Real-time monitoring system for physical exercise of teenagers

Abstract

The invention discloses a teenager physical exercise real-time monitoring system, and relates to the technical field of intelligent monitoring. An individualized health assessment model is constructed through space-time alignment processing of joint three-dimensional movement, electromyographic signals and plantar pressure and a bone age dynamic matching technology. Time sequence deviation is eliminated through unified clock marking and interpolation compensation, parameters such as bone stress deviation degree, muscle synergy asymmetry degree and plantar pressure unbalance degree are extracted, and comprehensive health indexes are generated through weighting. And in combination with a preset training path model, analyzing a health index trend in real time and triggering a grading adjustment instruction, dynamically optimizing training intensity, interval and action schemes, and correcting path model parameters through a closed-loop feedback mechanism to realize adaptive updating. The defects that traditional teenager physique monitoring data fusion is large in deviation and static evaluation hysteresis is high are overcome, real-time accurate monitoring and personalized training optimization of teenager physique exercise are achieved, and the sport injury risk is effectively reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent monitoring, and particularly relates to a real-time monitoring system for teenagers' physical exercise. Background Art

[0002] In the field of teenagers' physical fitness monitoring, existing technologies mostly rely on single-dimensional motion parameters or static physiological indicators for evaluation, and it is difficult to comprehensively reflect the biomechanical state in complex motion scenarios. Traditional systems collect joint, muscle, and plantar pressure data, and at the same time, based on a static evaluation model, use fixed thresholds to judge bone stress and muscle coordination status, ignoring the dynamic impact of teenagers' bone age development on motor ability, and there is a deviation between the evaluation result and the real physiological load.

[0003] In the prior art, the static evaluation model cannot adapt to the differences in biomechanical characteristics at different bone age stages, resulting in uneven applicability of the same training program for teenagers in the development period. These problems limit the practical value of the monitoring system and are difficult to meet the needs of personalized scientific training.

[0004] To address the above problems, there is an urgent need in this field for a monitoring system that can integrate multi-dimensional biomechanical parameters, dynamically adapt to individual development characteristics, and optimize training strategies in real time, so as to solve the deficiencies of traditional methods in data synchronization, evaluation accuracy, and decision-making real-time performance. Summary of the Invention

[0005] In view of the defects of large deviation in data fusion and strong lag in static evaluation in the existing technology for teenagers' physical fitness monitoring, the present invention provides a real-time monitoring system for teenagers' physical exercise.

[0006] To achieve the above object, the present invention is realized through the following technical solutions: A real-time monitoring system for teenagers' physical exercise, which adjusts training parameters based on the sensed data of a sensing device worn on a target person, includes: A data acquisition module, configured to receive three-dimensional joint motion data, electromyography signal data, plantar pressure distribution data in the current training stage, and bone age data in the physiological development stage; A standardized feature library, configured to simultaneously obtain a feature data set of standardized biomechanics in the historical training stage, and the feature data set includes a joint motion amplitude threshold, a muscle coordination symmetry range, and a plantar pressure distribution standard value that match the bone age data; A data processing module, configured to extract joint angle feature points in the three-dimensional joint motion data, the activation time series of the electromyography signal data, and the pressure areas of the plantar pressure distribution data, and perform spatio-temporal alignment processing with the feature data set to generate feature data; A health index generation module that calculates multiple biomechanical difference parameters reflecting bone stress deviation, muscle coordination asymmetry, and plantar pressure imbalance based on the aligned feature data. Generates a comprehensive health index based on the biomechanical difference parameters and records it as time series data. A dynamic path decision module, including: Dynamically compares the change trend of the comprehensive health index in the time series data with the change trend of the target health index in the corresponding stage of a preset training path model to generate a deviation direction and amplitude; the deviation direction is divided into positive deviation and negative deviation, and the positive deviation is used to increase the training amplitude, while the negative deviation is used to weaken the training amplitude. Generates an adjustment instruction according to the deviation direction and amplitude. An execution module for generating a training parameter adjustment amount including training intensity adjustment, time interval correction, and action alternative solutions according to the adjustment instruction through a preset rule mapping table.

[0007] Adopting the technical solution provided by the present invention, compared with the known public technology, it has the following beneficial effects: In view of the particularity of adolescent physical exercise monitoring, the present invention constructs a multi-dimensional and dynamic real-time monitoring system. Its core innovation lies in the precise collection of biomechanical data and its deep binding with the growth stage, effectively solving the problems of ignoring adolescent bone age differences and single evaluation dimensions in traditional monitoring methods. The system integrates joint three-dimensional motion, electromyography signals, plantar pressure, and bone age data, and relies on a standardized feature library to achieve spatio-temporal alignment processing, extracts key parameters such as joint angle feature points and muscle activation sequences, and combines the threshold range of bone age matching to eliminate abnormal data, ensuring the scientific nature of data processing; further generates a comprehensive health index (Hindex) through normalized weighted calculation, and dynamically adjusts the weight coefficients of dimensions such as bone stress, muscle coordination, and plantar pressure, making the evaluation results more in line with the physiological characteristics of different development stages of adolescents, and being more targeted than the traditional fixed-weight evaluation method.

[0008] The dynamic optimization mechanism of the present invention breaks through the limitations of the "one-way feedback" of traditional monitoring systems, and realizes the real-time intelligent optimization of the training plan through the closed-loop control of "monitoring - evaluation - adjustment - iteration". The dynamic path decision module dynamically compares the changing trend of health indicators based on time series with the target trend of the preset training path model, generates hierarchical adjustment instructions for positive / negative deviations (such as shortening / lengthening the training interval, increasing intensity or switching alternative actions), and corrects the parameters in the subsequent stages and records the version changes through the path model update module, forming a virtuous cycle of "data-driven - model self-learning - benchmark update". This mechanism can not only correct training deviations in a timely manner, but also continuously optimize the training path according to the actual effects, which is more timely and accurate than the traditional method relying on empirical adjustment, effectively reducing the risk of sports injuries and improving the exercise efficiency.

[0009] The present invention generates a dynamic simulation model of the skeletal-muscular system through real-time data, combines finite element analysis to predict the stress distribution trend in the future training cycle, and superimposes and displays it with the target value. The system can generate adjustment instructions in advance before potential risks occur (such as when the stress imbalance exceeds the threshold), realizing the leap from "post-intervention" to "forward-looking prevention". This technical means of "virtual-reality" synchronous mapping not only improves the intelligence level of the system, but also provides a safer and more scientific guarantee for teenagers' physical exercise through preventive adjustment, providing a new technical path for the formulation of personalized exercise intervention programs. Brief Description of the Drawings

[0010] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention, and those of ordinary skill in the art can also obtain other drawings without creative efforts based on these drawings.

[0011] Figure 1 It is the connection diagram of the system modules of the present invention; Figure 2 It is the structure diagram for extracting joint angle feature points of the present invention; Figure 3 It is the structure diagram of the path model update module of the present invention; Figure 4 It is the structure diagram of the biomechanical digital twin module of the present invention. Detailed Embodiments

[0012] It is easy to understand that according to the technical solution of the present invention, without changing the essence of the present invention, those of ordinary skill in the art can propose various interchangeable structural ways and implementation ways. Therefore, the following specific embodiments and the accompanying drawings are only exemplary descriptions of the technical solution of the present invention, and should not be regarded as all of the present invention or as a limitation or restriction on the technical solution of the present invention.

[0013] In the prior art, the physical fitness monitoring of teenagers mostly relies on single-sensor data or static evaluation models, and it is difficult to realize the collaborative analysis of multi-dimensional physiological parameters. When collecting joint movement, muscle activation and plantar pressure data by traditional methods, due to the asynchronous sampling timing of sensors, data fusion errors occur, and the dynamic correlation characteristics of the biomechanics of the movement chain cannot be accurately captured. Existing systems use fixed thresholds to evaluate bone stress and muscle coordination status, ignoring the impact of the bone age development differences of teenagers on their motor ability, resulting in the deviation of the risk assessment results from the actual physiological state. Especially in high-intensity scenarios such as strength training, the lack of real-time correlation analysis of plantar pressure distribution and joint load makes it difficult to timely warn of potential sports injury risks.

