Exercise risk assessment system based on big data

Abstract

The invention discloses an exercise risk assessment system based on big data, and relates to the technical field of exercise risk assessment. The exercise risk assessment system based on the big data comprises an exercise data acquisition and integration module, a data extraction and fusion module and a risk assessment module. According to the method, the sports data of the athlete are obtained through the wearable device, big data, training logs and other multi-angle modes, standardization processing is carried out, the method is not limited to manually filled sports information, scientificity is high, the multi-dimensional sports data are fused to form comprehensive features, one-sidedness of subsequent evaluation is avoided, and the method is suitable for popularization and application. And based on the obtained comprehensive characteristic dynamic evaluation, fatigue loss accumulation caused by long-time movement is considered, so that the risk evaluation result is more accurate, and the purpose of dynamically evaluating the strategy is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of sports risk assessment, and in particular to a sports risk assessment system based on big data. Background Art

[0002] Health and exercise venues rely primarily on the personal experience of coaches or instructors to provide health and exercise guidance to users. They lack long-term, systematic planning for users' exercise plans or goals, and their guidance lacks scientificity and consistency. Health and exercise venues also lack centralized management of users' health information. Coaches and instructors primarily focus on body shaping and lack a comprehensive understanding of health knowledge. Consequently, for users with chronic conditions or a history of illness, they are unable to develop scientific exercise plans or provide accurate health and exercise guidance, leading to increased exercise risks.

[0003] For example, a health exercise goal assessment method based on big data is disclosed in Chinese patent publication number CN113496767A. The health exercise guidance for users mainly relies on the information filled in by the users themselves, the users' historical physical examination data and medical records, and the various data before, during and after the users' exercise. The health exercise guidance has long-term and systematic planning for the users' exercise plans or exercise goals. The health exercise guidance is scientific and coherent, and the users' health information is continuously improved and updated. A separately applicable exercise health goal plan is provided for each target user, which can provide correct exercise health guidance and minimize exercise risks.

[0004] The above-mentioned evaluation method mainly relies on static data analysis and manual experience judgment. For example, the adaptive changes in athletes' training loads have a cumulative trend in risks, and it is difficult to quantify the cumulative effects of fatigue injuries. In addition, athletes' healthy exercise guidance mainly relies on information filled in by users themselves, resulting in the final evaluation method being unscientific. Summary of the Invention

[0005] In view of the shortcomings of the existing technology, the present invention provides a sports risk assessment system based on big data, which solves the problems raised in the above background technology.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: a sports risk assessment system based on big data, comprising: the sports data acquisition and integration module; used to obtain athlete sports data and perform standardization processing on the obtained data; Data extraction and fusion module: used to extract motion feature data from the normalized motion data and fuse them to generate a motion comprehensive feature vector; Risk assessment module: performs motion risk assessment based on comprehensive motion feature vectors and verifies accuracy using historical data in big data.

[0007] A further improvement to the technical solution of the present invention is that, in the sports data collection and integration module, the acquisition of athlete sports data includes: acquiring athlete biometric data based on wearable devices, acquiring athlete sports performance data based on big data and IMU inertial sensors, and acquiring behavioral and psychological data based on big data analysis of athlete training logs. Furthermore, the sports data and environmental data specifically include acquiring athlete biometric data based on wearable devices, acquiring athlete sports performance data based on big data and IMU inertial sensors, and acquiring behavioral and psychological data based on big data analysis of athlete training logs.

[0008] Biometric data such as heart rate, body temperature, muscle electrical signals, blood oxygen saturation, etc.; sports performance data such as movement trajectory, impact force and joint load, etc.; behavioral and psychological data such as training duration, intensity, recovery period, etc.

[0009] A further improvement of the technical solution of the present invention is that after obtaining the above-mentioned sports data related to the athletes, the Z-score standardization is performed. , where x is the original value, is the mean value obtained based on big data, Based on the standard deviation obtained from big data, the relevant motion data will eventually be standardized and processed to facilitate subsequent data extraction and risk assessment.

[0010] Furthermore, in order to ensure that the normalized motion data has time series characteristics, the present application further improves the above method. The motion data acquisition and integration module completes the normalization processing based on a dynamic normalization model, wherein the dynamic normalization model includes: ; Where, is the original motion data, such as joint angular velocity, ground reaction force, Normalize the time series data. is the average of the athletes’ historical data, is the standard deviation of historical data obtained based on big data, is the adaptive cycle, To prevent zero constant. Ultimately, individual differences are eliminated and time series characteristics are retained. In the above formula, Added in It is used to avoid the denominator from approaching zero due to insufficient initial data, and is usually set to 0.01. In the theory of super recovery, the human body's adaptation to training load conforms to the S-shaped curve. It is a small cycle division in the sports training cycle. In the explanation of this formula, it is used as an adaptive cycle and is usually taken as 30 days.

