A multi-modal adaptive student physical fitness training method and training system

Abstract

This invention discloses a multimodal adaptive student physical fitness training method and system, comprising: synchronously capturing multi-dimensional physiological and exercise-related data during student physical fitness training; deeply integrating and analyzing the data to mine its inherent correlations and generate feature information that comprehensively reflects physical fitness status; accurately locating weak links in various dimensions of physical fitness based on this feature information; clarifying physical fitness development trends by combining historical training data; and generating training parameter configurations suitable for individuals; providing real-time feedback on data changes during training and dynamically adjusting training parameters to form a closed-loop training chain. This method, through a multi-step detailed feature analysis, weakness diagnosis, trend prediction, and parameter optimization process, solves the shortcomings of traditional training methods, such as incomplete data analysis and a lack of personalization and dynamic adjustment capabilities, thus improving the accuracy, adaptability, and efficiency of physical fitness training. It is suitable for physical fitness improvement training for various student groups.

Description

Technical Field

[0001] This invention relates to the field of student physical training technology, and in particular to a multimodal adaptive student physical training method and training system. Background Technology

[0002] In the current context of comprehensively promoting quality education, students' physical fitness development has become an important dimension of talent cultivation. Traditional, standardized, and experience-based physical training models are no longer suitable for the differences in students' individual physiological conditions and physical fitness levels. During physical training, it is necessary to simultaneously acquire multi-dimensional physiological and athletic performance data of students, systematically analyze this data to identify strengths and weaknesses in physical development, formulate appropriate training plans based on the laws of students' physical development, and dynamically adjust the plans according to real-time conditions during training to achieve targeted and efficient physical training. Under this requirement, an integrated training model encompassing multi-dimensional data collection, comprehensive analysis, personalized plan generation, and dynamic optimization has become a key direction for improving the quality of students' physical training and provides a practical foundation for the research and development of related training methods and systems.

[0003] Existing technologies have two core drawbacks: First, data collection and analysis lack comprehensiveness and depth, focusing only on single types of data or simply integrating multi-source data without fully exploring the inherent relationships between different data. This results in inaccurate assessments of students' physical fitness and an inability to clearly identify specific weaknesses in their physical development. Second, training programs lack personalization and dynamic adjustment capabilities. They fail to scientifically plan based on students' past training experiences and physical development trends, and lack an effective real-time feedback and adjustment mechanism during training. This prevents key elements such as training content and intensity from being adapted and optimized according to real-time changes in students' physical fitness, affecting training effectiveness and efficiency. Summary of the Invention

[0004] In order to overcome the shortcomings and deficiencies of existing technologies, this invention provides a multimodal adaptive student physical fitness training method and training system.

[0005] The technical solution adopted in this invention is a multimodal adaptive student physical fitness training method, comprising the following steps: S1, synchronously capturing heart rate variability signals, electromyography signals, respiratory rate signals, blood oxygen saturation signals, and motion posture biomechanical signals during student physical fitness training through a multimodal physiological signal acquisition module, constructing a multi-dimensional physiological-motor fusion signal set; S2, inputting the fusion signal set into a multimodal physiological signal parsing convolutional network, and generating a high-dimensional feature vector matrix by mining the temporal-frequency-spatial domain correlation features of the signal through deep feature extraction links; S3, based on the high-dimensional feature vector matrix, calling the student physical fitness deficiency diagnosis and assessment model, and locating the deficiency through a multi-level feature matching and weight allocation mechanism. S4. Using a student's physical fitness trajectory prediction algorithm, combined with historical training data and current physical fitness weakness characteristics, a time-series prediction model is constructed to output physical fitness development trend data within a preset future period; S5. The physical fitness weakness data and development trend data are input into an adaptive physical fitness training intelligent optimization analysis platform, which generates training intensity, training duration, movement combinations, and interval period parameter configurations suitable for the individual through a multi-objective constraint optimization mechanism; S6. Based on the parameter configuration, a multimodal adaptive student physical fitness training execution process is initiated, providing real-time feedback on signal changes during the training process and dynamically adjusting the training parameter configuration to form a closed-loop training link.

[0006] Furthermore, the feature extraction expression of the multimodal physiological signal parsing convolutional network is as follows: ,in, It is a high-dimensional eigenvector matrix. For improved activation functions, These are the one-dimensional, two-dimensional, and three-dimensional convolution kernel weight matrices, respectively. These are the bias vectors for each convolutional layer. It is a collection of multimodal physiological-motor fusion signals. represents the weight coefficients of the attention mechanism, Conv1D, Conv2D, and Conv3D represent one-dimensional, two-dimensional, and three-dimensional convolution operations, respectively, and Attention(⋅) is the cross-modal attention calculation function.

[0007] Furthermore, the expression for locating weaknesses in the student physical fitness deficiency diagnostic assessment model is as follows: ,in, Quantify the assessment vector for physical fitness weaknesses. The number of feature dimensions, For the first Importance weights of class features For the first High-dimensional feature vectors This is a standard physical fitness feature template vector. This is the function for calculating feature similarity. For feature matching coefficients, The number of dimensions for physical fitness assessment. For the first Diagnostic coefficient of individual capacity dimension For the first Feature components of the individual energy dimension For the first Threshold vector of individual energy dimension, This is a logical activation function.

[0008] Furthermore, the trend output expression of the student physical fitness growth trajectory prediction algorithm is as follows: for Predicted physical growth at any given time For Long Short-Term Memory (LSTM) network computation functions, This is the high-dimensional feature vector at the current time. for Predicted physical fitness values ​​at any given time. For time-series dependency coefficients, For the weights of the gated recurrent units, The function is used to calculate the gated loop unit. Let this be the vector of the weakest physical condition at the current moment. for The hidden state vector at time step 1. For dynamic adjustment coefficients, The number of historical data windows. For the first The weighting coefficients of each historical window, For the first Training data vectors for each historical window, It is the hyperbolic tangent activation function.