[0014] To solve the above problems, the inventors found that there is a significant correlation between the spatio-temporal coupling relationship of multi-modal biomechanical data and the characteristics of bone age development, and an individualized health assessment is realized by establishing a dynamic path model. During the research process, it was found that the deviation degree of bone stress, the asymmetry degree of muscle coordination and the imbalance degree of plantar pressure show different weight distribution laws at different bone age stages, and thus a dynamic weight allocation mechanism based on bone age parameters was proposed. Further verified by experiments, the time series trend of multi-dimensional health indicators is matched with the preset path model to construct a training parameter optimization system with self-adaptive adjustment ability.

[0015] Specifically, the monitoring system first synchronously collects the three-dimensional movement trajectories of joints, surface electromyography signals and plantar pressure distribution data. The time sequence deviation between sensors is eliminated through spatio-temporal alignment processing, and key joint angle feature points, muscle activation time differences and pressure distribution ratio parameters are extracted. Combining with the biomechanical threshold library matched with bone age, quantitative indicators such as the deviation degree of bone stress and the asymmetry degree of muscle coordination are calculated, and a comprehensive health index is generated by weighting. When it is detected that the trend of the health index continuously deviates from the preset path, the system automatically triggers a hierarchical adjustment strategy: shorten the training interval and increase the intensity gradient for positive deviations, and extend the recovery period and switch to a low-risk alternative action plan for negative deviations. During continuous monitoring, the system dynamically updates the path model parameters according to real-time data, and optimizes the evaluation threshold and decision logic through a closed-loop feedback mechanism.

[0016] Compared with the prior art, traditional methods are limited by single-dimensional data analysis and static evaluation models, and it is difficult to adapt to the dynamic change characteristics of the adolescent growth and development stage. This solution innovatively integrates multi-source biomechanical data and bone age development parameters, and establishes an individualized health assessment system through spatio-temporal coupling analysis. Different from the traditional fixed-threshold judgment mechanism, this solution can dynamically adjust training parameters according to real-time trends, and continuously optimize decision-making accuracy through the self-learning function of the path model, significantly improving the monitoring reliability and safety guarantee ability in complex motion scenarios.

[0017] Through the above technical solution, this application effectively solves the problems of multi-modal data fusion deviation and static evaluation model lag, and realizes accurate health risk warning while ensuring real-time performance. The dynamic path decision mechanism takes into account individual development differences and exercise load characteristics, and the model parameter self-adaptive adjustment function ensures the accuracy of long-term monitoring. This method provides intelligent technical support for scientific physical training of adolescents, and is particularly suitable for scenarios of improving exercise ability and preventing injuries during the critical period of growth and development.

[0018] After introducing the basic concept of the present invention, the embodiments of the present invention will be specifically introduced below with reference to the accompanying drawings.

[0019] Embodiment

[0020] A real-time monitoring system for adolescent physical exercise in this embodiment, as Figure 1 shown, adjusts training parameters based on the sensed data of the sensing device worn on the target person, including: A data acquisition module, configured to receive three-dimensional joint motion data, electromyography signal data, plantar pressure distribution data in the current training stage, and bone age data in the physiological development stage; A standardized feature library, configured to simultaneously obtain a feature data set of standardized biomechanics in the historical training stage, and the feature data set includes a joint motion amplitude threshold, a muscle coordination symmetry range, and a plantar pressure distribution standard value that match the bone age data; A data processing module, configured to extract joint angle feature points in the three-dimensional joint motion data, the activation time series of the electromyography signal data, and the pressure area of the plantar pressure distribution data, and perform spatio-temporal alignment processing with the feature data set to generate feature data; A health index generation module, based on the aligned feature data, calculates a plurality of biomechanical difference parameters reflecting bone stress deviation, muscle coordination asymmetry, and plantar pressure imbalance; Generates a comprehensive health index according to the biomechanical difference parameters, and records it as time series data; A dynamic path decision module, including: Dynamically compare the change trend of the comprehensive health index in the time series data with the change trend of the target health index in the corresponding stage of the preset training path model to generate the deviation direction and amplitude; the deviation direction is divided into positive deviation and negative deviation, the positive deviation is used to increase the training amplitude, and the negative deviation is used to weaken the training amplitude; Generate an adjustment instruction according to the deviation direction and amplitude; An execution module, configured to generate a training parameter adjustment amount including training intensity adjustment, time interval correction, and action alternative solutions according to the adjustment instruction through a preset rule mapping table.

[0021] Among them, the joint three-dimensional motion data refers to the three-dimensional space coordinates and angle change data of each joint (such as knee, ankle, hip joint) during the training of teenagers obtained by an inertial measurement unit (IMU) or an optical motion capture device. Specifically, it can be implemented by an Xsens inertial sensor or a Vicon optical camera, and is used to record the dynamic trajectory of the joint during the movement in real time. The electromyography signal data refers to the bioelectric signals of the target muscle groups (such as quadriceps femoris, gastrocnemius) collected by surface electromyography (sEMG) sensors. Specifically, it can be implemented by a Delsys Trigno wireless electromyograph, and is used to reflect the timing characteristics of muscle activation. The plantar pressure distribution data refers to the pressure value distribution data of each area of the sole (heel, forefoot, arch) obtained by a pressure sensing insole or a force platform. Specifically, it can be implemented by a Novel Pedar-X insole pressure system, and is used to quantify the balance of foot force. The bone age data refers to the physiological development stage data of teenagers predicted by left wrist X-ray bone age assessment (such as Greulich-Pyle method) or bioinformatics model, and is used to match the personalized biomechanical assessment benchmark.

[0022] Among them, the standardized biomechanical feature dataset refers to a benchmark library constructed based on large-sample teenager biomechanical experiment data, which is labeled by sports medicine experts or trained by machine learning models, and includes joint movement amplitude thresholds (such as the safe range of knee joint flexion angle), muscle coordination symmetry range (such as the allowable value of the electromyography activation time difference between the left and right sides), and plantar pressure distribution standard values (such as the heel-forefoot pressure ratio benchmark value) matching each bone age stage, serving as a dynamic reference benchmark for the assessment of teenagers' physical exercise.

[0023] Among them, the spatio-temporal alignment processing refers to unifying the time stamps of the joint three-dimensional motion data, the sampling frequencies of the electromyography signals, and the acquisition periods of the plantar pressure data to the same time coordinate system, and eliminating the device space offset through three-dimensional coordinate registration (such as SVD rigid transformation), which is used to ensure the consistency of multi-source data in the time and space dimensions and provide a unified benchmark for subsequent feature extraction.

[0024] Among them, the biomechanical difference parameter refers to the degree of abnormal movement of bones, muscles, and the sole of the foot quantified by mathematical methods, which is used to construct multi-dimensional health assessment indicators. Among them, the comprehensive health indicator is used to simplify the trend analysis of the evolution of training effects; the time series data refers to the set of comprehensive health indicators recorded in the order of training stages, which is specifically stored in a time series database (such as InfluxDB) or a CSV file, and is used to reveal the time correlation and dynamic deviation characteristics of the changes in health indicators.

[0025] Among them, the preset training path model refers to a digital training plan that includes the target health indicators at each bone age stage, the time nodes of the training cycle, and the allowable deviation range. The adjustment instruction refers to generating a hierarchical correction command according to the deviation degree between the actual change rate and the target change rate of the comprehensive health indicator, which is specifically triggered by a rule engine (such as the first threshold interval of positive deviation triggers shortening the training interval) or a classification model (such as a gradient boosting tree classifier), and is used to automatically adjust the training intensity or action plan.

[0026] Among them, the training parameter adjustment amount refers to the specific correction values for training intensity, time interval, and action substitution, which are specifically obtained by a look-up table method (such as a mapping table of deviation amplitude and intensity increment) or a regression model (such as linear regression to calculate the adjusted duration of the interval), and is used to optimize the execution strategy of subsequent training stages; the path model update refers to the data iteration process of feeding back the adjustment amount to the preset training plan in real time, which is specifically implemented by version control (recording the change time and the effective stage identifier) or a parameter rewriting mechanism, and is used to maintain the health indicator within the allowable deviation range of the model, forming a closed-loop control of "monitoring - evaluation - adjustment - iteration".