[0011] A further improvement of the technical solution of the present invention is that the data extraction and fusion module completes the fusion of multiple motion feature data based on a feature fusion model, wherein the feature fusion model includes: ; Where, is the feature-damage association matrix obtained by graph convolutional network training, The safety threshold can be determined by the threshold given by the International Society of Sports Medicine. is the biomechanical nonlinear coefficient, such as 1.5 for bones and 2.0 for ligaments. is the time series feature weight, It is the temporal anomaly degree extracted by the long short-term memory network. The feature fusion model derives the single parameter risk contribution based on the Hill model, specifically: , where is the actual stress of the moving tissue, is the yield strength of the moving tissue, is the material hardening index, and is 1.5-3.0 for human tissue. The above model is a biomechanical response model. In order to facilitate the integration of human motion data, a human feature space mapping is constructed, specifically: ,in The feature-injury correlation matrix, i.e., the anatomical structure correlation matrix, is used to construct the biomechanical response model layer with the biomechanical relationship of the human motion chain. Finally, the time dimension feature is introduced and LSTM is used to capture the compensatory action pattern, i.e. , and finally fuse the above multiple models to obtain the feature fusion model disclosed in this application.

[0012] A further improvement of the technical solution of the present invention is that the risk assessment module is based on the input of the comprehensive motion feature vector and outputs a real-time risk index. , specifically including the following evaluation functions: ; Where, is the real-time risk index, is the motion comprehensive feature vector, is the injury weight coefficient, reflecting the clinical importance of different injury types. is the fatigue attenuation coefficient, which reflects the impact of historical risks on the current situation. It is the cumulative risk score, reflecting the fatigue damage superposition effect caused by the superposition of exercise time. In this application, the initial risk model of the Logistic regression framework is considered in the exercise risk assessment based on the comprehensive motion feature vector, specifically: , while this risk assessment model is applied to static assessment. In the present technical solution, the assessment of motion analysis is often affected by the time dimension. Therefore, this application introduces the time dimension and uses the integral term to simulate the fatigue characteristics of human tissue under long-term exercise, specifically: Where, is the real-time risk index, is the motion comprehensive feature vector, is the injury weight coefficient, reflecting the clinical importance of different injury types. Fatigue attenuation coefficient, which reflects the degree of influence of exercise time superposition on the current state, is adjusted according to the type of exercise. For example, the fast attenuation of sprinting is 0.02 / s, and the long-term accumulation of marathon is 0.005 / s. It is the cumulative risk score, reflecting the cumulative effect of fatigue damage caused by continuous exercise.

[0013] A further improvement of the technical solution of the present invention is that the integral term of the evaluation function in the risk assessment module also includes: site hardness Temperature and humidity , and through transfer learning, based on the athlete's baseline data, such as joint mobility and maximum oxygen uptake, optimize the fatigue attenuation coefficient and safety thresholds .

[0014] A further improvement of the technical solution of the present invention is that the risk assessment module uses historical data in big data to verify its accuracy, including the following steps: The historical data is divided into training set and test set. K-fold cross validation is used to randomly divide the data into K subsets. K-1 subsets are used to train the model in turn, and the remaining 1 subset is used for testing. After repeating K times, the average value is taken to obtain the verification model.

[0015] A further improvement of the technical solution of the present invention is that the risk assessment module uses historical data in big data to verify accuracy, and further includes the following steps: The high-risk features output by the sports risk assessment are matched with the actual sports injury records in the validation model. The chi-square test is used to verify whether there is a significant difference between the model prediction results and the actual risk distribution. If the model involves probabilistic prediction, the Brier score can be used to evaluate the calibration of the predicted probability.

[0016] Compared with the existing technology, the beneficial effects of the present invention are: obtaining athletes' sports data from multiple angles through wearable devices, big data, training logs, etc., and performing standardized processing. It is not limited to manually filled-in sports information, and is more scientific. After the multi-dimensional sports data is integrated, comprehensive features are formed to avoid one-sidedness in subsequent evaluations. Dynamic evaluation is performed based on the obtained comprehensive features, taking into account the accumulated fatigue loss caused by long-term exercise, so that the risk assessment results are more accurate, thereby achieving the purpose of dynamic evaluation strategy. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 This is a flowchart of the sports risk assessment system based on big data. DETAILED DESCRIPTION

[0018] Various exemplary embodiments, features, and aspects of the present application will be described in detail below with reference to the accompanying drawings. The same reference numerals in the accompanying drawings represent elements with the same or similar functions. Although various aspects of the embodiments are shown in the accompanying drawings, the drawings are not necessarily drawn to scale unless otherwise indicated.

