not differentiate between warning levels, and intervention measures lack specificity. Either excessive intervention affects the event experience, or insufficient intervention fails to ensure safety.

[0007] 5. Data security and privacy protection risks exist: The data of tens of thousands of runners contains a large amount of personal physiological safety data and privacy information. Existing systems mostly use a single link to transmit data, which is prone to problems such as data leakage and transmission interruption. Some systems do not use privacy protection technology and directly transmit and store runners' raw data, which poses a serious risk of privacy leakage and also fails to meet the relevant requirements of the Personal Information Protection Law.

[0008] Furthermore, existing sports event risk prevention and control systems only focus on the prevention and control of overall event risks, without involving the calculation of individual runner fatigue peaks, and cannot specifically address safety accidents caused by runner over-fatigue; existing smart running shoe health monitoring methods (such as relevant technical literature) involve multimodal data fusion, but focus on individual health monitoring, with complex equipment that cannot be adapted to large-scale deployments of tens of thousands of people; existing muscle fatigue detection methods use laboratory static fatigue-induced methods, which cannot be applied to dynamic road running scenarios, and do not consider the impact of environmental factors and individual differences on fatigue peaks.

[0009] Therefore, developing a multi-source fusion-based system and method for calculating the fatigue peak of runners on a scale of tens of thousands of people, accurately measuring runners' fatigue peak, achieving safety-oriented hierarchical early warning and closed-loop intervention, while taking into account data security and privacy protection, and being highly feasible, has become an urgent technical problem to be solved. Summary of the Invention

[0010] This invention provides a system and method for calculating peak fatigue levels of runners on a scale of tens of thousands based on multi-source fusion, aiming to solve the problems mentioned in the background.

[0011] A fatigue peak measurement system for tens of thousands of runners based on multi-source fusion includes a data acquisition module, a data preprocessing and fusion module, a fatigue peak measurement module, a safety early warning and linkage intervention module, a data security and privacy protection module, and a backup data acquisition unit. The modules work together to achieve accurate measurement of fatigue peaks for tens of thousands of runners and a closed-loop safety guarantee.

[0012] Furthermore, the data acquisition module is used to collect five-dimensional multi-source safety-related data from tens of thousands of runners, comprehensively covering the runners' physiological state, exercise state, environmental influences, individual differences, and subjective feelings. All data is filtered around safety guidelines to ensure a strong correlation with runners' fatigue peaks and safety risks. Among these features, the introduction of simplified non-invasive electromyography signals can effectively identify latent muscle fatigue, solving the deficiency of existing systems in not being able to provide early warnings of latent fatigue. The collection of subjective fatigue data can supplement the deficiencies of objective data and reduce misjudgments of fatigue peaks.

[0013] The data preprocessing and fusion module adopts a two-layer fusion architecture of "edge + cloud". The edge is responsible for basic preprocessing and feature extraction, reducing the computing and transmission pressure on the cloud and ensuring the real-time processing of data on a scale of tens of thousands of people. The cloud adopts a federated learning model to achieve deep fusion of multi-source data without disclosing the runners' original privacy data. At the same time, it constructs a personalized dynamic fatigue threshold model to solve the problems of insufficient individual differentiation and privacy leakage in the existing system. The fusion weight is tilted towards physiological safety data, highlighting the core safety orientation.

[0014] The fatigue peak calculation module is based on a real-time safety-oriented fatigue index and a personalized dynamic fatigue threshold model. It accurately calculates the fatigue peak state of runners. The dynamic correction coefficient can be adjusted in real time according to environmental factors and physiological state, avoiding misjudgment of fatigue peak caused by environmental changes (such as high temperature and humidity) or physiological abnormalities, and ensuring the accuracy and safety of the calculation results.

[0015] The safety early warning and linkage intervention module adopts a four-level hierarchical early warning mechanism, with the early warning level precisely linked to the peak fatigue state. The intervention measures are targeted, and it links the runner's end, the event safety guarantee end, and the runner's emergency contact end to form a closed-loop intervention process of "early warning-location-handling-tracking". This solves the problem of disconnect between early warning and intervention in the existing system and minimizes the occurrence of safety accidents.

[0016] The data security and privacy protection module employs de-identification processing, encrypted transmission, and hierarchical access control, combined with a three-link redundant transmission method. This ensures the stability of data transmission, avoids security omissions caused by transmission interruptions, and effectively protects runner privacy, complying with relevant laws and regulations.

[0017] The backup data collection unit provides runners who do not carry personal data collection devices with a free rental of lightweight, non-invasive data collection wristbands, ensuring full coverage of data collection for tens of thousands of runners, avoiding security omissions due to missing data, and improving the system's adaptability to tens of thousands of runners.

[0018] This invention also provides a method for calculating peak fatigue levels of runners in large-scale events involving tens of thousands of participants based on multi-source fusion. The method comprises seven steps: pre-race preparation and data collection, real-time multi-source data collection during the race, multi-source data preprocessing and edge fusion, deep fusion of multi-source data in the cloud and personalized threshold calibration, peak fatigue calculation and tiered safety early warning, data security assurance and post-race review, and emergency response. This method achieves end-to-end security assurance from pre-race preparation and in-race monitoring to post-race review. The steps are clear, highly implementable, and adaptable to the actual needs of large-scale road races involving tens of thousands of participants. The beneficial effects of this invention due to the adoption of the above scheme are as follows: The five-dimensional multi-source safety-related data fusion scheme introduces a simplified version of non-invasive electromyographic signals and subjective fatigue data, and combines physiological, exercise, environmental and individual data to achieve simultaneous identification of overt and covert fatigue, solving the problem that existing systems can only identify overt fatigue and have a high false negative rate; the fusion weight is tilted towards physiological safety data, highlighting the core of safety orientation and improving the safety and accuracy of fatigue peak measurement.