[0027] The core innovation of this application lies in constructing a closed-loop feedback control system based on biomechanical difference parameters and a preset training path. By comparing the evolution trend of comprehensive health indicators during the training of teenagers with the target path of the preset model in real time, it automatically generates training parameter adjustment instructions and updates the path model, solving the technical bottleneck that traditional physical exercise monitoring technologies lack a dynamic evaluation system and adaptive control ability, and achieving the leap from "experience guidance" to "data-driven".

[0028] The working process and principle of this application are as follows: First, the data acquisition module receives the three-dimensional joint motion data, electromyogram signal data, plantar pressure distribution data, and bone age data in the current training stage; the standardized feature library synchronously obtains the historical standardized biomechanical feature dataset (including joint motion amplitude thresholds, muscle coordination symmetry ranges, etc.) that matches the current bone age; the data processing module extracts joint angle feature points (such as the key value of the knee joint flexion angle), electromyogram activation time series, and plantar pressure areas, and performs spatio-temporal alignment with the standardized dataset to generate feature data; the health index generation module calculates difference parameters such as bone stress deviation, muscle coordination asymmetry, and plantar pressure imbalance based on the aligned features, generates a comprehensive health index through normalization and dynamic weighting, and records it as time series data; the dynamic path decision module applies a sliding window analysis to the time series data, calculates the actual change rate and compares it with the target change rate of the preset training path model, and generates a hierarchical adjustment instruction according to the deviation direction and amplitude; the execution module converts the adjustment instruction into specific parameter adjustment amounts for training intensity adjustment, time interval correction, and action alternative solutions through a preset rule mapping table; finally, the path model update module corrects parameters such as time nodes and allowable deviation ranges in subsequent stages to maintain the comprehensive health index within the allowable range of the model, so as to achieve dynamic monitoring and intelligent optimization of adolescent physical exercise, ensure that the training process conforms to the laws of growth and development, and reduce the risk of sports injuries.

[0029] As Figure 2 shown, this application further proposes that the extraction of joint angle feature points includes: Based on a preset kinematic feature template, identify the key angle values of the knee joint flexion angle, ankle joint dorsiflexion angle, and hip joint abduction angle from the three-dimensional joint motion data; Perform timestamp alignment on the key angle values. When the angle fluctuation of 5 consecutive frames of data is less than the fluctuation range, mark it as a stable feature point; Eliminate abnormal angle data that exceeds the amplitude threshold range matched by the bone age data.

[0030] Specifically, the preset kinematic feature template adopts a hierarchical kinematic feature recognition strategy. Based on the joint range of motion template library constructed according to the standards of the International Society of Biomechanics, it includes the gold standard angle curves of 12 basic motion patterns (such as walking, jumping, and squatting). When the three-dimensional joint motion data stream is input, the dynamic time warping (DTW) algorithm runs in real time for pattern matching, and key phase points (such as the moment of maximum knee flexion) are synchronously identified in three dimensions: the knee flexion angle, the ankle dorsiflexion angle, and the hip abduction angle. The feature verification engine monitors the stability of the angle sequence through sliding window variance analysis (such as window width 200ms, step size 50ms). When the fluctuation amplitude of 5 consecutive frames (corresponding to 100ms) of data is less than the preset bone age dynamic threshold (such as ±3° at 12 years old, ±5° at 15 years old), the feature point locking mechanism is triggered, and the joint angle value at this moment is written into the stable feature library. The abnormal data filtering module is connected to the bone age - joint range of motion database in real time, and uses the Mahalanobis Distance to calculate the deviation degree of the current angle from the statistical distribution of the same bone age group, and hard deletes the outliers beyond the 3σ range.

[0031] In the data processing pipeline, after the candidate feature points are output based on the preset kinematic feature template, time series statistics are injected for secondary verification, and the two achieve millisecond-level data exchange through a shared memory buffer. When it is detected that the fluctuation variance of the candidate feature points exceeds the threshold, the template parameter self-correction mechanism is immediately triggered, and the matching similarity tolerance is dynamically adjusted according to the current motion mode (decreasing from the default 0.8 to 0.6). The bone age adaptation anomaly filter establishes two-way communication with the feature library, not only eliminating abnormal data, but also feeding back the valid features to the bone age - joint range of motion database, and the normal angle range of each bone age segment is updated in real time through an online learning algorithm (such as stochastic gradient descent), forming a dynamically evolving human kinematic benchmark.

[0032] Compared with the traditional scheme, the existing joint angle analysis systems generally face the problem that the traditional fixed threshold template cannot adapt to the variation of individual motion patterns, resulting in mis-matching of walking and running actions. This scheme uses DTW dynamic regularization and self-correction mechanism to improve the cross-action type recognition accuracy; the traditional method uses single-frame threshold judgment and is vulnerable to motion artifacts (such as instantaneous displacement of sensors). This scheme introduces multi-frame joint variance analysis, combines the dynamic fluctuation range of bone age, and reduces the false feature point recognition rate; the existing systems usually perform abnormal filtering at the end of data processing, resulting in the contamination of the intermediate analysis process by invalid features; this scheme constructs a feed-forward - feedback dual-channel filtering mechanism to intercept abnormal data at the source of feature extraction.

[0033] This application constructs an intelligent extraction system for joint angle features for biomechanical assessment. Through the collaborative processing of three stages: kinematic feature template matching, dynamic stability determination, and bone age adaptation anomaly filtering, it realizes the accurate capture and reliable analysis of joint movement features. A spatio-temporal consistency guarantee mechanism is established in the feature extraction stage to improve the effective feature capture rate of joint angle data, while reducing the misjudgment rate caused by motion artifacts, providing a high-confidence input source for subsequent biomechanical modeling. Through the deep collaboration of kinematic template matching, dynamic stability verification, and bone age intelligent filtering, this solution constructs a joint angle feature extraction system with self-evolution ability. This application not only achieves high-precision feature capture, but also enables the system to adapt to different motion scenarios and individual differences through real-time data verification and parameter optimization, establishing a new generation of feature engineering benchmark for accurate sports biomechanical analysis and providing reliable technical support for real-time motion intervention decision-making.

[0034] This application further proposes that the biomechanical difference parameters include: Bone stress deviation degree, which is the relative difference between the real-time bone stress value and the safety threshold of bone age data matching; Muscle coordination asymmetry degree, which is the product of the absolute value of the activation time difference between the left and right target muscle groups and the signal intensity ratio; Plantar pressure imbalance degree, which is the absolute deviation of the pressure ratio between the lateral heel and the forefoot and the standard value.

[0035] Specifically, through the dynamic coupling analysis of bone stress deviation degree, muscle coordination asymmetry degree, and plantar pressure imbalance degree, a panoramic quantitative assessment of motion risk is realized. Among them, the bone stress deviation degree reflects the physiological safety boundary of the skeletal system load, the muscle coordination asymmetry degree represents the neuromuscular control efficiency, and the plantar pressure imbalance degree indicates the motion posture stability. The three generate a comprehensive health index through a weighted fusion model, forming a biomechanical assessment framework covering the trinity of "structure - control - posture".

[0036] The bone stress deviation degree is calculated in real time based on a finite element biomechanical simulation engine. The three-dimensional joint movement trajectory data is input into a personalized bone model adapted to the bone age (for example, an epiphyseal plate layered grid model is used at 12 years old, and a trabecular microstructure model is used at 15 years old). The relative deviation rate (ΔS = (Speak - Ssafe) / Ssafe×100%) of the stress peak Speak and the safety threshold Ssafe in the growth plate area is solved through a dynamic load mapping algorithm. The muscle coordination asymmetry degree uses a time-frequency joint analysis method for dual-channel electromyography signals to perform Hilbert-Huang transform on the activation sequences of the target muscle groups on the left and right sides (such as the vastus lateralis and semitendinosus), and extracts the phase difference Δt and the energy spectrum ratio RE within the time window of 0 - 50 ms to construct the asymmetry index Masym = Δt×|1 - Ssafe|. The plantar pressure imbalance degree is calculated through plantar partition pressure integration operation, and the pressure ratio Pratio of the unilateral heel area (Zone1-3) and the forefoot area (Zone4-6) in the midstance phase is calculated and dynamically compared with the standard value Pstd matched to the bone age to generate the pressure imbalance degree Dp = |Pratio - Pstd|.