[0019] The word “exemplary” is used exclusively herein to mean “serving as an example, example, or illustration.” Any embodiment described herein as “exemplary” is not necessarily to be construed as preferred or advantageous over other embodiments.

[0020] In addition, numerous specific details are provided in the following specific examples to better illustrate the present application. Those skilled in the art will appreciate that the present application can be practiced without certain specific details. In some instances, methods, means, and components well known to those skilled in the art are not described in detail in order to highlight the main purpose of the present application.

[0021] The present invention provides a sports risk assessment system based on big data, the system comprising: Sports data collection and integration module; used to obtain athletes' sports data and perform standardized processing on the acquired data; Furthermore, the sports data and environmental data specifically include obtaining athletes' biometric data based on wearable devices, collecting athletes' sports performance data based on big data and IMU inertial sensors, and obtaining behavioral and psychological data based on big data analysis of athletes' training logs.

[0022] Biometric data such as heart rate, body temperature, muscle electrical signals, blood oxygen saturation, etc.; sports performance data such as movement trajectory, impact force and joint load, etc.; behavioral and psychological data such as training duration, intensity, recovery period, etc.

[0023] After obtaining the sports data related to the above athletes, the Z-score standardization is used. , where x is the original value, is the mean value obtained based on big data, Based on the standard deviation obtained from big data, the relevant motion data will eventually be standardized and processed to facilitate subsequent data extraction and risk assessment.

[0024] Furthermore, in order to ensure that the normalized motion data has time series characteristics, the present application further improves the above method as follows: ; Where, is the original motion data, such as joint angular velocity, ground reaction force, Normalize the time series data. is the average of the athletes’ historical data, is the standard deviation of historical data obtained based on big data, is the adaptive cycle, To prevent zero constants, ultimately eliminate individual differences, and preserve time series characteristics, the athlete's historical data mean is a baseline value calculated through a sliding or weighted average of the athlete's long-term accumulated sports data. 3 Raw sports data is instantaneous monitoring values ​​collected in real time. The relationship between the two is reflected in the fact that the historical mean serves as a baseline for dynamic normalization, eliminating individual differences and ensuring comparability of data across different athletes.

[0025] The standard deviation of historical data can refer to the grouping statistical method of risk assessment of financial institutions in the existing technology. The specific steps are: collecting historical data of athletes in the same sport, such as the jumping height of basketball players; establishing data groups according to dimensions such as sports category and body part; calculating the standard deviation of each group of data and storing it in the standard parameter library. This is common knowledge in this technical field, so it will not be elaborated on here.

[0026] The e in the formula is a natural constant, which is used here to construct a dynamic attenuation function. Its physical meaning is that when the duration of the movement increases, the impact of historical data on the current standardization decays exponentially, avoiding excessive weighting of early data.

[0027] In the above formula, Added in It is used to avoid the denominator from approaching zero due to insufficient initial data, and is usually set to 0.01. In the theory of super recovery, the human body's adaptation to training load conforms to the S-shaped curve. It is a small cycle division in the sports training cycle. In the explanation of this formula, it is used as an adaptive cycle and is usually taken as 30 days.

[0028] In the above formula, in addition to obtaining standardized motion data, the obtained motion data can be further analyzed. For example, assuming the athlete's real-time hip angle =172°(15 days), historical baseline =168°, =3.2°, =30, =0.01; final: , , ; The normalized value of 0.775 is in the safe range [0.5, 1.2], so the normalized model of this application can be used as a judgment suggestion for the action standard.

[0029] Data extraction and fusion module: used to extract motion feature data from the normalized motion data and fuse them to generate a motion comprehensive feature vector; The data extraction and fusion module completes the fusion of multi-motion feature data based on the feature fusion model, which includes: ; Where, is the feature-damage association matrix obtained by graph convolutional network training, The safety threshold can be determined by the threshold given by the International Society of Sports Medicine. is the biomechanical nonlinear coefficient, such as 1.5 for bones and 2.0 for ligaments. is the time series feature weight, Temporal anomaly extracted for long short-term memory network.