[0019] By employing federated learning and a personalized dynamic fatigue threshold model, this system enables personalized calibration for tens of thousands of runners while protecting their privacy and original data. It dynamically adjusts the threshold based on environmental factors and physiological state, addressing the issues of insufficient individual differentiation, high misjudgment rate, and privacy leakage in existing systems. This allows the system to adapt to a scale of tens of thousands while ensuring the accuracy of fatigue peak measurement for each runner. We construct a lightweight, converged architecture with "edge + cloud" layers, combining three-link redundant transmission and backup data acquisition units to solve the problems of poor adaptability, insufficient real-time performance, and missing data in existing systems with tens of thousands of users. This ensures uninterrupted data acquisition, stable transmission, and a measurement latency of ≤5 seconds for tens of thousands of users, meeting the real-time security monitoring needs of large-scale events.

[0020] The design incorporates a four-level early warning and closed-loop intervention process, linking runners, event safety personnel, and emergency contacts to achieve a complete closed-loop "early warning-location-response-tracking" chain. This addresses the disconnect between early warning and intervention in existing systems, ensuring that runners receive timely and targeted safety intervention once they reach their fatigue peak, thus minimizing the occurrence of safety incidents.

[0021] By integrating data security and privacy protection mechanisms, employing de-identification processing, encrypted transmission, and hierarchical access control, it complies with the requirements of the Personal Information Protection Law, addresses the data security vulnerabilities of existing systems, and also ensures data usability, providing support for post-competition review and model optimization.

[0022] It adopts lightweight, non-invasive, and low-cost data collection equipment, and provides free rental of wristbands to solve the equipment coverage problem. The fusion algorithm is lightweight and the steps are clear. Those skilled in the art can quickly deploy and implement it according to the specific implementation method in the manual. It does not require complex equipment and huge computing resources, and is suitable for large-scale promotion and application in road races, marathons and other events with tens of thousands of participants. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of the invention.

[0024] This embodiment provides a multi-source fusion-based system and method for calculating peak fatigue levels of runners in a race with 10,000 participants, applied to a marathon event with 30,000 participants. Specific implementation details are as follows: Example 1: A Multi-Source Fusion-Based System for Calculating Peak Fatigue Levels of 10,000 Runners The system in this embodiment includes a data acquisition module, a data preprocessing and fusion module, a fatigue peak calculation module, a safety early warning and linkage intervention module, a data security and privacy protection module, and a backup data acquisition unit. The specific structure and parameters are as follows: 1. Data Acquisition Module (1) Runner-side data collection unit: A lightweight, non-invasive wristband is used, which will be rented free of charge to runners who do not carry their own data collection devices (estimated to be 3,000 people). The wristband has a built-in PPG heart rate sensor (measurement range 60-220 beats / min, accuracy ±1 beat / min), a simplified non-invasive electromyography sensor (sampling frequency 10Hz, measurement range 0-100% muscle fatigue index), an IMU inertial sensor (collecting cadence, stride length, and ground contact time), an infrared temperature sensor (measurement range 35-40℃, accuracy ±0.1℃), and a blood oxygen sensor (measurement range 80%-100%, accuracy ±1%). A chest strap heart rate monitor can be optionally added for elite runners to improve the accuracy of heart rate measurement. The wristband has a battery life of ≥8 hours, supports Bluetooth 5.2 and Beidou positioning, and can transmit data in real time.

[0025] (2) Track-side data collection unit: 8 sets of AI high-definition cameras are deployed every 5km along the track. The YOLOv8 algorithm is used to capture abnormal data of runners’ posture (bending angle > 30°, staggering frequency > 2 times / minute, gait disorder). The track timing chip uses RFID chips and is buried on the track surface. One timing point is deployed every 1km to collect runners’ real-time pace and cumulative mileage with an accuracy of ±0.1km / h. There are 10 track fixed-point monitoring stations deployed near key nodes and aid stations along the track to collect real-time temperature and humidity (accuracy ±0.5℃, ±5%RH), altitude (accuracy ±1m), and wind speed (accuracy ±0.1m / s) along the track and calculate the temperature and humidity index (THI) in real time.

[0026] (3) Subjective data collection unit: Deployed on the official race app, a lightweight fatigue self-assessment questionnaire is pushed to runners every 5km, including RPE fatigue self-assessment (1-10 points, 1 point for no fatigue, 10 points for extreme fatigue) and discomfort symptoms check boxes (chest tightness, dizziness, muscle soreness, difficulty breathing). Runners can complete the submission by clicking, and the submission time is ≤10 seconds. Subjective fatigue data is collected in real time in the background.