[0037] During the dynamic training cycle, through a spatio-temporal feature fusion engine for joint analysis, the bone stress deviation degree (ΔS) is used as the structural risk baseline. When ΔS>15%, the muscle coordination compensation mechanism is triggered, and the system automatically increases the evaluation weight of Masym to 0.6 to prevent compensatory injuries by enhancing the muscle coordination monitoring sensitivity. The plantar pressure imbalance degree (Dp) and ΔS form a composite warning signal. When Dp>0.2 and ΔS>10%, it is determined as a high fall risk state, and the plantar tactile feedback device is activated to implement immediate posture correction. 120 groups of feature vectors are generated per second for the three parameters and input into the LSTM time series prediction network. The contribution weights of each parameter to the comprehensive health index are dynamically adjusted through the gating unit to achieve early prediction of risk evolution.

[0038] Compared with the traditional scheme, most existing biomechanical evaluation systems adopt a single-parameter threshold alarm mechanism. For example, only the overlimit of bone stress or the abnormality of electromyography signal amplitude is monitored, and the composite risk scenario of multi-system decompensation cannot be recognized. However, through the three-parameter dynamic coupling model of this scheme, three typical risk modes can be accurately distinguished: Structural overload (dominated by ΔS): When ΔS>20% and Masym<0.4, it is determined as pure mechanical overload, and an instruction to reduce the training intensity is generated; Controlled compensation (dominated by Masym): When Masym>0.6 and ΔS<15%, it is recognized as neuromuscular control decompensation, and bilateral electromyography biofeedback training is triggered; Composite imbalance (type of three-parameter synchronous growth): When ΔS, Masym, and Dp exceed the threshold for 5 consecutive seconds, it is determined as a precursor to systemic collapse, and an emergency brake is executed and an alternative training plan is pushed.

[0039] This solution reduces the misjudgment rate of compound sports injuries, shortens the warning response time of bone stress, realizes early intervention through the early compensatory signals of Masym, improves the accuracy of the correlation recognition between plantar pressure imbalance and abnormal muscle coordination. Through the spatio-temporal coupling and dynamic weight adjustment of three-level parameters, this solution constructs an intelligent evaluation system with the ability of physiological system interaction perception, enabling the identification of sports risks to leap from traditional single-dimensional threshold judgment to multi-system collaborative state diagnosis, providing a new generation of biomechanical analysis paradigm for precise sports intervention.

[0040] This application further proposes to generate comprehensive health indicators, including: Normalize the bone stress deviation degree, muscle coordination asymmetry degree, and plantar pressure imbalance degree; Dynamically adjust the weight coefficients according to bone age data, and perform weighted summation to generate comprehensive health indicators: Hindex = α × Sdev + β × Masym + λ × |Rp - γ| Where Sdev represents the bone stress deviation degree, Masym represents the muscle coordination asymmetry degree, and |Rp - γ| represents the plantar pressure imbalance degree. α, β, and λ respectively represent the corresponding weight coefficients, which are dynamically adjusted according to bone age data. Rp represents the actual plantar pressure-related parameter, and γ is a reference value. The greater the absolute value of the difference between the two, the more serious the plantar pressure imbalance.

[0041] The weight adjustment dynamically configures the fusion coefficients according to bone age data. When the bone age < 13 years old, set α = 0.6, β = 0.25, λ = 0.15 to highlight the sensitivity of bone development; when the bone age ≥ 16 years old, adjust to α = 0.4, β = 0.35, λ = 0.25 to strengthen the weight of muscle coordination monitoring. The fusion engine calculates Hindex in real time through a sliding window (such as setting the window width to 2 seconds and the step size to 0.5 seconds), and triggers an alarm when the continuous window exceeds the threshold.

[0042] During the data stream processing, the three parameters are precisely aligned through a hardware-level time synchronization framework: the bone stress data (10ms cycle) and the electromyogram signal (5ms cycle) use the PTP protocol hardware clock synchronization, and the plantar pressure data (20ms cycle) is matched to the unified time axis through interpolation resampling. The global timestamp generation module attaches an absolute time tag with μs-level accuracy to each data packet to ensure that the phase deviation of the three within the fusion window is less than 1ms. The weight dynamic regulator interacts with the timestamp engine in real time. When the detected timing deviation exceeds the tolerance (such as the bone stress data delay > 3ms), the weight coefficient of the affected parameter is automatically reduced (such as temporarily reducing α by 20%) to avoid asynchronous data contaminating the fusion result.

[0043] Compared with traditional solutions, existing health assessment systems generally have three major defects. The traditional fixed-weight model cannot adapt to the physiological characteristic changes during the rapid development period of teenagers, resulting in assessment errors for teenage users during the rapid development period. This solution improves the cross-age assessment consistency through dynamic bone age weight mapping; there are timing jitters in traditional software-level time synchronization, causing phase misalignment between the pressure peak and joint angle; this solution combines hardware clock synchronization with interpolation compensation to compress the alignment error of multi-source data; traditional Min-Max normalization ignores the population distribution characteristics of bone age, and the parameters of 12-year-old and 15-year-old users are forced to be scaled to the same interval. This solution uses bone age clustering Z-score processing to retain the specificity of physiological development and improve the sensitivity of anomaly detection.

[0044] This application constructs a comprehensive health assessment system based on the bone age response mechanism. Through the dynamic normalization processing and adaptive weight fusion of multi-source biomechanical parameters, it realizes the accurate quantification of sports health risks. This solution establishes a spatio-temporal consistency guarantee mechanism in the index generation stage, enabling the three of bone stress, muscle coordination, and plantar pressure to complete feature fusion under a unified spatio-temporal benchmark, compressing the comprehensive assessment error of multi-modal data, and providing a highly reliable decision-making basis for real-time sports intervention. Through spatio-temporal alignment normalization processing and bone age response-based weight fusion, this solution constructs a health assessment engine with physiological adaptation capabilities. This application not only breaks through the limitations of traditional static models, but also through the deep coordination of hardware-level time synchronization and dynamic parameter adjustment, realizes the accurate fusion of multi-source biomechanical data, enabling the sports health risk assessment to leap from single-parameter threshold judgment to multi-system joint state diagnosis, and establishing a new generation of technical standards for intelligent sports management systems.

[0045] This application further proposes that the construction method of the preset training path model includes: Obtain at least 100,000 cases of adolescent biomechanical experimental data to generate an experimental dataset, analyze the experimental dataset through the random forest algorithm, and group them to establish the benchmark values of target health indicators, and determine the allowable deviation Δ: When the bone age of the experimental data is 12 - 14 years old: the allowable deviation Δ = 0.3; When the bone age of the experimental data is 15 - 18 years old: the allowable deviation Δ = 0.2; When the actual health indicators exceed the allowable deviation Δ for 5 consecutive times, trigger the model self-learning mechanism to update the allowable deviation Δ.

[0046] Specifically, obtain at least 100,000 cases of adolescent biomechanical experimental datasets through parameter normalization, calculate the population means and standard deviations of the bone stress deviation degree, muscle coordination asymmetry degree, and plantar pressure imbalance degree by bone age grouping, and use the Z-score algorithm to generate standardized parameters to eliminate the influence of bone age differences on the parameter amplitudes.

[0047] In this embodiment, a bone age-weight mapping rule is established through random forest model analysis: For the 12 - 14-year-old stage: α = 0.6, β = 0.25, λ = 0.15; For the 15 - 18-year-old stage: α = 0.4, β = 0.35, λ = 0.25; When the detected bone age is in the critical zone (such as 14.5 - 15.5 years old), the Sigmoid function is used to smoothly transition the weight coefficients to avoid sudden changes in the evaluation results.