[0009] Furthermore, the parameter configuration generation expression of the adaptive physical training intelligent optimization analysis platform is as follows: ,in, Configure vectors for training parameters. This is the computation function for the non-dominated sorting genetic algorithm III. To train the set of constraints, For feature fusion operators, Principal component analysis function, For element-wise product operators, To optimize the coefficients for the parameters, For training intensity-duration adaptation matrix, For the weighting coefficients of the action combination, Configure vectors for intermittent periods.

[0010] Furthermore, the closed-loop adjustment expression for the multimodal adaptive student physical training is: ,in, This is a dynamically adjusted set of training parameters. This is a function for calculating real-time signal feedback. For real-time multimodal signals during the training process, Configure vectors for initial training parameters. For feedback adjustment coefficients, A function for calculating feature differences. For real-time high-dimensional feature vectors, As the initial high-dimensional feature vector, Update the operator for the parameter. Update function for parameters, To adjust the step size coefficient.

[0011] Further, S3 includes the following sub-steps: S31, splitting the high-dimensional feature vector matrix according to the feature dimensions corresponding to endurance, strength, speed, agility, and flexibility to form a unique feature subset for each physical fitness dimension. Each subset includes the temporal correlation feature, frequency response feature, and spatial distribution feature under that dimension; S32, calling the feature matching layer of the student physical fitness weakness diagnosis and assessment model, comparing the unique feature subset for each physical fitness dimension with the preset standard physical fitness feature template library layer by layer, and filtering out feature items that deviate from the standard range by calculating feature distance and similarity value; S33, based on the filtered deviating feature items, assigning corresponding influence weights to each deviating feature item through the model's weight allocation mechanism. The weight allocation is dynamically allocated according to the correlation degree of the feature with physical fitness performance and the reliability of the signal; S34, integrating the quantitative assessment results of the weaknesses of each physical fitness dimension according to the number of deviating feature items, the degree of deviation, and the corresponding influence weights, and clarifying the priority ranking of each weakness in the overall physical fitness system.

[0012] Further, S4 includes the following sub-steps: S41, collecting students' multimodal physiological signal data, physical fitness test results data, and training execution parameter data from previous preset periods, organizing them according to time series, and constructing a historical training dataset, which includes feature vectors for each time node and corresponding physical fitness status labels; S42, fusing the historical training dataset with the quantitative assessment results of physical fitness weaknesses output in S3 to generate the input dataset for the time-series prediction model, ensuring data consistency through feature splicing and dimension alignment during the fusion process; S43, calling the student physical fitness growth trajectory prediction algorithm, inputting the input dataset into the algorithm's time-series feature extraction module, mining long-term dependencies and short-term fluctuation patterns in the data, and constructing a time-series prediction model for physical fitness development; S44, through the model's prediction output module, based on the constructed time-series prediction model, outputting predicted values ​​of physical fitness status at different time nodes within the future preset period, forming complete physical fitness growth trajectory data.

[0013] Further, S5 includes the following sub-steps: S51, inputting the quantitative assessment results of physical weakness items and the predicted data of physical growth trajectory into the constraint setting module of the adaptive physical training intelligent optimization analysis platform, setting multi-dimensional constraint parameters such as the upper limit of training intensity, the range of movement difficulty levels, and the threshold of the interval cycle; S52, the platform's optimization objective construction module constructs a multi-objective optimization function based on the input data and constraint parameters, with the main objectives being to maximize the efficiency of improving weakness items, balance the training load, and rationalize the training cycle; S53, calling the platform's optimization algorithm execution module, iteratively optimizing the training intensity, training duration, movement combination, and interval cycle parameters through the multi-objective constraint optimization mechanism, generating multiple sets of candidate parameter configuration schemes; S54, through the platform's scheme evaluation and screening module, comprehensively evaluating the candidate parameter configuration schemes according to preset evaluation indicators, and selecting the optimal training parameter configuration scheme as the output result.

[0014] A multimodal adaptive student physical fitness training system, applied to a multimodal adaptive student physical fitness training method, includes: a multimodal physiological-motor signal synchronous acquisition unit, a high-dimensional feature intelligent analysis unit, a physical fitness deficiency accurate diagnosis unit, a growth trajectory time-series prediction unit, a training parameter adaptive optimization unit, and a closed-loop training execution and adjustment unit. Each unit is bidirectionally connected via a data bus and a control bus. The multimodal physiological-motor signal synchronous acquisition unit captures heart rate variability, electromyography, respiratory rate, blood oxygen saturation, and movement posture biomechanical signals during student physical fitness training and transmits them to the high-dimensional feature intelligent analysis unit. The high-dimensional feature intelligent analysis unit extracts features from the received signals using a multimodal physiological signal analysis convolutional network to generate a high-dimensional feature vector matrix. The data is then sent to the Physical Fitness Shortcomings Precision Diagnosis Unit; based on a high-dimensional feature vector matrix, the Physical Fitness Shortcomings Precision Diagnosis Unit locates the weak points in physical fitness through the student physical fitness shortcomings diagnosis and assessment model and transmits the results to the Growth Trajectory Time Series Prediction Unit; the Growth Trajectory Time Series Prediction Unit uses the student physical fitness growth trajectory prediction algorithm to output physical fitness development trend data, which, together with the weakness data, is sent to the Training Parameter Adaptive Optimization Unit; the Training Parameter Adaptive Optimization Unit generates the optimal training parameter configuration through the adaptive physical fitness training intelligent optimization analysis platform and transmits it to the closed-loop training execution and adjustment unit; the closed-loop training execution and adjustment unit starts the training process based on the parameter configuration, collects signal changes during the training process in real time and feeds them back to the high-dimensional feature intelligent analysis unit, dynamically adjusts the training parameters to conduct multimodal adaptive physical fitness training.