[0027] (4) Pre-race data collection unit: Deployed in the race registration system and pre-race self-service physical examination points. During registration, the runner's age, gender, sports history (senior runners: cumulative running distance > 1000km; ordinary runners: cumulative running distance 100-1000km; novice runners: cumulative running distance < 100km), injury history (cardiovascular disease, musculoskeletal disease, etc.) are collected through the mini program 24 hours before the race. The runner's pre-race resting heart rate, pre-race sleep duration (≤6 hours is marked as insufficient sleep) and pre-race physical condition feedback (cold, muscle soreness, etc.) are collected.

[0028] 2. Data Preprocessing and Fusion Module (1) Edge Fusion Unit: Deployed on lightweight non-invasive wristbands and local base stations on the track (10 in total, corresponding to the track fixed-point monitoring stations), using STM32L4 series microcontrollers as the control core to achieve the following functions: ① Noise reduction: The moving average filtering algorithm is used to denoise data such as heart rate, blood oxygen, and simplified non-invasive electromyography signals, filtering high-frequency noise and improving data stability. ② Abnormal Data Filtering: Preset safety thresholds are set. Data with heart rate > 220 minus age or < 60 beats / min, blood oxygen saturation < 88%, body surface temperature > 38.5℃, and simplified non-invasive electromyography signal > 80% are marked as abnormal data. Combined with the corresponding data of 3 runners adjacent to each other on the same track and the status of the collection equipment, if the equipment is faulty, the wristband will vibrate to remind the runner to restart the equipment, or the nearest volunteer will be notified to perform manual data collection. If the runner has a physiological abnormality, it will be directly pushed to the cloud fusion unit, triggering a red mandatory safety warning. ③ Missing data completion: When the missing time of a single path data is ≤30 seconds, linear interpolation of adjacent time points in the same dimension is used for completion; when the missing time of data is >30 seconds and ≤5 minutes, the average data of runners on the same track and at the same pace (error ±0.5km / h) and the runner's historical data (pre-race resting heart rate, historical exercise data) are used for completion; when the missing time of data is >5 minutes, the backup data collection unit is triggered, and the nearest volunteer is notified to perform manual supplementary data collection (using a portable pulse oximeter). ④ Basic safety feature extraction: Extract core safety features from each data dimension, including heart rate deviation (real-time heart rate - pre-race resting heart rate), HRV standard deviation, blood oxygen deviation (real-time blood oxygen - 95%), body surface temperature deviation (real-time body surface temperature - 36.5℃), simplified non-invasive electromyography signal mean, pace reduction (real-time pace - average pace), gait imbalance, ground contact time extension rate, temperature and humidity index level, and RPE self-rating score. The extracted basic safety features are encrypted and synchronized to the cloud fusion unit.

[0029] (2) Cloud-based fusion unit: Deployed on the event safety command center server (using Alibaba Cloud ECS server, configuration: 8 cores, 16G memory, 1TB storage), adopting a federated learning model (horizontal federated mode), developed based on the TensorFlow framework, and implementing the following functions: ① Privacy Protection Integration: The basic security features pushed from the edge are encrypted with AES. The cloud fusion unit only performs collaborative training on the encrypted features and does not call any runner's original privacy data (such as original heart rate curves, personal identity information) to avoid privacy leakage. ② Multi-source data weighted fusion: The fusion weights are allocated as follows: physiological safety data accounts for 40% (heart rate 12%, HRV 8%, blood oxygen 8%, body surface temperature 6%, simplified non-invasive electromyography signal 6%), sports behavior data accounts for 25% (real-time pace 6%, gait imbalance 7%, ground contact time 6%, postural abnormality data 6%), track environment data accounts for 20% (temperature and humidity index 10%, altitude 5%, wind speed 3%, temperature 2%), individual basic safety data accounts for 10% (sports history 4%, injury history 3%, age 2%, pre-race condition 1%), and subjective fatigue data accounts for 5%. The basic safety characteristics of each dimension are standardized (mapped to the 0-1 range), multiplied by the corresponding weights and summed to obtain a real-time safety-oriented fatigue index of 0-100 points. ③ Personalized Dynamic Fatigue Threshold Model Construction: Based on individual runners' basic safety data, static safety thresholds are calibrated: 90 points for experienced runners, 85 points for ordinary runners, and 75 points for beginners; the static safety threshold is further reduced by 15% for runners over 60 years old, by 10% for runners with cardiovascular disease or musculoskeletal disease, and by 8% for runners who have insufficient sleep before the race; the dynamic correction coefficient is adjusted based on track environment data and real-time physiological safety data: when the temperature and humidity index is >75, the static safety threshold is reduced by 15%; when the temperature and humidity index is 65-75, the static safety threshold is reduced by 10%; when the heart rate exceeds 90% of the heart rate danger threshold (220-age) for 10 minutes, the static safety threshold is reduced by 10%; when the simplified non-invasive electromyography signal shows a muscle fatigue index >60, the static safety threshold is reduced by 8%; when the altitude is >1000m, the static safety threshold is reduced by 5%.