[0048] The initial allowable deviation Δ is set to Δ = 0.3 for 12 - 14 years old and Δ = 0.2 for 15 - 18 years old. When the actual H index When it exceeds the allowable deviation Δ for 5 consecutive times, the gradient descent optimization algorithm is triggered, and the Δ value is dynamically updated based on the latest 100 sets of data. The update amplitude is limited to ±0.05 per time to prevent overfitting.

[0049] Compared with the traditional scheme, the traditional fixed Δ value (such as uniformly setting Δ = 0.25) cannot adapt to the non-linear characteristics of bone age development. This scheme enables the Δ value to dynamically adapt to the individual development rhythm through a self-learning mechanism. The traditional artificial experience ignores the development coordination relationship of the musculoskeletal system. This scheme is based on the random forest feature importance analysis of 100,000 cases of data, making the weight allocation conform to medical evidence-based basis; the traditional Min-Max normalization forces different bone age parameters into the same interval, resulting in the compression of the plantar pressure characteristics of 12-year-old users; this scheme uses bone age grouping Z-score processing to retain the true distribution form of the parameters in physiological development.

[0050] This application constructs a comprehensive health assessment system based on the bone age response mechanism. Through the dynamic normalization processing of multi-source biomechanical parameters, adaptive weight fusion, and deviation self-learning optimization, it realizes the accurate quantification and dynamic tracking of sports health risks. This scheme establishes a strong coupling relationship between physiological development characteristics and dynamic weights in the index generation stage, and through the self-learning mechanism, it realizes the continuous evolution of the evaluation model, improves the cross-age evaluation consistency of health indicators, and provides high-precision decision support for personalized sports management. Through the deep coordination of multi-source parameter spatio-temporal alignment, bone age-driven weight fusion, and deviation self-learning optimization, this scheme constructs an intelligent evaluation system with physiological adaptation ability.

[0051] This application further proposes that the dynamic comparison and generation of adjustment instructions include: Apply sliding window analysis to the comprehensive health indicators in the time series data, calculate the actual change rate of the health indicators within the window period, and compare it with the target change rate in the corresponding stage of the preset training path model; When the actual change rate continuously exceeds the allowable deviation range of the target change rate, generate a graded adjustment instruction according to the deviation direction.

[0052] This application constructs an intelligent training path management system based on the evolution of real-time health indicators. Through the synergistic effect of sliding window dynamic monitoring, deviation elasticity evaluation, and model self-learning optimization, precise dynamic regulation of the training plan is achieved. This solution breaks through the limitations of traditional static threshold management, establishes an adaptive decision-making system with evolutionary capabilities, shortens the early warning response time for training path deviation, improves the accuracy of dynamic adjustment, and provides real-time closed-loop management support for the scientific training of teenagers.

[0053] Specifically, after generating the comprehensive health indicators, sliding window analysis is applied to the comprehensive health indicator data in the time series. During the sliding window analysis process, the actual change rate of the health indicators within the window period is calculated, and the calculation of the actual change rate is based on the changes in the comprehensive health indicators within this window period. Then it is compared with the target change rate at the corresponding stage in the preset training path model. The preset training path model is trained based on a large amount of health data, and the target change rate at the corresponding stage is the ideal reference value for the change of health indicators at this stage.

[0054] When the actual change rate continuously exceeds the allowable deviation range of the target change rate, a hierarchical adjustment instruction is generated according to the deviation direction. Here, the hierarchical adjustment instruction is determined based on the deviation direction and the degree of excess. Different degrees and directions of deviation correspond to different levels of adjustment instructions, and these adjustment instructions can be used for subsequent health interventions, rehabilitation plan formulation, etc., so as to achieve effective monitoring and precise intervention of the individual's health status, ensuring that the individual's health status can be maintained within the ideal range or develop in the ideal direction. The system adopts a dual-engine collaborative architecture, deploys a sliding window analyzer (such as setting the window width to 30 seconds and the step size to 5 seconds), and continuously calculates the change slope Kactual of the health indicator Hindex within the window period. The motion artifact interference is eliminated through a Kalman filter. When the actual change rate continuously exceeds the allowable deviation range of the target change rate, for example, when the absolute deviation |ΔK| between Kactual and the preset path target slope Ktarget of three consecutive windows is > △, the hierarchical adjustment mechanism is triggered.

[0055] Compared with the traditional solution, the traditional batch processing mode (calculating once per minute) results in a delay in deviation response; this solution adopts a streaming sliding window analysis to improve the monitoring frequency; the traditional threshold adjustment relies on manual experience and there is a risk of medical rationality. This solution constructs a clinical knowledge graph verification module to ensure that each update of the △ value complies with the orthopedic safety rules.

[0056] Through the deep collaboration of real-time tracking with a sliding window, elastic deviation thresholds, and enhanced self-learning, the training efficiency is dynamically improved by adjusting instructions, the accuracy of the high-intensity action library is activated, the elastic threshold regulator compresses the Δ value optimization cycle to meet the needs of the growth spurt period, and the closed-loop feedback mechanism reduces the deviation rate of training plan execution and the incidence of sports injuries. This application constructs an intelligent training management center with evolutionary capabilities, breaks through the static management mode of traditional solutions, and realizes the full-link closed loop of "millisecond-level monitoring - intelligent decision-making - medical verification - model evolution".

[0057] This application further proposes that the trigger logic for hierarchical adjustment of instructions includes: Obtain the deviation direction and judge the deviation direction; If it is a positive deviation (actual change rate > target change rate + Δ): When the magnitude of the positive deviation is within the first threshold interval, generate an instruction to shorten the training interval; When the magnitude of the positive deviation is within the second threshold interval, generate an instruction to advance to the next training stage and increase the intensity; If it is a negative deviation (actual change rate < target change rate - Δ): When the magnitude of the negative deviation is within the first threshold interval, generate an instruction to extend the training interval; When the magnitude of the negative deviation is within the second threshold interval, generate an instruction to force a switch to an alternative action library, which is selected from historical low-risk actions.

[0058] This application constructs an adaptive training management system based on real-time biomechanical state perception. Through a triple collaborative mechanism of hierarchical deviation recognition, elastic instruction generation, and medical safety verification, it realizes precise dynamic regulation of the training process. This solution breaks through the traditional static threshold management paradigm, establishes an intelligent decision-making system with real-time response and self-learning capabilities, improves the medical compliance of training adjustment instructions, and compresses the instruction execution response speed to the 800ms level, providing closed-loop management support for scientific training of teenagers.

[0059] Specifically, in one embodiment, real-time deviation analysis deploys a sliding window (window width 30 seconds, step size 5 seconds) to continuously calculate the change rate Kactual of the health index Hindex, and eliminates motion artifact interference through a Kalman filter. When it is detected that Kactual exceeds the target change rate Ktarget ± △ three times continuously, a hierarchical decision tree is triggered: When there is a positive deviation (Kactual > Ktarget + △): First-level instruction (first threshold interval: △ < deviation ≤ 1.5△): Dynamically shorten the training interval Δt_new = Δt × [1 - 0.2 × (deviation / △)], and the minimum interval is not less than the safety threshold Δtmin = 45 seconds; Secondary instruction (second threshold range: deviation > 1.5△): Activate the next-stage training protocol in advance, with the intensity increase coefficient β = 1 + 0.15×(deviation / △), and synchronously inject a high-difficulty action library (risk coefficient < 0.3); When there is a negative deviation (Kactual < Ktarget - △): Primary instruction (first threshold range: △ < deviation ≤ 1.5△): Extend the training interval Δt_new = Δt×[1 + 0.3×(deviation / △)], with the maximum interval not exceeding Δt_max = 120 seconds; Secondary instruction (second threshold range: deviation > 1.5△): Forcefully switch to an alternative action library, construct an alternative solution pool based on historical low-risk actions (injury incidence rate < 5%), and use cosine similarity to match the current motion pattern (similarity > 85%).

[0060] Meanwhile, the medical safety verification integrates a clinical knowledge graph (including orthopedic safety rules) to perform real-time compliance verification on the generated adjustment instructions. When it is detected that an instruction may trigger a risk (such as the predicted value of bone stress after intensity increase > Ssafe×1.2), the instruction correction mechanism is immediately triggered, and an alternative solution is generated within 50ms through a reinforcement learning model.