[0015] Beneficial Effects: This invention proposes a multimodal adaptive student physical fitness training method and system. Through simultaneous multi-dimensional data acquisition and in-depth integration and analysis, it overcomes the shortcomings of existing technologies in data acquisition and analysis, which lack comprehensiveness and depth. This method simultaneously captures multiple types of data related to students' physiological state and athletic performance, fully exploring the intrinsic correlations between different data points to achieve accurate judgment of students' physical fitness status and clearly define specific weaknesses in dimensions such as endurance and strength, providing a scientific basis for subsequent training. Simultaneously, by combining students' historical training data with physical fitness development patterns to formulate personalized training plans and constructing a real-time feedback and adjustment mechanism during the training process, it completely solves the problem of existing training plans lacking personalization and dynamic adjustment capabilities. Key elements such as training content and intensity can be adapted and optimized according to real-time changes in students' physical fitness status. This not only significantly improves the targeting and efficiency of physical fitness training but also steadily promotes the continuous improvement of students' physical fitness according to their own development trajectory, taking into account the scientific nature and adaptability of training, meeting the individual differences of different students, and comprehensively improving the overall quality and effect of physical fitness training. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating the overall process of the method of the present invention.

[0017] Figure 2 This is a flowchart of method step S3 of the present invention;

[0018] Figure 3 This is a flowchart of method step S4 of the present invention;

[0019] Figure 4 This is a flowchart of step S5 of the method of the present invention;

[0020] Figure 5 This is a diagram showing the system unit composition of the present invention. Detailed Implementation

[0021] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. The application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0022] like Figure 1 As shown, a multimodal adaptive student physical fitness training method includes the following steps: S1, synchronously capturing heart rate variability signals, electromyography signals, respiratory rate signals, blood oxygen saturation signals, and motion posture biomechanical signals during student physical fitness training through a multimodal physiological signal acquisition module, constructing a multi-dimensional physiological-motor fusion signal set; S2, inputting the fusion signal set into a multimodal physiological signal parsing convolutional network, mining the temporal-frequency-spatial domain correlation features of the signals through deep feature extraction links, and generating a high-dimensional feature vector matrix; S3, based on the high-dimensional feature vector matrix, calling the student physical fitness deficiency diagnosis and assessment model, and locating the student's endurance deficiency through a multi-level feature matching and weight allocation mechanism. S4. Using a student physical fitness growth trajectory prediction algorithm, combined with historical training data and current physical fitness weakness characteristics, a time-series prediction model is constructed to output physical fitness development trend data within a preset future period; S5. The physical fitness weakness data and development trend data are input into an adaptive physical fitness training intelligent optimization analysis platform, which generates training intensity, training duration, movement combination, and interval period parameter configurations suitable for individuals through a multi-objective constraint optimization mechanism; S6. Based on the parameter configuration, a multimodal adaptive student physical fitness training execution process is initiated, providing real-time feedback on signal changes during the training process and dynamically adjusting the training parameter configuration to form a closed-loop training link.

[0023] Step S1 involves multi-dimensional data synchronous acquisition via a multimodal physiological signal acquisition module. This module integrates a heart rate variability sensor, an electromyography (EMG) sensor, a respiratory rate sensor, a blood oxygen saturation sensor, and a motion posture biomechanics sensor. Each sensor's sampling frequency is set to 200 Hz to 500 Hz to ensure timely and complete signal acquisition. During acquisition, the heart rate variability sensor is attached to the student's chest to capture continuous heart rate interval time-series data; the EMG sensor is attached to the surface of the student's main exerting muscle groups in the limbs to collect bioelectrical signals during muscle contraction; the respiratory rate sensor is fixed to the student's abdomen to convert changes in chest cavity movement into respiratory cycle data; the blood oxygen saturation sensor is worn on the student's fingertips to capture the percentage of oxyhemoglobin in the blood in real time; and the motion posture biomechanics sensor is integrated into the joints of the student's athletic shoes and clothing to collect acceleration, angular velocity, and pressure distribution data during running, jumping, and stretching movements. Each sensor uploads data to the data acquisition terminal in real time via a wireless transmission protocol. The terminal performs timestamp alignment on the data to construct a multi-dimensional physiological-motor fusion signal set, which includes five types of signals, each with a duration of 10 to 30 minutes. This provides comprehensive and synchronous raw data support for subsequent feature analysis.

[0024] Step S2 inputs the constructed multi-dimensional physiological-motor fusion signal set into a preset multimodal physiological signal parsing convolutional network. This network includes an input layer, three convolutional layers, two pooling layers, one feature fusion layer, and an output layer. The kernel sizes of each convolutional layer are set to 3×3, 5×5, and 7×7, respectively, with a stride of 1 and SAME padding. The pooling layer uses max pooling with a kernel size of 2×2 and a stride of 2. During network operation, the input layer receives the fused signal set and converts it into a standardized input format. The first convolutional layer performs one-dimensional convolution operations on the input data to extract the basic temporal features of the signal. The second convolutional layer uses two-dimensional convolution operations to mine the temporal-frequency domain correlation features of the signal. The third convolutional layer uses three-dimensional convolution operations to capture the temporal-frequency-spatial domain cross features of the signal. An activation operation is performed after each convolutional operation to enhance the nonlinear expressive power of the features. The pooling layer downsamples the convolutional feature map to retain key features and reduce data dimensionality. The feature fusion layer concatenates and assigns weights to the feature maps output by each convolutional layer. The weight coefficients are set to 0.2 to 0.4 depending on the importance of different features to physical fitness assessment. Finally, the output layer generates a high-dimensional feature vector matrix with dimensions of 128 to 256. This matrix condenses the key information in the fused signal and provides accurate feature input for the diagnosis of physical fitness shortcomings.