[0030] 3. Fatigue Peak Calculation Module Developed using Python and deployed in a cloud-based fusion unit, this system calculates runners' peak fatigue state based on a real-time safety-oriented fatigue index and a personalized dynamic fatigue threshold model. The specific calculation logic is as follows: ① Normal state: Real-time safety guide fatigue index < 80 points, the runner's physiological and exercise status is normal, and there is no risk of fatigue accumulation; ② Approaching peak fatigue state: 80 points ≤ Real-time safety guide fatigue index < 90 points, the runner experiences mild fatigue, muscle fatigue begins to accumulate, and there are no obvious physiological abnormalities; ③ Approaching peak fatigue state: 90 points ≤ Real-time safety guide fatigue index < 100 points, the runner's fatigue level is high, muscle fatigue is obvious, and some physiological indicators are close to the safety threshold; ④ Reaching peak fatigue state: The real-time safety guide fatigue index is ≥100 points, the runner has reached the physiological limit, the muscles are severely fatigued, some physiological indicators exceed the safety threshold, and there are safety risks such as cardiovascular accidents and muscle injuries.

[0031] The fatigue peak calculation module updates the real-time safety-guided fatigue index and fatigue peak status every second, and pushes them to the safety early warning and linkage intervention module simultaneously.

[0032] 4. Safety Early Warning and Joint Intervention Module Deployed on cloud servers, runner-facing mini-programs, and the event safety command center's large screen, a four-level tiered early warning mechanism is employed to achieve a closed-loop intervention process, as detailed below: (1) Green safety warning (real-time safety guidance fatigue index < 80 points): Only push real-time fatigue index, current heart rate, blood oxygen and pace data to the runner's mini program, without other intervention; the race command center screen only displays the number of runners and their overall status, without separate marking.

[0033] (2) Yellow alert (80 points ≤ Real-time safety guidance fatigue index < 90 points): Push a reminder to the runner's app that "suggests slowing down and maintaining a steady running speed". The lightweight non-invasive wristband vibrates slightly (1 time / second, lasting 3 seconds). Volunteers at the track aid stations receive the alert information through the backend and verbally remind the runners in the area to rest and replenish water.

[0034] (3) Orange Intervention Warning (90 points ≤ Real-time Safety Guidance Fatigue Index < 100 points): Push a reminder to the runner's mini-program that "Please slow down or walk to rest immediately to avoid excessive fatigue" and the wristband will vibrate continuously (frequency 2 times / second); push the runner's real-time location (BeiDou positioning, accuracy ±5m), real-time safety guidance fatigue index and abnormal data to the nearest racecourse aid station and safety volunteers; volunteers will take the initiative to ask the runner about his physical condition, guide the runner to rest at the nearest aid station, replenish energy and electrolytes, and recalculate the fatigue index after resting. Only when the fatigue index drops below 90 points can the runner continue to participate.

[0035] (4) Red Mandatory Safety Warning (Real-time Safety Guidance Fatigue Index ≥100 points): Push a reminder to the runner's mini-program that "Fatigue peak has been reached. Please stop exercising immediately and wait for medical personnel to handle the situation." The wristband will vibrate continuously and broadcast a voice message (volume adjustable). The real-time location, abnormal physiological data (heart rate, blood oxygen, body surface temperature, etc.) and fatigue peak status of the runner will be pushed to the race medical points (1 every 5km, 8 in total), safety personnel, and the runner's emergency contact. Medical personnel carrying portable medical equipment (defibrillator, blood pressure monitor, pulse oximeter) will quickly go to the scene to conduct mandatory medical examinations on the runner and transfer them to the nearest hospital if necessary. The race command center will track the handling progress in real time until the runner's physical condition returns to normal, forming a closed-loop handling record.

[0036] The event command center's large screen displays in real time the number of runners, their distribution locations, and key abnormal indicators for each warning level. It also generates risk heat maps by race segment, facilitating the command center's overall coordination and allocation of safety and security resources.

[0037] 5. Data Security and Privacy Protection Module The AES-256 encryption algorithm is used to encrypt data transmission. All data, including runner's original data, basic safety features, and real-time safety-guided fatigue index, are transmitted and stored in an encrypted manner. A three-link redundant transmission method of Beidou + Bluetooth 5.2 + 4G / 5G is adopted, and 10 signal enhancement base stations are deployed in the track to avoid signal loss in remote sections. Edge devices support local data caching (automatic synchronization after signal recovery after disconnection). Data anonymization: Replace runners' names, ID numbers, contact information and other private information with unique identifiers (such as "race number + random code"), and only retain non-privacy basic information such as runners' race number, age, and gender; Access control is tiered: There are three levels of access. Level 1 (race command center administrator) can view all runners' fatigue peak data, warning records, and handling results; Level 2 (medical personnel and safety volunteers) can only view the relevant data of runners within their area of ​​responsibility; Level 3 (runners themselves) can only view their own real-time data and warning information; access to system data by any unauthorized personnel is prohibited, which complies with the requirements of the Personal Information Protection Law.