[0030] The biomechanical nonlinear coefficient describes the degree of nonlinearity in the stress-strain relationship, such as the sudden increase in tendon stiffness in the late stages of stretching. It is typically set between 1.5 and 3.0 and is derived by fitting the Hill model through material testing. The safety threshold represents the critical biomechanical load that human tissue can withstand, such as the maximum torsional torque of the knee joint of 300 Nm. Exceeding this threshold increases the risk of injury. The International Federation of Sports Medicine measures this through biomechanical experiments and adjusts it based on the type of exercise. This is common knowledge in the field, so I won't elaborate on it here. The feature-injury association matrix is ​​constructed by analyzing the anatomical atlas using a graph convolutional network (GCN). Nodes represent joints / muscles, edges represent mechanical transmission pathways, and edge weights are derived from injury cases. This matrix is ​​based on previously derived data and is commonly used in sports risk assessment in this technical field, so we will not elaborate on it here. The temporal anomaly degree is determined by an LSTM network analyzing motion sequences, such as the landing angles of 10 consecutive jumps. Larger output values ​​indicate abnormal compensation patterns, such as a 20% increase in knee internal rotation. The calculation process is as follows: An LSTM network is used to learn an athlete's historical motion sequences, such as 100 sets of standard jump data. The motion sequences are collected in real time and fed into the trained LSTM. The mean squared error (MSE) between the predicted output and the actual sensor output is calculated and normalized using a sigmoid function to an anomaly degree value between 0 and 1. For example, if an abnormal landing posture results in an MSE greater than 0.7, an alert is triggered.

[0031] The feature fusion model derives the single parameter risk contribution based on the Hill model, specifically: , where is the actual stress of the moving tissue, is the yield strength of the moving tissue, is the material hardening index, and is 1.5-3.0 for human tissue. The above model is a biomechanical response model. In order to facilitate the integration of human motion data, a human feature space mapping is constructed, specifically: ,in The feature-injury correlation matrix, i.e., the anatomical structure correlation matrix, is used to construct the biomechanical response model layer with the biomechanical relationship of the human motion chain. Finally, the time dimension feature is introduced and LSTM is used to capture the compensatory action pattern, i.e. , and finally fuse the above multiple models to obtain the feature fusion model disclosed in this application.

[0032] Risk assessment module: performs motion risk assessment based on comprehensive motion feature vectors and verifies accuracy using historical data in big data.

[0033] In this application, the initial risk model of the Logistic regression framework is considered in the motion risk assessment based on the comprehensive motion feature vector, specifically: , while this risk assessment model is applied to static assessment. In the present technical solution, the assessment of motion analysis is often affected by the time dimension. Therefore, this application introduces the time dimension and uses the integral term to simulate the fatigue characteristics of human tissue under long-term exercise, specifically: Where, is the real-time risk index, is the motion comprehensive feature vector, is the injury weight coefficient, reflecting the clinical importance of different injury types. Fatigue attenuation coefficient, which reflects the degree of influence of exercise time superposition on the current state, is adjusted according to the type of exercise. For example, the fast attenuation of sprinting is 0.02 / s, and the long-term accumulation of marathon is 0.005 / s. It is the cumulative risk score, reflecting the cumulative effect of fatigue damage caused by continuous exercise.

[0034] Therefore, the final motion analysis and evaluation of the athlete in this application is as follows, assuming that the athlete's knee internal rotation angular velocity is , the safety threshold is , the asymmetry of the ground reaction force is , LSTM detects the abnormality of landing mode In terms of parameter setting, based on big data , , , , static items =1.2×2.3+0.9×1.8+0.5×0.67=5.01, , when the historical average risk R=0.1, , an emergency risk warning is triggered.

[0035] The integral term of the evaluation function in the risk assessment module also includes: site hardness Temperature and humidity , and through transfer learning, based on the athlete's baseline data, such as joint mobility and maximum oxygen uptake, optimize the fatigue attenuation coefficient and safety thresholds .

[0036] The risk assessment module uses historical data from big data to verify accuracy, including the following steps: The historical data is divided into training set and test set. K-fold cross validation is used to randomly divide the data into K subsets. K-1 subsets are used to train the model in turn, and the remaining 1 subset is used for testing. After repeating K times, the average value is taken to obtain the verification model.

[0037] The risk assessment module uses historical data from big data to verify accuracy and also includes the following steps: The high-risk features output by the sports risk assessment are matched with the actual sports injury records in the validation model. The chi-square test is used to verify whether there is a significant difference between the model prediction results and the actual risk distribution. If the model involves probabilistic prediction, the Brier score can be used to evaluate the calibration of the predicted probability.