[0061] During the operation cycle, the dual engines achieve deep cooperation through a high-speed data bus. The biomechanical sensing array transmits Hindex time-series data to the deviation analysis engine at a frequency of 200Hz, and simultaneously transmits raw parameters such as real-time stress values and electromyogram synergy degrees to the safety verification engine; the correction instructions from the safety verification engine are transmitted back to the deviation analysis engine through a priority interrupt channel, triggering dynamic adjustment of decision tree parameters (such as reducing the sensitivity of the △ value); after the daily training is completed, the system automatically analyzes the instruction execution effect data (such as injury incidence rate, training efficiency improvement rate), and uses the Q-learning algorithm to optimize the grading threshold (△ value) and intensity adjustment coefficient, with the update range limited to ±0.05 / day to ensure that the evolution process complies with the medical safety boundary.

[0062] Through the above technical solutions, through the deep cooperation of hierarchical deviation recognition, real-time safety verification, and model self-learning, this solution constructs an intelligent training control center with the ability to evolve. It breaks through the static management shackles of traditional solutions and realizes the full-link closed-loop of "millisecond-level perception - intelligent decision-making - medical escort - dynamic evolution".

[0063] As Figure 3 shown, the present application further proposes that the real-time monitoring system for teenagers' physical exercise also includes a path model update module, which is used to update the preset training path model according to the training parameter adjustment amount to maintain the comprehensive health index within the allowable deviation range of the training path model; Among them, the preset training path model includes the target health indicators, training cycle time nodes, and allowable deviation ranges preset for each bone age stage, serving as the dynamic benchmark for the actual training process.

[0064] Among them, the preset training path model covers the target health indicators, training cycle time nodes, and allowable deviation ranges preset for each bone age stage. During the actual training process, the corresponding target health indicators are set according to the characteristics of different bone age stages. The training cycle time nodes stipulate the training progress and status that should be achieved at different time stages. The allowable deviation range provides a certain flexible space for the entire training process, ensuring that the system will not make misjudgments due to minor fluctuations.

[0065] During the training process, the preset training path model is updated in real time according to the adjustment amount of training parameters. For example, when certain parameters change during actual training, relevant parts of the preset training path model are adjusted according to the specific situation. For instance, if it is found that the training progress of a certain bone age stage is ahead of or behind the preset training cycle time node, the subsequent time node arrangements and the setting of target health indicators are adjusted accordingly to keep it within the allowable deviation range.

[0066] At the same time, when generating the comprehensive health indicator, the bone stress deviation degree, muscle coordination asymmetry degree, and plantar pressure imbalance degree are first normalized, which makes these indicators in different dimensions comparable and fusible. Then, the weight coefficients are dynamically adjusted according to the bone age data, and weighted summation is performed to generate the comprehensive health indicator. The dynamically adjusted weight coefficients and the change situation of the comprehensive health indicator are fed back to the path model update module, providing an important basis for its update of the preset training path model.

[0067] Through the close interaction and collaborative work of the above parts, the entire system can effectively adjust the training path model according to the actual training situation, making the comprehensive health indicator always maintain within the allowable deviation range, thereby providing a high-precision and highly adaptable dynamic benchmark for training and significantly improving the training effect and quality.

[0068] This application further proposes that updating the preset training path model includes: Model parameter correction, adjusting the time node interval and allowable deviation range of the subsequent stage according to the actual health indicator trend; Version identification generation, recording the model change time, effective stage identification, and adjusted deviation range, and associating them with historical training data; Closed-loop control execution, writing the updated path model into the edge computing device as the real-time comparison benchmark for the next monitoring cycle.

[0069] Among them, model parameter correction refers to dynamically adjusting the time node interval and allowable deviation range in the subsequent training stage based on the real-time trends of actual health indicators (such as real-time risk coefficient, muscle coordination asymmetry, dynamic pressure load index, etc.). Specifically, online learning algorithms in machine learning (such as stochastic gradient descent) can be used. By calculating the deviation between the actual indicator and the target indicator (Δ = actual value - target value), and combining the statistical laws of historical training data (such as the time series distribution of deviations), the time node interval (such as shortening the original 3 days / stage to 2 days / stage) and the allowable deviation range (such as widening ±10% to ±15%) are automatically optimized. For example, when it is monitored that the risk coefficient of teenagers is continuously lower than the target value for 2 consecutive weeks, the system will shorten the time node interval of the next stage to accelerate the training progress; if the risk coefficient frequently exceeds the original deviation range, the allowable deviation will be expanded to improve the model's inclusiveness.

[0070] Version identification generation refers to generating a unique version identification for each model change, recording the model change time, the effective stage identification (such as "stage 3_v2"), and the adjusted deviation range, and associating and storing them with historical training data (such as the risk coefficient and pressure ratio in the previous 30 days) through a hash algorithm. In specific implementation, the version identification includes a timestamp (accurate to milliseconds), a stage number (such as "T3" representing the third stage), an adjustment type (such as "time node shortening"), and a check code (to ensure that the data has not been tampered with), and the version information is bound to the corresponding historical data block through blockchain technology to form a traceable model change chain.

[0071] Closed-loop control execution refers to writing the updated path model into edge computing devices (such as smart bracelets, treadmill controllers) through a low-latency communication protocol (such as MQTT) as the real-time comparison benchmark for the next monitoring cycle. After receiving the new model, the edge device immediately switches to the time node and deviation range of the latest version to dynamically evaluate the real-time collected health indicators. For example, when the edge device receives the "stage 3_v2" model (the time node interval is shortened to 2 days, and the allowable deviation is widened to ±15%), it will use the parameters of this model to calculate the risk coefficient and trigger interventions in subsequent monitoring to ensure that the training path is synchronized with the actual development status.

[0072] Specifically, when the system detects that the actual risk coefficient of a certain adolescent is lower than the target value of the original model for 5 consecutive days (deviation Δ = -0.2), the model parameter correction module calculates through an online learning algorithm that the time node interval for the next stage needs to be shortened from 3 days to 2 days, and the allowable deviation range is adjusted from ±10% to ±12%. The version identification generation module then generates the version number "T4_v1", records the change time, the effective stage (the 4th stage), and the adjusted parameters, and associates and stores them with the risk coefficient data of the previous 30 days in the blockchain through a hashing algorithm. The closed-loop control execution module pushes the "T4_v1" model to the adolescent's smart bracelet through the MQTT protocol. After receiving the model, the bracelet immediately updates the training path parameters stored locally, and uses the new time node (evaluated every 2 days) and the deviation range (±12%) for real-time comparison in subsequent monitoring to ensure that the training progress matches the actual physical improvement speed of the adolescent.

[0073] Compared with the prior art, the traditional preset training path model uses static parameters (such as a fixed 3 days / stage, ±10% deviation), which cannot be dynamically adjusted according to the individual training effect, resulting in problems such as "too slow training progress" or "too strict risk assessment". In this solution, the training path is adaptively optimized through model parameter correction (such as accelerating or slowing down the progress according to the actual risk coefficient), the traceability of model changes is ensured through version identification generation (to avoid evaluation chaos caused by incorrect parameter modification), and seamless connection between model update and monitoring is achieved through closed-loop control execution (the edge device immediately applies the new model). In the prior art, model updates require manual intervention and have a high delay (usually more than 24 hours). In this solution, through online learning and low-latency communication, the model update cycle is shortened to the minute level, and the response speed of parameter adjustment is improved.

[0074] Through the above technical solutions, this application effectively solves the problem of the disconnection between the preset training path model and the actual training effect. Model parameter correction dynamically optimizes the training progress according to real-time data, version identification generation ensures the traceability and security of model changes, and closed-loop control execution realizes the real-time implementation of model updates. The matching degree between the training path and the physical development of adolescents is improved, the misjudgment rate of risk assessment is reduced, and the scientificity and security of adolescent physical exercise are significantly improved. Through the real-time interaction between actual training data and the model, the adaptive adjustment of the training path is realized, and the technical problem that the traditional static model cannot adapt to individual development differences and dynamic training scenarios is solved.