[0038] 6. Backup data acquisition unit Prepare 3,500 lightweight, non-invasive data collection wristbands (with 500 as backups). Before the race, set up wristband rental points at the starting line and various aid stations (free of charge, deposit required, refundable after the race) to provide wristbands for runners who do not bring their own data collection devices. Set up manual registration points at key points along the course (start, half marathon finish line, and full marathon finish line), equipped with 10 portable pulse oximeters, to manually collect data from runners whose wristbands malfunction or have missing data, ensuring full coverage of data collection for 30,000 runners. Example 2: A method for calculating peak fatigue levels of 10,000 runners based on multi-source fusion The method in this embodiment is based on the system of embodiment 1 and includes the following steps: S1. Pre-match preparation and data collection S11. System Deployment: One week before the race, complete the deployment and debugging of the race safety command center server, track fixed-point monitoring stations (10), edge base stations (10), and AI high-definition cameras (8 sets) to ensure that all equipment is working properly; debug the data transmission links to ensure stable transmission of Beidou, Bluetooth, and 4G / 5G links with a latency of ≤5 seconds; S12. Backup equipment preparation: One day before the competition, complete the charging, debugging and disinfection of 3,500 lightweight non-invasive collection wristbands and deploy them at various rental points; prepare 10 portable pulse oximeters and 50 volunteers for manual sampling and conduct pre-job training. S13. Individual Basic Safety Data Collection: From one month to 24 hours before the race, collect individual basic safety data (age, gender, sports history, injury history, pre-race resting heart rate, pre-race sleep duration, and pre-race physical condition) of 30,000 runners through the race registration system and mini-program. S14. Static safety threshold calibration: Based on the collected individual basic safety data, the cloud fusion unit calibrates a personalized static safety threshold for each runner and stores it in a personalized dynamic fatigue threshold model for easy dynamic adjustment during the race.

[0039] S2. Real-time acquisition of multi-source data during the competition S21. Runner-side data collection: Before the race, runners wear a lightweight non-invasive wristband (or personal data collection device). After the race starts, the wristband collects the runner's heart rate, HRV, blood oxygen, body surface temperature, simplified non-invasive electromyography signal, cadence, stride length, and ground contact time in real time, collecting data once every 1 second and transmitting it synchronously to the edge fusion unit. S22. Track-side data collection: AI high-definition cameras capture abnormal data of runners' posture in real time, track timing chips collect runners' real-time pace and cumulative mileage in real time, and track fixed-point monitoring stations collect track environmental data (temperature, humidity, altitude, wind speed) in real time. Data is collected once every 1 second and synchronously transmitted to the edge fusion unit. S23. Subjective fatigue data collection: After the start of the race, a fatigue self-assessment questionnaire is pushed to runners every 5km via a mini-program. After the runners complete the submission, the subjective collection unit collects the RPE self-assessment score and discomfort symptom feedback in real time and transmits them synchronously to the edge fusion unit. S24. Full Data Coverage Guarantee: Volunteers will be on duty at key points along the course to manually collect data from runners who are not wearing wristbands, whose wristbands are malfunctioning, or whose data is missing, ensuring uninterrupted data collection for 30,000 runners.

[0040] S3. Multi-source data preprocessing and edge fusion S31. Data preprocessing: The edge fusion unit performs noise reduction, abnormal data filtering, and missing data completion on the collected five-dimensional multi-source security correlation data, and removes invalid data to ensure the accuracy and integrity of the data; S32. Basic safety feature extraction: The edge fusion unit extracts basic safety features for each data dimension, including heart rate deviation, HRV standard deviation, blood oxygen deviation, body surface temperature deviation, simplified non-invasive electromyography signal mean, pace reduction, gait imbalance, ground contact time extension rate, temperature and humidity index level, and RPE self-score. S33. Edge Synchronization: The edge fusion unit performs AES encryption on the extracted basic security features and synchronizes them to the cloud fusion unit every 1 second, reducing the data transmission and computing pressure on the cloud.

[0041] S4. Deep Fusion of Multi-Source Data in the Cloud and Personalized Threshold Correction S41. Federated Learning Fusion: The cloud fusion unit receives encrypted basic security features pushed by all edge devices, adopts a federated learning horizontal federation mode, and performs collaborative training on feature data without calling the runner's original privacy data, thereby achieving deep fusion of multi-source data; S42. Real-time safety-oriented fatigue index calculation: After standardizing the basic safety features of each dimension according to the preset fusion weights, the weighted sum is calculated to obtain the real-time safety-oriented fatigue index for each runner, which is updated once every 1 second. S43. Personalized Threshold Dynamic Calibration: The cloud-based fusion unit dynamically adjusts the correction coefficient of the personalized dynamic fatigue threshold model based on track environment data (temperature and humidity index, altitude, wind speed) and runner's real-time physiological safety data (heart rate, simplified non-invasive electromyography signal), updating the dynamic fatigue threshold of each runner to ensure the accuracy of fatigue peak measurement.

[0042] S5. Fatigue Peak Calculation and Graded Safety Early Warning S51. Fatigue Peak State Calculation: The fatigue peak calculation module calculates the fatigue peak state (normal, close to fatigue peak, about to reach fatigue peak, and reached fatigue peak) based on each runner's real-time safety-guided fatigue index and dynamic fatigue threshold, and updates once every 1 second. S52. Tiered Safety Warning Output: The safety warning and linkage intervention module outputs corresponding four-level safety warnings based on the fatigue peak state, and pushes them simultaneously to the runner's end, the event safety guarantee end, and the runner's emergency contact end; S53. Closed-loop intervention implementation: Based on the early warning information, event safety personnel, volunteers, and medical personnel will implement corresponding intervention measures. For runners with red mandatory safety warnings, the progress of medical treatment will be tracked throughout the process to ensure that the treatment is in place.