[0025] Step S3, based on the high-dimensional feature vector matrix generated in step S2, calls the preset student physical fitness deficiency diagnosis and assessment model, which includes a feature splitting module, a feature matching module, a weight allocation module, and a result integration module. The feature decomposition module divides the high-dimensional feature vector matrix into five exclusive feature subsets based on five physical fitness dimensions: endurance, strength, speed, agility, and flexibility. Each subset includes 20 to 50 feature components corresponding to that dimension. The feature matching module compares each exclusive feature subset with a pre-set standard physical fitness feature template library, which includes standard feature data for students of different ages and genders. The comparison process combines feature distance calculation and similarity analysis, setting a similarity threshold of 0.7 to filter out deviation features with similarity below the threshold. The weight allocation module assigns each deviation feature a weight of 0.1 to 0.3 based on its impact on the performance of each physical fitness dimension. The degree of impact is determined through statistical analysis of historical training data. The results integration module performs weighted calculations on the number, degree of deviation, and corresponding impact weights of deviation features for each physical fitness dimension, generating quantitative assessment results for weaknesses in each physical fitness dimension. It clarifies the quantitative score of each weakness and its priority ranking among the five physical fitness dimensions, providing precise target guidance for the subsequent training program development.

[0026] Step S4 uses a student physical fitness growth trajectory prediction algorithm, combined with the quantitative assessment results of physical fitness weaknesses output in step S3 and the student's historical training data. The algorithm's preset historical data window length is 3 to 12 months, the time interval is 1 week, and the prediction period is set to 1 to 6 months. First, multimodal physiological signal data, physical fitness test results, and training execution parameter data of students within corresponding time windows are collected and sorted according to time series to construct a historical training dataset containing 12 to 48 sets of data, each set of data associated with a corresponding physical fitness status label. Then, the historical training dataset and the quantitative assessment results of physical fitness weaknesses are input into the algorithm's time-series feature extraction module. This module uses a sliding window technique to extract long-term dependency features and short-term fluctuation features from the data, with the sliding window size set to 3 to 5 time nodes. Based on the extracted time-series features, the algorithm constructs a time-series prediction model. The model learns the changing patterns of physical fitness status in historical data and combines this with current weakness features to predict future physical fitness development trends. Finally, it outputs the predicted physical fitness status values ​​for each month within the future prediction period, forming physical fitness growth trajectory data including five dimensions: endurance, strength, speed, agility, and flexibility, providing a forward-looking basis for training parameter optimization.

[0027] Step S5 inputs the quantitative assessment results of physical weaknesses from Step S3 and the predicted data of physical growth trajectory from Step S4 into the adaptive physical training intelligent optimization analysis platform. This platform includes a constraint setting module, an optimization goal construction module, an optimization algorithm execution module, and a scheme evaluation and screening module. The constraint setting module sets the upper limit of training intensity to 60% to 80% of the student's maximum exercise load, the movement difficulty level range to 1 to 5, the interval period threshold to 30 seconds to 3 minutes, and sets a constraint parameter that the total training time does not exceed 90 minutes. The optimization goal construction module constructs a multi-objective optimization function with the core objectives of improving the efficiency of weak points, balancing the training load, and ensuring the rationality of the training cycle. The weights of each objective are set to 0.4, 0.3, and 0.3, respectively. The optimization algorithm execution module adopts a multi-objective constraint optimization mechanism to optimize training intensity, training duration, movement combinations, and... The parameters such as the interval period are iteratively optimized, with the number of iterations set to 100 to 200 times and the convergence threshold set to 0.001, generating 3 to 5 sets of candidate parameter configuration schemes. The scheme evaluation and screening module comprehensively evaluates the candidate schemes based on preset evaluation indicators, including the potential for improvement of weaknesses, training load adaptability, and scientific nature of the action combination. Finally, the optimal training parameter configuration scheme with the highest comprehensive score is selected. This scheme clearly defines the number of executions of each type of training action, the duration of each set, the interval time between sets, and the overall training process, providing a specific basis for training execution.

[0028] Step S6, based on the optimal training parameter configuration scheme output in step S5, initiates the multimodal adaptive student physical fitness training execution process. The training execution equipment includes intelligent sports equipment, wearable monitoring devices, and a central control terminal, which achieve real-time data interaction through a wireless communication protocol. After training begins, the intelligent exercise equipment automatically adjusts operating parameters such as training intensity and resistance level according to the parameter configuration scheme. The wearable monitoring device continues the acquisition frequency of step S1, capturing multimodal physiological-motor signals such as heart rate variability, electromyography, and respiratory rate in real time during training, and transmitting data to the central control terminal every 5 seconds. After receiving the real-time signal, the central control terminal synchronously calls the multimodal physiological signal parsing convolutional network to extract real-time high-dimensional feature vectors, compares and analyzes them with the initial feature vectors generated in step S2, and calculates the feature difference degree. The training parameter configuration is dynamically adjusted according to the feature difference degree. When the difference degree is higher than 0.2, the training intensity is reduced or increased according to the preset adjustment step size of 0.1, the interval period is adjusted by increasing or decreasing by 10 seconds to 30 seconds, and the movement combination is optimized in sequence according to the improvement of weak points. Through real-time data feedback and dynamic parameter adjustment, a closed-loop training link of "acquisition-analysis-evaluation-adjustment" is formed to ensure that the training process is always adapted to the student's real-time physical condition and maximize the training effect.