[0075] As Figure 4 shown, this application further proposes that the real-time monitoring system for adolescent physical exercise further includes a biomechanical digital twin module: Generate a dynamic simulation model of bones and muscles in a virtual environment based on real-time three-dimensional joint motion data and plantar pressure distribution data; Predict the stress distribution trend in the next 10 training cycles through finite element analysis and superimpose and display it with the target value of the preset path model; When the predicted trend deviates from the target value by more than the preset deviation threshold, generate a pre-adjustment instruction in advance and push it to the processing terminal.

[0076] Specifically, the system collects real-time three-dimensional joint motion data and plantar pressure distribution data, creates a dynamic simulation model of bones and muscles in a virtual environment (such as a three-dimensional scene simulated by a computer), and intuitively displays the dynamic changes of bones and muscles during movement. With the help of finite element analysis, based on the current motion data and model, predict the stress distribution trend of bones and muscles in the next 10 training cycles (the training cycle here can be a specific time period such as sports training or rehabilitation training). At the same time, the system superimposes and displays the predicted stress distribution trend with the target value of the preset path model (the pre-set and expected stress distribution state to be achieved), and intuitively compares the difference between the predicted trend and the target value. The system sets a preset deviation threshold (that is, the maximum deviation range allowed between the predicted trend and the target value). When the predicted stress distribution trend deviates from the target value by more than this preset deviation threshold, it means that the current motion state is not conducive to the training effect or may cause damage to the body. At this time, the system generates pre-adjustment instructions in advance (such as suggestions for adjusting the training intensity, changing the movement posture, etc.), and pushes these instructions to the processing terminal (such as a computer, mobile phone, smart wearable device, etc., facilitating relevant personnel to obtain and take corresponding measures).

[0077] As a preferred embodiment, the solution of this application is specifically implemented as follows: By deploying a lightweight bone-muscle multibody dynamics model, the virtual bone system is driven in real time based on three-dimensional joint motion data (200Hz) and plantar pressure distribution (500Hz). An improved inverse dynamics algorithm (InverseDynamics) is used to solve the bone joint moments, and the muscle force-length relationship parameters are dynamically adjusted in combination with the activation state of electromyography signals (1000Hz) to achieve biomechanical state synchronization with μs-level accuracy. The finite element analysis (FEA) accelerated by GPU is integrated. Based on the current bone stress distribution and motion trajectory characteristics, stress evolution cloud maps for the next Y (default Y = 30) training cycles are generated through a time series extrapolation algorithm. The adaptive mesh refinement technique (minimum mesh size 0.1mm) is adopted to dynamically optimize the calculation resource allocation, so that the prediction time for a single 30-cycle prediction is controlled within 2 seconds. A pre-adjustment strategy library based on deep reinforcement learning (DRL) is constructed. When the deviation of the predicted stress trend from the target path exceeds the threshold (Δ>0.15), the Monte Carlo tree search (MCTS) algorithm is started, and an optimization instruction set including training intensity correction, action alternative plans, and recovery cycle adjustment is generated within 500ms and pushed to the training terminal in real time through the edge computing node.

[0078] During the operation cycle, deep interaction is achieved through the high-speed data bus and the model interface. The biomechanical sensing array delivers the six-degree-of-freedom pose of the joint, the plantar pressure heat map, and the electromyography activation waveform to the twin engine with microsecond-level accuracy, and the spatio-temporal alignment error of multi-source data is ensured to be <0.5ms through hardware-level clock synchronization (PTP protocol). The virtual bone system updates its dynamic state every 5ms to generate a holographic digital mirror of biomechanical parameters. The finite element engine receives a snapshot of the current bone stress field every 30 seconds, extrapolates the future stress distribution trend through a spatio-temporal convolutional network (ST-Conv), and dynamically superimposes and compares the prediction results with the target path model. When the predicted stress value in the key growth plate area (such as the proximal tibia) exceeds the safety threshold (Ssafe×1.25), the pre-decision process is immediately triggered to generate adjustment instructions for the next 3 training cycles. The pre-adjustment instruction set is directly connected to the edge computing node through the RDMA protocol, and the safety verification (including clinical rule verification and execution effect simulation) is completed within 150ms. The verified instructions are injected into the training control system in the form of a priority queue to ensure that the intervention measures are started 2 training cycles before the target deviation actually occurs.

[0079] Compared with the traditional solution, the traditional solution only monitors the current biomechanical state and cannot predict future risks, resulting in the inability to give early warnings for skeletal stress injury events. This solution advances the risk identification window through finite element extrapolation, improving the early warning success rate. The traditional manual adjustment mode needs to initiate intervention after the injury index appears. This solution realizes automated pre-decision response, advancing the intervention timing. There are calculation errors in joint torque calculation of the traditional inverse dynamics algorithm; this solution compresses the calculation error through a muscle activation compensation mechanism driven by electromyography signals.

[0080] Through the deep coordination of real-time digital twin construction, multi-physical field prediction and deduction, and intelligent advanced decision-making, a training management system with the ability to predict the future has been constructed. This technology breaks through the technical boundaries of the traditional monitoring system, realizes a new management mode of "real-time mirroring - future deduction - advanced control", elevates the scientific training of teenagers from passive response to active prevention in a new dimension, sets a new benchmark for intelligent pre-decision technology in the field of sports health, and realizes the forward-looking prevention and control of training risks and the adaptive optimization of training programs.

[0081] The following is a complete embodiment of the real-time monitoring system for teenagers' physical exercise in this application: A teenager wears an inertial sensor array (sampling rate 200Hz) to accurately record the joint movement trajectories during his basketball training. During the jump shot after a quick stop, the sensor captures in real time the hip abduction angle (peak value 58°±2°), knee flexion angle (maximum 112°±3°), and ankle dorsiflexion angle (dynamic range 25° - 35°), fully presenting the motion posture characteristics. Surface electromyography electrodes (Noraxon DTS system, sampling rate 1500Hz) are attached to the rectus femoris and the lateral head of the gastrocnemius. During the squat training, the system records a delay in the activation of the left rectus femoris (Δt = 18ms) and the maximum contraction intensity of the right gastrocnemius (RMS = 1.2mV), revealing the characteristics of bilateral muscle strength imbalance. A piezoresistive smart insole (Tekscan F-Scan system, 100 sensing points / cm²) monitors the plantar pressure distribution during change-of-direction running. The data shows that the peak pressure in the heel area of the left foot reaches 450 kPa (380 kPa for the right foot), and the forefoot pressure ratio deviation ΔP = 0.23. An EOS low-dose three-dimensional imaging system is used to obtain the development status of the carpal bones, and the BA is determined to be 14.3 years old through the TW3 bone age scoring system, and the skeletal maturity parameters are determined by matching the physiological development database.

[0082] Multi-source data hardware-level synchronization is achieved through the IEEE 1588 PTP protocol (error < 1ms). The dynamic time warping algorithm (DTW) is applied to phase-align the knee joint angle curve (200Hz), the rectus femoris myoelectric envelope (downsampled from 1500Hz to 200Hz), and the heel pressure time series (100Hz). Based on a standard finite element model of a 14-year-old male skeleton (mesh accuracy 0.5mm), joint moment data is input to calculate the dynamic stress distribution in the proximal tibial growth plate area. During continuous jump training, the stress peak is monitored to reach 28MPa (safety threshold Ssafe = 32MPa). The Hilbert-Huang transform (IMF components = 5) is performed on the bilateral rectus femoris myoelectric signals to extract time-frequency features. The co-contraction asymmetry index Masym = 0.18 (Δt = 22ms, intensity ratio R = 0.85) is calculated. The comprehensive health index is calculated using the dynamic weight formula: Hindex = 0.5×(28 / 32) + 0.3×0.18 + 0.2×0.23 = 0.63 (weight coefficients for 14-year-old bone age α = 0.5, β = 0.3, λ = 0.2). The preset training path model shows the target Hindex = 0.68 ± 0.15 in the 3rd week. The current value of 0.63 triggers a secondary deviation warning (Δ = 0.05, lasting for 3 analysis windows). Adjustment instructions are generated: Training intensity adjustment: Increase the squat load coefficient β = 1.1 (original 1.0); Action alternative: Enable the low-impact action library (lunge → wall sit, similarity 89%); Recovery period optimization: Extend the interval between sets to 90s (original 60s).