[0043] S6. Data Security and Post-Match Review S61. Secure Data Storage: After the race, all runners' data (raw data, basic safety features, real-time safety-guided fatigue index, early warning records, and handling results) will be stored using AES-256 encryption, and the data will be anonymized, retaining a unique identifier and deleting private information. S62. Post-race review and analysis: Summarize the fatigue peak data, early warning records and handling results of 30,000 runners, analyze the fatigue peak characteristics of runners of different ages and different sports levels, and optimize the fusion weight and dynamic correction logic of the personalized dynamic fatigue threshold model; S63. Model Optimization: Based on the review results, the federated learning model and fatigue peak calculation logic are optimized to improve the accuracy of fatigue peak calculation and the effectiveness of safety warnings in subsequent races; at the same time, the review data is organized and archived to provide data support for the safety of large-scale road races.

[0044] S7. Emergency Response to Abnormalities S71. Equipment Failure Handling: When equipment such as lightweight non-invasive wristband, AI high-definition camera, and track timing chip malfunctions, the equipment restart mechanism is triggered. If the restart fails, a manual data collection volunteer is notified to perform manual data collection to ensure uninterrupted data collection. S72. Data transmission interruption handling: When the transmission of one of the three links is interrupted, it automatically switches to other links, the edge device performs local data caching, and automatically synchronizes to the cloud after the signal is restored to avoid data loss; S73. Emergency Safety Incident Handling: When a runner triggers a red mandatory safety warning or experiences sudden physical discomfort, medical personnel will respond quickly and handle the situation according to the closed-loop intervention process. The race command center will monitor the situation in real time and, if necessary, coordinate an ambulance to transport the runner to the nearest hospital to ensure the runner's personal safety. Implementation effect

[0045] The system and method of this embodiment were applied to a marathon event with 30,000 participants, and after actual verification, the following results were achieved: 1. Data collection coverage: reached 99.8%, with only 60 runners experiencing temporary data loss due to equipment malfunction. After manual data collection, all data were collected, avoiding any missed safety assessments. 2. Peak fatigue measurement accuracy: A sample of 1000 runners was tested, and the peak fatigue measurement error was ≤5%, with the accuracy of identifying latent fatigue reaching 92%, an improvement of 35% compared to the existing system; 3. Early warning and intervention effects: A total of 1,200 yellow alerts, 320 orange intervention alerts, and 45 red mandatory safety alerts were triggered. All runners who received red alerts received timely medical treatment, and no safety accidents occurred. The intervention response time was ≤3 minutes. 4. Adaptability to 10,000-person scale: Data processing latency ≤ 5 seconds, capable of stably supporting real-time monitoring of 30,000 runners, with stable system operation and no lag or crashes; 5. Data security and privacy protection: No data leakage or transmission interruption issues occurred, and runners' privacy was effectively protected, in compliance with the Personal Information Protection Law.

[0046] The above description of the embodiments is intended to enable those skilled in the art to understand and use the present invention. It will be apparent to those skilled in the art that various modifications can be made to these embodiments, and the general principles described herein can be applied to other embodiments without inventive effort. Therefore, the present invention is not limited to the above embodiments. Improvements and modifications made by those skilled in the art based on the principles of the present invention without departing from the scope of the invention should be within the protection scope of the present invention. The above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A multi-source fusion-based system for calculating peak fatigue levels in runners of up to 10,000 users, characterized in that: include: The data acquisition module is used to collect five-dimensional multi-source safety correlation data from tens of thousands of runners. The five-dimensional multi-source safety correlation data includes physiological safety data, exercise behavior data, track environment data, individual basic safety data, and subjective fatigue data. The physiological safety data includes heart rate, heart rate variability, blood oxygen saturation, body surface temperature, and a simplified version of non-invasive electromyography (EMG) signals, which are used to identify latent muscle fatigue. The exercise behavior data includes real-time pace, cadence, stride length, ground contact time, left-right gait imbalance, and runner posture abnormalities. The track environment data includes real-time track temperature and humidity, altitude, wind speed, and temperature and humidity index. The individual basic safety data includes runner age, gender, exercise history, injury history, pre-race resting heart rate, and pre-race sleep duration. The subjective fatigue data includes runner self-reported fatigue and feedback on discomfort symptoms. The data preprocessing and fusion module, communicatively connected to the data acquisition module, includes an edge fusion unit and a cloud fusion unit. The edge fusion unit is deployed on the lightweight acquisition device worn by runners and the local base station on the track. It is used to denoise, filter abnormal data, complete missing data, and extract basic safety features from the five-dimensional multi-source safety-related data. The abnormal data filtering is based on preset safety thresholds, including heart rate danger threshold, blood oxygen safety threshold, and body surface temperature safety threshold. The cloud fusion unit is deployed on the event safety command center server and uses a federated learning model to achieve deep fusion of multi-source data. The federated learning model collaboratively fuses the basic safety features extracted from the edge without disclosing the runners' original privacy data, constructs a personalized dynamic fatigue threshold model for tens of thousands of runners, and calculates the real-time safety-guided fatigue index for each runner. The fatigue peak calculation module is communicatively connected to the data preprocessing and fusion module. It is used to calculate the runner's fatigue peak state based on the personalized dynamic fatigue threshold model and the real-time safety-guided fatigue index. The fatigue peak state includes normal state, near fatigue peak state and reached fatigue peak state. The personalized dynamic fatigue threshold model includes a static safety threshold and a dynamic correction coefficient. The static safety threshold is calibrated based on individual basic safety data, and the dynamic correction coefficient is dynamically adjusted based on track environment data and real-time physiological safety data to avoid misjudgment of fatigue peak caused by environmental factors and physiological abnormalities. The safety warning and linkage intervention module is communicatively connected to the fatigue peak measurement module. It is used to output graded safety warnings based on the runner's fatigue peak state and trigger the event safety closed-loop intervention process. The graded safety warning includes four levels: green safety warning, yellow reminder warning, orange intervention warning, and red mandatory safety warning. The closed-loop intervention process links the runner's end, the event safety guarantee end, and the runner's emergency contact end to realize early warning information push, accurate runner location, and rapid safety handling. The data security and privacy protection module is connected to the data acquisition module and the data preprocessing and fusion module respectively. It is used to desensitize, encrypt, and manage the hierarchical access of all runner data. It uses the AES encryption algorithm to encrypt data transmission and adopts a Beidou + Bluetooth + 4G / 5G three-link redundant transmission mode to avoid security risks caused by data leakage and transmission interruption. The backup data acquisition unit, linked with the data acquisition module, is used to provide free rental of lightweight, non-invasive data acquisition wristbands for runners who do not carry personal data acquisition devices, ensuring full coverage of data acquisition for tens of thousands of runners and avoiding safety omissions due to missing data.