[0029] Preferably, the feature extraction expression of the multimodal physiological signal parsing convolutional network is: ,in, It is a high-dimensional eigenvector matrix. For improved activation functions, These are the one-dimensional, two-dimensional, and three-dimensional convolution kernel weight matrices, respectively. These are the bias vectors for each convolutional layer. It is a collection of multimodal physiological-motor fusion signals. represents the weight coefficients of the attention mechanism, Conv1D, Conv2D, and Conv3D represent one-dimensional, two-dimensional, and three-dimensional convolution operations, respectively, and Attention(⋅) is the cross-modal attention calculation function.

[0030] Specifically, the feature extraction process of the multimodal physiological signal parsing convolutional network is derived based on the multidimensional correlation characteristics of multimodal signals in the time, frequency, and spatial domains. Since single-dimensional convolution cannot fully exploit signal cross-features, a feature extraction architecture using one-dimensional, two-dimensional, and three-dimensional convolutional layers is adopted. First, one-dimensional convolution captures basic time-domain features; then, two-dimensional convolution mines time-frequency domain correlation features; and finally, three-dimensional convolution captures global cross-features. Simultaneously, a cross-modal attention mechanism is introduced to strengthen the weights of key signals. During the derivation process, the initial value range of the weight matrix for each convolutional layer was determined through extensive experimental statistics. The convolutional kernel weight matrix was set to a numerical range suitable for physiological-motor signals based on signal type differences. The bias vector value fluctuated slightly around 0 to avoid initial output offset. The attention mechanism weight coefficients were determined to be between 0.3 and 0.6 after multiple iterations of optimization. An improved activation function adapted to the feature distribution of multimodal signals was selected to ensure the non-linear expressive power of feature extraction. In practice, the multimodal fused signal is first input into the network, and convolution operations are performed sequentially according to the preset kernel size, stride and padding method. After each round of convolution, the corresponding bias vector is superimposed and processed by the activation function. Then, the signal is reduced in dimensionality by the pooling layer. Finally, the cross-modal features are dynamically weighted and fused through the attention mechanism to output a high-dimensional feature vector matrix. The parameter values ​​and derivation logic ensure that the physical fitness-related features in the multimodal signal can be extracted comprehensively and accurately, providing reliable input for subsequent diagnosis and assessment.

[0031] Preferably, the expression for locating weaknesses in the student physical fitness deficiency diagnostic assessment model is as follows: ,in, Quantify the assessment vector for physical fitness weaknesses. The number of feature dimensions, For the first Importance weights of class features For the first High-dimensional feature vectors This is a standard physical fitness feature template vector. This is the function for calculating feature similarity. For feature matching coefficients, The number of dimensions for physical fitness assessment. For the first Diagnostic coefficient of individual capacity dimension For the first Feature components of the individual energy dimension For the first Threshold vector of individual energy dimension, This is a logical activation function.

[0032] Specifically, the weakness location expression of the student physical fitness deficiency diagnosis and assessment model is derived based on the principle of feature similarity matching and weighted quantization. Since physical fitness weaknesses need to be judged comprehensively by combining the degree of feature deviation and the weight of influence, a framework combining feature similarity calculation and weight allocation is constructed. Feature items deviating from the standard are screened through feature distance and similarity analysis, and then weights are assigned according to the degree of influence and quantified for assessment. In the derivation process, the number of feature dimensions is determined to be 15 to 25 based on the core indicators of physical fitness assessment. The feature importance weight is set to a range of 0.1 to 0.5 based on historical data statistical analysis of the correlation between different features and physical fitness performance. The feature matching coefficient is adjusted to 0.6 to 0.9 based on signal reliability. The physical fitness assessment dimensions are fixed at five core dimensions: endurance, strength, speed, agility, and flexibility. The diagnostic coefficient is calibrated experimentally to 0.8 to 1.2. The threshold vector is set with reference to standard physical fitness data for students of different age groups. A logical activation function is used to map the assessment results to the 0-1 interval for quantification. During implementation, the high-dimensional feature vector matrix is ​​first split to obtain a subset specific to each physical fitness dimension. After comparison with the standard template library, the deviating feature items are selected and weighted according to the preset weight coefficients. The quantitative evaluation vector is output through the activation function. The parameter values ​​are verified through a large number of samples to ensure that the weak items of each physical fitness dimension can be accurately located, providing a clear goal for the formulation of the training plan.

[0033] Preferably, the trend output expression of the student physical fitness growth trajectory prediction algorithm is: for Predicted physical growth at any given time For Long Short-Term Memory (LSTM) network computation functions, This is the high-dimensional feature vector at the current time. for Predicted physical fitness values ​​at any given time. For time-series dependency coefficients, For the weights of the gated recurrent units, The function is used to calculate the gated loop unit. Let this be the vector of the weakest physical condition at the current moment. for The hidden state vector at time step 1. For dynamic adjustment coefficients, The number of historical data windows. For the first The weighting coefficients of each historical window, For the first Training data vectors for each historical window, It is the hyperbolic tangent activation function.