[0083] Finite element analysis predicts the cumulative trend of tibial stress in the next 5 training sessions (10 cycles), showing that the 8th training will exceed the safety threshold (predicted value 34MPa). The system pushes adjustment suggestions 2 cycles in advance: Introduce a shock-absorbing insole (heel pressure reduced by 18%) and limit the number of continuous jumps ≤ 15 times / group; When the actual Hindex deviates continuously for 3 times > 0.12, the incremental learning Δnew = 0.15×[1 + 0.1×(0.63 - 0.68)] = 0.1425 is triggered, and the allowable deviation Δ is adjusted from 0.15 to 0.1425 after update.

[0084] This embodiment demonstrates the complete application process of the system in adolescent basketball special training. Through the synergistic effect of multi-dimensional biomechanical monitoring, intelligent dynamic adjustment, and advanced risk prediction, accurate monitoring of the synchronization analysis error of the joint-muscle-sole multi-system is achieved. The bone age-driven model reduces the evaluation error of 14-year-old users, advances the sports injury risk identification window, and blockchain encrypted transmission ensures zero leakage of privacy data.

[0085] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements will not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A real-time monitoring system for teenagers' physical exercise, characterized in that, Comprising: A data acquisition module, configured to receive three-dimensional joint motion data, electromyogram signal data, plantar pressure distribution data during the current training stage, and bone age data during the physiological development stage; A standardized feature library, configured to simultaneously obtain a feature dataset of standardized biomechanics during the historical training stage, the feature dataset including joint motion amplitude thresholds, muscle coordination symmetry ranges, and plantar pressure distribution standard values that match the bone age data; A data processing module, configured to extract joint angle feature points from the three-dimensional joint motion data, the activation time series of the electromyogram signal data, and the pressure regions of the plantar pressure distribution data, and perform spatio-temporal alignment processing with the feature dataset to generate feature data; A health index generation module, based on the aligned feature data, calculates a plurality of biomechanical difference parameters reflecting bone stress deviation, muscle coordination asymmetry, and plantar pressure imbalance; Generates a comprehensive health index according to the biomechanical difference parameters and records it as time series data; A dynamic path decision module, including: Dynamically comparing the change trend of the comprehensive health index in the time series data with the change trend of the target health index in the corresponding stage of the preset training path model to generate a deviation direction and amplitude; the deviation direction is divided into positive deviation and negative deviation, the positive deviation is used to increase the training amplitude, and the negative deviation is used to weaken the training amplitude; Generates an adjustment instruction according to the deviation direction and amplitude; An execution module, configured to generate a training parameter adjustment amount including training intensity adjustment, time interval correction, and action alternative according to the adjustment instruction through a preset rule mapping table.

2. The real-time monitoring system for teenagers' physical exercise according to claim 1, characterized in that, Extracting the joint angle feature points includes: Based on a preset kinematic feature template, identifying the key angle values of the knee flexion angle, ankle dorsiflexion angle, and hip abduction angle from the three-dimensional joint motion data; Performing timestamp alignment on the key angle values, and when the angle fluctuation of 5 consecutive frames of data is less than the fluctuation interval, marking it as a stable feature point; Eliminating abnormal angle data that exceeds the amplitude threshold range matching the bone age data.

3. The real-time monitoring system for teenagers' physical exercise according to claim 1, characterized in that, The biomechanical difference parameters include: Bone stress deviation, the bone stress deviation is the relative difference between the real-time bone stress value and the safety threshold matching the bone age data; Muscle coordination asymmetry, the muscle coordination asymmetry is the product of the absolute value of the activation time difference between the left and right target muscle groups and the signal strength ratio; Plantar pressure imbalance, the plantar pressure imbalance is the absolute deviation of the ratio of the lateral heel to forefoot pressure to the standard value.

4. The real-time monitoring system for teenagers' physical exercise according to claim 3, characterized in that The generating of the comprehensive health index includes: Normalizing the bone stress deviation, the muscle coordination asymmetry, and the plantar pressure imbalance; Dynamically adjusting the weight coefficients according to the bone age data, and performing weighted summation to generate the comprehensive health index: Hindex = α×Sdev + β×Masym + λ×|Rp - γ| Where Sdev represents the bone stress deviation, Masym represents the muscle coordination asymmetry, and |Rp - γ| represents the plantar pressure imbalance, and α, β, and λ respectively represent the corresponding weight coefficients, which are dynamically adjusted according to the bone age data.

5. The real-time monitoring system for teenagers' physical exercise according to claim 1, characterized in that The construction method of the preset training path model includes: Obtain experimental data to generate an experimental data set, analyze the experimental data set through the random forest algorithm, and group them to establish the benchmark value of the target health index, and determine the allowable deviation Δ: When the bone age of the experimental data is 12 - 14 years old: the allowable deviation Δ = 0.3; When the bone age of the experimental data is 15 - 18 years old: the allowable deviation Δ = 0.2; When the actual health index exceeds the allowable deviation Δ for 5 consecutive times, trigger the model self-learning mechanism and update the allowable deviation Δ.

6. The real-time monitoring system for teenagers' physical exercise according to claim 5, characterized in that The dynamic comparison and generation of the adjustment instruction include: Apply sliding window analysis to the comprehensive health index in the time series data, calculate the actual change rate of the comprehensive health index within the window period, and compare it with the target change rate in the corresponding stage of the preset training path model to generate the deviation direction and amplitude; When the actual change rate continuously exceeds the allowable deviation range of the target change rate, generate a hierarchical adjustment instruction according to the deviation direction.

7. The real-time monitoring system for teenagers' physical exercise according to claim 6, characterized in that, The trigger logic of the hierarchical adjustment instruction includes: Obtain the deviation direction, and judge whether the deviation direction is the positive deviation. The positive deviation means that the actual change rate is greater than the sum of the target change rate and the allowable deviation Δ. If so, make the following decisions according to the positive deviation amplitude: When the amplitude of the positive deviation is within the first threshold interval, generate an instruction to shorten the training interval; When the amplitude of the positive deviation is within the second threshold interval, generate an instruction to advance to the next training stage and increase the intensity; Otherwise, it is judged as the negative deviation. The negative deviation means that the actual change rate is less than the difference between the target change rate and the allowable deviation Δ. Make the following decisions according to the negative deviation amplitude: When the amplitude of the negative deviation is within the first threshold interval, generate an instruction to extend the training interval; When the amplitude of the negative deviation is within the second threshold interval, generate an instruction to forcibly switch to the alternative action library, and the alternative action library is selected from the historical low-risk actions.

8. The real-time monitoring system for teenagers' physical exercise according to claim 1, characterized in that, The real-time monitoring system for adolescent physical exercise also includes a path model update module, which is used to update the preset training path model according to the training parameter adjustment amount to maintain the comprehensive health index within the allowable deviation range of the training path model; Among them, the preset training path model includes the target health index, training cycle time nodes and allowable deviation range preset for each bone age stage, as the dynamic benchmark for the actual training process.

9. The real-time monitoring system for teenagers' physical exercise according to claim 8, characterized in that Updating the preset training path model includes: Model parameter correction, which adjusts the time node interval and allowable deviation range of the subsequent stage according to the actual health index trend; Version identification generation, which records the model change time, effective stage identification and adjusted deviation range, and associates them with the historical training data; Closed-loop control execution, which writes the updated path model into the edge computing device as the real-time comparison benchmark for the next monitoring cycle.

10. The real-time monitoring system for teenagers' physical exercise according to claim 1, characterized in that, The real-time monitoring system for adolescent physical exercise also includes a biomechanical digital twin module: Generate a dynamic simulation model of bones and muscles in a virtual environment based on the real-time three-dimensional joint motion data and the plantar pressure distribution data; Predict the stress distribution trend in the next 10 training cycles through finite element analysis and superimpose and display it with the target value of the preset path model; When the predicted trend deviates from the target value by more than the preset deviation threshold, generate a pre-adjustment instruction in advance and push it to the processing terminal.

Images