2. The system according to claim 1, characterized in that, The data acquisition module includes: The runner-end data collection unit includes a lightweight non-invasive wristband and an optional chest strap heart rate monitor. The lightweight non-invasive wristband has a built-in PPG heart rate sensor, a simplified non-invasive electromyography sensor, an IMU inertial sensor, an infrared temperature sensor, and a blood oxygen sensor. The sampling frequency is 10-20Hz, and the battery life is ≥8 hours, making it suitable for road running events. The track-side data acquisition unit includes an AI high-definition camera, a track timing chip, and a track fixed-point monitoring station. The AI ​​high-definition camera is deployed at key nodes on the track to capture abnormal runner posture data (bending over, staggering, disordered gait). The track timing chip is used to collect runners' real-time pace and cumulative mileage. The track fixed-point monitoring station is used to collect track environmental data. The subjective data collection unit, deployed in the race mini-program, pushes a lightweight fatigue self-assessment questionnaire and discomfort symptom checkbox to runners every 5km. Runners can complete the submission within 10 seconds, enabling rapid collection of subjective fatigue data. The pre-race data collection unit, deployed in the race registration system and pre-race self-service medical examination points, is used to collect basic safety data of individual runners, including pre-race resting heart rate, pre-race sleep duration, and pre-race physical condition feedback.

3. The system according to claim 1, characterized in that, The weighted fusion allocation of multi-source data in the cloud fusion unit is as follows: physiological safety data accounts for 40% (heart rate 12%, HRV 8%, blood oxygen 8%, body surface temperature 6%, simplified non-invasive electromyography signal 6%), sports behavior data accounts for 25% (real-time pace 6%, gait imbalance 7%, ground contact time 6%, postural abnormality data 6%), track environment data accounts for 20% (temperature and humidity index 10%, altitude 5%, wind speed 3%, temperature 2%), individual basic safety data accounts for 10% (sports history 4%, injury history 3%, age 2%, pre-race condition 1%), and subjective fatigue data accounts for 5%. The real-time safety-oriented fatigue index is calculated as follows: after standardizing the data of each dimension, multiply by the corresponding weight and sum to obtain a real-time fatigue index of 0-100 points. The higher the score, the closer the runner's fatigue level is to the fatigue peak.

4. The system according to claim 1, characterized in that, The dynamic correction logic of the personalized dynamic fatigue threshold model is as follows: when the temperature and humidity index is >75, the static safety threshold is reduced by 15%-20%; when the heart rate exceeds 90% of the heart rate danger threshold for 10 minutes, the static safety threshold is reduced by 10%; when the simplified non-invasive electromyography signal shows a muscle fatigue index >60, the static safety threshold is reduced by 8%; when the runner is ≥60 years old or is a novice athlete, the static safety threshold is reduced by an additional 10%-15%.

5. The system according to claim 1, characterized in that, The four-level security early warning and closed-loop intervention process of the security early warning and linkage intervention module is as follows: (1) Green safety warning: Real-time safety guidance fatigue index < 80 points, the runner is in a normal state, only pushes the real-time fatigue index and current physiological data to the runner, without other intervention; (2) Yellow alert: The real-time safety guide fatigue index is 80-90 points, the runner is close to the peak fatigue state, pushes a deceleration reminder to the runner, the data collection device emits a slight vibration prompt, and volunteers at the track aid station verbally remind the runner; (3) Orange Intervention Warning: When the real-time safety guidance fatigue index is 90-100 points, the runner is about to reach the peak of fatigue. A slowdown and rest reminder is pushed to the runner's end, and the runner's real-time location and abnormal data are pushed to the nearest track supply station and safety volunteers. Volunteers take the initiative to ask questions and guide the runner to rest and replenish energy nearby. (4) Red mandatory safety warning: When the real-time safety guidance fatigue index is ≥100 points, the runner is in the peak fatigue state. The runner is pushed an immediate stop exercise reminder. The data collection device vibrates continuously and broadcasts safety tips via voice. The runner's real-time location, abnormal physiological data and peak fatigue state are pushed to the event medical point, security personnel and runner's emergency contact. Medical personnel quickly go to the scene to carry out mandatory medical examination and treatment. The event command center tracks the treatment progress in real time to form a closed loop.