[0034] Specifically, the trend output calculation process of the student physical fitness growth trajectory prediction algorithm is derived based on the long-term dependence and short-term fluctuation characteristics of time-series data. Because physical fitness growth has both temporal continuity and dynamic changes, a single prediction model cannot simultaneously account for both long-term trends and short-term fluctuations. Therefore, the advantages of Long Short-Term Memory (LSTM) networks and gated recurrent units are integrated, while historical window data product terms are introduced to strengthen the influence of cyclical patterns. In the derivation process, the number of historical data windows is determined to be 6 to 12 based on the characteristics of physical fitness training cycles. The temporal dependence coefficient is set to 0.4 to 0.7 based on statistical analysis of the temporal correlation strength of physical fitness changes. The weight of the gated recurrent unit is set to 0.3 to 0.5 based on the optimization experience of time-series prediction models. The dynamic adjustment coefficient is set to 0.2 to 0.4 based on individual student differences. The historical window weight coefficient is allocated according to the time decay law, with recent data having a higher weight than older data. A hyperbolic tangent activation function is used to constrain the output range of the historical data product term. During implementation, historical training data within a preset period is first collected and organized according to time series. This data is then integrated with the current weak point features and input into the algorithm. Long Short-Term Memory (LSTM) networks are used to mine long-term dependency features, and gated recurrent units capture short-term fluctuation patterns. Historical window data is superimposed after weight allocation and activation function processing, and the predicted physical fitness values ​​at each time point are output. The logic and parameter values ​​of this formula are fully adapted to the temporal characteristics of physical fitness growth, and can accurately predict future physical fitness development trends, providing a forward-looking basis for training optimization.

[0035] Preferably, the parameter configuration generation expression of the adaptive physical training intelligent optimization analysis platform is: ,in, Configure vectors for training parameters. This is the computation function for the non-dominated sorting genetic algorithm III. To train the set of constraints, For feature fusion operators, Principal component analysis function, For element-wise product operators, To optimize the coefficients for the parameters, For training intensity-duration adaptation matrix, For the weighting coefficients of the action combination, Configure vectors for intermittent periods.

[0036] Specifically, the parameter configuration generation of the adaptive physical training intelligent optimization analysis platform is derived based on multi-objective constraint optimization theory. Since the training parameters need to simultaneously meet multiple objectives such as improving weaknesses, balancing workload, and ensuring reasonable cycles, the non-dominated sorting genetic algorithm III is used as the core optimization algorithm. Principal component analysis is combined to simplify feature dimensions, and parameter adaptation is achieved through feature fusion and element-wise multiplication. During the derivation process, the set of training constraints is determined based on students' physical endurance and exercise safety standards, including key constraints such as the upper limit of training intensity and the range of movement difficulty. The parameter optimization coefficients are experimentally calibrated to 0.5 to 0.8. The training intensity-duration adaptation matrix is ​​set with reference to the principles of exercise physiology. The weight coefficients of movement combinations are allocated from 0.2 to 0.5 according to the need to improve weaknesses. The interval cycle configuration vector is set according to the corresponding movement type and intensity. During implementation, the data on weak points and trend predictions are first input into the platform. A multi-objective optimization function and constraints are set, and the non-dominated sorting genetic algorithm III is used for 100 to 200 iterations to find the optimal solution. In each iteration, the optimal solution is selected according to the fitness function. The features after dimensionality reduction by principal component analysis, the fitness matrix, and the weight coefficients are fused to generate multiple sets of candidate parameter configurations. After evaluation and selection, the optimal solution is output. This formula, through the derivation logic of multi-objective optimization and feature fusion, combined with scientific parameter values, ensures that the generated training parameter configuration is both targeted and feasible.

[0037] Preferably, the closed-loop adjustment expression for the multimodal adaptive student physical training is: ,in, This is a dynamically adjusted set of training parameters. This is a function for calculating real-time signal feedback. For real-time multimodal signals during the training process, Configure vectors for initial training parameters. For feedback adjustment coefficients, A function for calculating feature differences. For real-time high-dimensional feature vectors, As the initial high-dimensional feature vector, Update the operator for the parameter. Update function for parameters, To adjust the step size coefficient.

[0038] Specifically, the closed-loop adjustment expression for multimodal adaptive student physical training is derived based on the principles of real-time feedback and dynamic adaptation. Because students' physical condition changes in real time during training, parameters need to be dynamically adjusted through a feedback mechanism. Therefore, three main modules are introduced: real-time signal feedback calculation, feature difference analysis, and parameter update, constructing the closed-loop adjustment logic. During the derivation, the feedback adjustment coefficient is set to 0.1 to 0.3 based on the signal feedback sensitivity requirements. The feature difference calculation function adopts a measurement method adapted to multimodal feature changes, and the adjustment step size coefficient is experimentally determined to be 0.05 to 0.15 to ensure the smoothness and timeliness of parameter adjustment. During implementation, multimodal signals are collected in real time during training and input into the feedback calculation module. These signals are compared and analyzed with the initial training parameter configuration to calculate the difference between real-time and initial features. When the difference exceeds a preset threshold, the training parameters are dynamically updated according to the feedback adjustment coefficient and the adjustment step size coefficient. The update process uses a parameter update function to adapt and adjust key elements such as training intensity and interval periods. Element-wise multiplication ensures that the adjusted parameters accurately match the real-time physical condition. The formula takes into full account the dynamic nature of the training process. The parameter values ​​have been verified through extensive practice. It can quickly respond to changes in students' physical condition and ensure that the training parameters are always in the optimal fit through a closed-loop adjustment mechanism, thereby maximizing the training effect.