[0038] In a specific implementation, the present application provides a computer storage medium and a corresponding data processing unit, wherein the computer storage medium is capable of storing a computer program that, when executed by the data processing unit, executes the invention of the big data-based sports risk assessment system provided by the present invention and some or all of the steps in each embodiment. The storage medium can be a magnetic disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0039] Those skilled in the art will clearly understand that the technical solutions in the embodiments of the present invention can be implemented using computer programs and their corresponding general-purpose hardware platforms. Based on this understanding, the technical solutions in the embodiments of the present invention, or the portion that contributes to the prior art, can be embodied in the form of a computer program, i.e., a software product. This computer program software product can be stored in a storage medium and includes instructions for enabling a device including a data processing unit (such as a personal computer, server, single-chip microcomputer, MCU, or network device) to execute the methods described in various embodiments of the present invention or certain portions of these embodiments.

[0040] The present invention provides a big data-based sports risk assessment system. There are numerous methods and approaches for implementing this technical solution. The foregoing description is merely a preferred embodiment of the present invention. It should be noted that those skilled in the art may make improvements and modifications without departing from the principles of the present invention, and such improvements and modifications are also within the scope of protection of the present invention. Any components not specified in this embodiment may be implemented using existing technologies.

Claims

1. The sports risk assessment system based on big data is characterized by: The system comprises: Sports data collection and integration module; used to obtain athletes' sports data and perform standardized processing on the acquired data; Data extraction and fusion module: used to extract motion feature data from the normalized motion data and fuse them to generate a motion comprehensive feature vector; Risk assessment module: performs motion risk assessment based on comprehensive motion feature vectors and verifies accuracy using historical data in big data.

2. The big data-based sports risk assessment system according to claim 1, characterized in that: In the sports data collection and integration module, the acquisition of athlete sports data includes: acquiring athlete biometric data based on wearable devices, acquiring athlete sports performance data based on big data and IMU, and acquiring behavioral and psychological data based on big data analysis of athlete training logs.

3. The sports risk assessment system based on big data according to claim 1, characterized in that: The motion data collection and integration module completes the standardization process based on a dynamic standardization model, wherein the dynamic standardization model includes: ; Where, is the original motion data, Normalize the time series data. is the average of the athletes’ historical data, is the standard deviation of historical data obtained based on big data, is the adaptive cycle, To prevent division by zero constant.

4. The sports risk assessment system based on big data according to claim 3, characterized in that: The data extraction and fusion module completes the fusion of multi-motion feature data based on a feature fusion model, wherein the feature fusion model includes: ; Where, is the feature-damage association matrix obtained by graph convolutional network training, is the safety threshold, is the biomechanical nonlinear coefficient, is the time series feature weight, Temporal anomaly extracted for long short-term memory network.

5. The sports risk assessment system based on big data according to claim 1, characterized in that: The risk assessment module is based on the input of the comprehensive feature vector of the movement and outputs a real-time risk index. , specifically including the following evaluation functions: ; Where, is the real-time risk index, is the motion comprehensive feature vector, is the injury weight coefficient, reflecting the clinical importance of different injury types. is the fatigue attenuation coefficient, which reflects the impact of historical risks on the current situation. It is the cumulative risk score, reflecting the cumulative effect of fatigue damage caused by the superposition of exercise time.

6. The big data-based sports risk assessment system according to claim 5, characterized in that: The integral term of the evaluation function in the risk assessment module also includes: site hardness Temperature and humidity , and through transfer learning, based on the athlete's baseline data, optimize the fatigue attenuation coefficient and safety thresholds .

7. The sports risk assessment system based on big data according to claim 1, characterized in that: The risk assessment module uses historical data in big data to verify accuracy, including the following steps: The historical data is divided into training set and test set. K-fold cross validation is used to randomly divide the data into K subsets. K-1 subsets are used to train the model in turn, and the remaining 1 subset is used for testing. After repeating K times, the average value is taken to obtain the verification model.

8. The big data-based sports risk assessment system according to claim 7, characterized in that: The risk assessment module uses historical data in big data to verify accuracy and also includes the following steps: The high-risk features output by the sports risk assessment are matched with the actual sports injury records in the validation model. The chi-square test is used to verify whether there is a significant difference between the model prediction results and the actual risk distribution. If the model involves probabilistic prediction, the Brier score can be used to evaluate the calibration of the predicted probability.

Images