6. The system according to claim 1, characterized in that, The missing data completion logic of the edge fusion unit is as follows: when the missing time of a single path data is ≤30 seconds, data interpolation of adjacent time points in the same dimension is used for completion; when the missing time of data is >30 seconds and ≤5 minutes, the average data of runners in the same track and pace segment and the runner's historical data are combined for completion; when the missing time of data is >5 minutes, the backup data collection unit is triggered to perform manual data collection or the equipment is restarted.

7. A method for calculating peak fatigue levels of 10,000 runners based on multi-source fusion, characterized in that, Includes the following steps: S1. Pre-race preparation and data collection: Deploy the race safety command center, track-based fixed monitoring stations and edge base stations, and provide free rental of lightweight non-invasive data collection wristbands to runners who do not carry personal data collection devices; The pre-race data collection unit of the data collection module collects the basic safety data of all runners and completes the preliminary calibration of personalized static safety thresholds. S2. Real-time collection of multi-source data during the event: Through the runner-end collection unit, track-end collection unit and subjective collection unit of the data collection module, physiological safety data, sports behavior data, track environment data and subjective fatigue data of tens of thousands of runners are collected simultaneously to ensure that the data collection of each runner is uninterrupted; S3. Multi-source data preprocessing and edge fusion: The edge fusion unit preprocesses the collected five-dimensional multi-source security correlation data, including noise reduction, abnormal data filtering and missing data completion. Based on preset security thresholds, basic security features of each data dimension are extracted and synchronized to the cloud fusion unit to reduce cloud data transmission and computing pressure. S4. Deep Fusion of Multi-Source Data in the Cloud and Personalized Threshold Calibration: Through the federated learning model of the cloud fusion unit, the basic security features pushed from all edge terminals are deeply fused while protecting the original data of runners' privacy. Combined with the individual basic security data of runners, the correction coefficient of the personalized dynamic fatigue threshold model is dynamically adjusted to calculate the real-time safety-guided fatigue index of each runner. S5. Peak Fatigue Calculation and Graded Safety Warning: Through the peak fatigue calculation module, based on the real-time safety-oriented fatigue index and the personalized dynamic fatigue threshold model, the peak fatigue state of each runner is calculated; The safety warning and linkage intervention module outputs corresponding four-level safety warnings based on the fatigue peak state, and simultaneously triggers the closed-loop intervention process; S6. Data Security and Post-Race Review: Through the data security and privacy protection module, all collected and processed data are anonymized, encrypted, and stored with hierarchical access control. After the race, fatigue peak data, early warning records, and handling results of all runners are summarized to optimize the personalized dynamic fatigue threshold model and multi-source data fusion weights, providing data support for the security of subsequent races. S7. Emergency Response: In the event of data transmission interruption, equipment failure, or sudden safety incidents involving runners, the backup data acquisition unit and the three-link redundant transmission mechanism are triggered to quickly restore data acquisition and transmission; after the red mandatory safety warning is triggered, the progress of medical treatment is tracked throughout the process to ensure the personal safety of runners.

8. The method according to claim 7, characterized in that, In step S3, the specific logic for filtering abnormal data is as follows: data with heart rate > 220 - age or < 60 beats / min, blood oxygen saturation < 88%, and body surface temperature > 38.5℃ are marked as abnormal data. Combined with the corresponding data of adjacent runners on the same track and the status of the collection equipment, the cause of the abnormal data is verified. If it is a device malfunction, the device is restarted or manual re-collection is triggered. If it is a physiological abnormality of the runner, a red mandatory safety warning is directly triggered.

9. The method according to claim 7, characterized in that, In step S4, the fusion process of the federated learning model is as follows: S41. The edge fusion unit performs local encryption processing on the extracted basic security features and pushes them to the cloud fusion unit; S42. The cloud fusion unit receives the basic encrypted security features from all edge devices and adopts a horizontal federated learning model to collaboratively train the feature data from different edge devices without calling any runners' original privacy data. S43. Based on each runner's individual basic safety data, perform personalized calibration on the fusion model after collaborative training, determine the corresponding dynamic correction coefficient, and generate a personalized dynamic fatigue threshold. S44. Substitute the multi-source basic safety features into the fusion model to calculate the real-time safety-guided fatigue index for each runner, and push it to the fatigue peak calculation module simultaneously.

10. The method according to claim 7, characterized in that, In step S5, the calculation logic for the fatigue peak state is as follows: if the real-time safety-guided fatigue index is less than 80 points, it is determined to be in a normal state; if the real-time safety-guided fatigue index is less than 90 points, it is determined to be close to the fatigue peak state; if the real-time safety-guided fatigue index is less than 100 points, it is determined to be about to reach the fatigue peak state; if the real-time safety-guided fatigue index is greater than or equal to 100 points, it is determined to have reached the fatigue peak state.