[0039] Preferred, such as Figure 2 As shown, step S3 includes the following sub-steps: S31, splitting the high-dimensional feature vector matrix according to the feature dimensions corresponding to endurance, strength, speed, agility, and flexibility to form a unique feature subset for each physical fitness dimension. Each subset includes the temporal correlation feature, frequency response feature, and spatial distribution feature under that dimension; S32, calling the feature matching layer of the student physical fitness weakness diagnosis and assessment model, comparing the unique feature subset for each physical fitness dimension with the preset standard physical fitness feature template library layer by layer, and filtering out feature items that deviate from the standard range by calculating feature distance and similarity value; S33, based on the filtered deviating feature items, assigning corresponding influence weights to each deviating feature item through the model's weight allocation mechanism. The weight allocation is dynamically allocated according to the correlation degree of the feature with physical fitness performance and the reliability of the signal; S34, integrating the number, degree of deviation, and corresponding influence weights of the deviating feature items to form a quantitative assessment result of the weaknesses in each physical fitness dimension, and clarifying the priority ranking of each weakness in the overall physical fitness system.

[0040] Specifically, step S3 includes four sub-steps, S31 to S34, which are closely linked to accurately pinpoint physical weaknesses. S31 first splits the high-dimensional feature vector matrix generated in step S2 into five core physical dimensions: endurance, strength, speed, agility, and flexibility. Each dimension corresponds to a feature subset containing 20 to 35 feature components, including temporal correlation features, frequency response features, and spatial distribution features. The splitting process is completed using feature label matching and dimension classification algorithms to ensure the specificity and completeness of each subset's features. S32 activates the feature matching layer of the student physical weakness diagnosis and assessment model, calling a pre-set standard physical feature template library. This library includes standard feature data for students of different ages and genders. During comparison, a feature component-by-feature matching method is used to calculate feature distance and similarity values. A similarity threshold of 0.75 is set, and features below this threshold are filtered out. S33 assigns influence weights to deviation feature items through the model's weight allocation mechanism. The weight values ​​range from 0.1 to 0.4, and the allocation is based on the correlation between the feature and physical performance and the reliability of signal acquisition. The correlation is determined through statistical analysis of historical training data, and the signal reliability is calibrated based on the sensor acquisition error rate. S34 performs a weighted summation of the number of deviation feature items, the degree of deviation, and the corresponding influence weights for each physical performance dimension to generate a quantitative evaluation result. The weak items are prioritized according to the evaluation scores from high to low, and the ranking results are directly used to formulate subsequent training plans. The entire step-by-step implementation process is realized through modular computation, ensuring the accuracy and efficiency of weak item identification and providing a clear goal orientation for personalized training.

[0041] Preferred, such as Figure 3 As shown, S4 includes the following sub-steps: S41, collecting students' multimodal physiological signal data, physical fitness test results data, and training execution parameter data from previous preset periods, organizing them according to time series, and constructing a historical training dataset, which includes feature vectors for each time node and corresponding physical fitness status labels; S42, fusing the historical training dataset with the quantitative assessment results of physical fitness weaknesses output in S3 to generate the input dataset for the time series prediction model, ensuring data consistency through feature splicing and dimension alignment during the fusion process; S43, calling the student physical fitness growth trajectory prediction algorithm, inputting the input dataset into the algorithm's time series feature extraction module, mining the long-term dependencies and short-term fluctuation patterns in the data, and constructing a time series prediction model for physical fitness development; S44, through the model's prediction output module, based on the constructed time series prediction model, outputting predicted values ​​of physical fitness status at different time nodes within the future preset period, forming complete physical fitness growth trajectory data.

[0042] Specifically, step S4 comprises four sub-steps, S41 to S44, constructing a complete implementation process around predicting physical fitness growth trajectories. S41 collects students' historical training data from the past 3 to 12 months, including multimodal physiological signal data, physical fitness test results, and training execution parameter data. Data collection is conducted weekly, with each collection lasting 15 to 30 minutes. After collection, the data is sorted and organized according to time series, constructing a historical training dataset containing 12 to 48 sets of data. Each set of data is associated with a corresponding physical fitness status label, which is divided into 5 levels based on physical fitness achievement levels. S42 fuses the historical training dataset with the quantitative assessment results of physical fitness weaknesses output from step S3. The fusion process uses feature splicing, first aligning the dimensions of the two types of data to ensure a uniform dimensionality of 128 to 256 dimensions, then eliminating dimensional differences through data normalization to generate the input dataset for the time-series prediction model. S43 invokes the time-series feature extraction module of the student physical fitness growth trajectory prediction algorithm, employing a sliding window technique to extract data features. The sliding window size is set to 3 to 5 time nodes, with a window step size of 1. This process mines long-term dependencies and short-term fluctuation patterns in the data. Based on the extracted features, a time-series prediction model is constructed, with 200 to 300 training iterations and a convergence threshold of 0.001. S44, through the model's prediction output module, outputs predicted physical fitness status values ​​for each month within the next 1 to 6 months, based on the constructed time-series prediction model. The prediction dimensions include five core aspects: endurance, strength, speed, agility, and flexibility, forming complete physical fitness growth trajectory data. This step-by-step implementation process, through time-series data mining and model prediction, provides a forward-looking basis for optimizing training parameters, ensuring the scientific validity and adaptability of the training plan.

[0043] Preferred, such as Figure 4 As shown, S5 includes the following sub-steps: S51, inputting the quantitative assessment results of physical weakness items and the predicted data of physical growth trajectory into the constraint setting module of the adaptive physical training intelligent optimization analysis platform, setting multi-dimensional constraint parameters such as the upper limit of training intensity, the range of movement difficulty levels, and the threshold of the interval cycle; S52, the platform's optimization objective construction module constructs a multi-objective optimization function based on the input data and constraint parameters, with the main objectives being to maximize the efficiency of improving weakness items, balance the training load, and rationalize the training cycle; S53, calling the platform's optimization algorithm execution module, iteratively optimizing the training intensity, training duration, movement combination, and interval cycle parameters through the multi-objective constraint optimization mechanism, generating multiple sets of candidate parameter configuration schemes; S54, through the platform's scheme evaluation and screening module, comprehensively evaluating the candidate parameter configuration schemes according to the preset evaluation indicators, and selecting the optimal training parameter configuration scheme as the output result.

[0044] Specifically, step S5 involves adaptively optimizing the training parameters. S51 inputs the quantitative assessment results of physical weaknesses and the predicted data of physical growth trajectory into the constraint setting module of the adaptive physical training intelligent optimization analysis platform, setting multi-dimensional constraint parameters. These parameters include an upper limit for training intensity of 60% to 80% of the student's maximum exercise load, a range of movement difficulty levels from 1 to 5, an interval threshold of 30 seconds to 3 minutes, and a total training duration constraint of no more than 90 minutes. These constraint parameters are determined based on the student's age, physical fitness level, and exercise safety standards. S52, the platform's optimization objective construction module, based on the input data and constraint parameters, constructs a multi-objective optimization function. The core optimization objectives are maximizing the efficiency of improving weaknesses, balancing the training load, and rationalizing the training cycle. The weights of each objective are set to 0.4, 0.3, and 0.3, respectively. The weight allocation is determined using the analytic hierarchy process (AHP) to ensure a balanced consideration of each objective. S53 invokes the platform's optimization algorithm execution module, employing a multi-objective constraint optimization mechanism for iterative optimization. The number of iterations is set to 100 to 200. Each iteration selects the optimal solution based on the fitness function. Optimization parameters include training intensity, training duration, action combinations, and interval periods. Action combinations include 8 to 12 basic training actions, sorted and combined according to the need to improve weaknesses. S54 uses the platform's scheme evaluation and screening module to comprehensively evaluate candidate parameter configuration schemes based on preset evaluation indicators. Evaluation indicators include the potential for improving weaknesses, training load adaptability, and the scientific nature of action combinations. Each indicator has a maximum score of 10 points. The optimal training parameter configuration scheme with a comprehensive score of not less than 8.5 points is selected. This step-by-step implementation process, through constraint setting, objective construction, iterative optimization, and scheme screening, ensures that the generated training parameter configuration is targeted, feasible, and efficient, providing a precise basis for training execution.

[0045] like Figure 5As shown, a multimodal adaptive student physical fitness training system is applied to a multimodal adaptive student physical fitness training method. The system includes: a multimodal physiological-motor signal synchronous acquisition unit, a high-dimensional feature intelligent analysis unit, a physical fitness deficiency accurate diagnosis unit, a growth trajectory time-series prediction unit, a training parameter adaptive optimization unit, and a closed-loop training execution and adjustment unit. Each unit is bidirectionally connected to a control bus via a data bus. The multimodal physiological-motor signal synchronous acquisition unit captures heart rate variability, electromyography, respiratory rate, blood oxygen saturation, and movement posture mechanical signals during student physical fitness training and transmits them to the high-dimensional feature intelligent analysis unit. The high-dimensional feature intelligent analysis unit extracts features from the received signals using a multimodal physiological signal analysis convolutional network to generate high-dimensional feature vectors. The matrix is ​​sent to the physical fitness deficiency precise diagnosis unit; the physical fitness deficiency precise diagnosis unit, based on the high-dimensional feature vector matrix, locates the physical fitness weaknesses through the student physical fitness deficiency diagnosis and assessment model and transmits the results to the growth trajectory time series prediction unit; the growth trajectory time series prediction unit uses the student physical fitness growth trajectory prediction algorithm to output physical fitness development trend data, which, together with the weakness data, is sent to the training parameter adaptive optimization unit; the training parameter adaptive optimization unit generates the optimal training parameter configuration through the adaptive physical fitness training intelligent optimization analysis platform and transmits it to the closed-loop training execution and adjustment unit; the closed-loop training execution and adjustment unit starts the training process based on the parameter configuration, collects signal changes in real time during the training process and feeds them back to the high-dimensional feature intelligent analysis unit, dynamically adjusts the training parameters to carry out multimodal adaptive physical fitness training.

[0046] The formula in this invention integrates scalar and vector parameters into the same computational framework. First, standardization eliminates dimensional differences and the influence of different types of parameters. Then, a unified computational dimension is established based on the physical meaning of the parameters and their correlation with physical training. For example, scalar parameters include attention mechanism weight coefficients, feature matching coefficients, and dynamic adjustment coefficients. These parameters have no directional attributes and only reflect the degree of influence through their numerical value. Vector parameters include multimodal physiological-motor fusion signal sets, high-dimensional feature vector matrices, and quantitative assessment vectors of physical weaknesses, encompassing feature information across multiple dimensions. Before computation, scalar parameters are normalized within a preset range, such as adjusting feature importance weights to a uniform range of 0.1 to 0.5. Vector parameters are converted from vectors of different lengths into a matrix form of a unified dimension through feature dimension alignment and data normalization, such as standardizing feature vectors from 128 to 256 dimensions into a unified 200-dimensional matrix. At the same time, by clarifying the functional positioning of parameters in physical training, such as scalar parameters being used to adjust weights and vector parameters being used to provide feature basis, the two form a complementary calculation relationship in the formula, ensuring that different types of parameters can participate in the calculation in a coordinated manner.

[0047] The formula design fully considers the compatibility of scalar and vector operations, achieving organic integration of the two through reasonable operational logic and structural design, and optimizing the calculation path based on the characteristics of the specific parameters of this invention. For example, in the formula related to multimodal physiological signal analysis, the multimodal fusion signal in vector form is used to extract features through convolution operations, and then combined linearly with the scalar form convolution kernel weight matrix and bias vector to achieve feature enhancement; in the formula for diagnosing physical fitness deficiencies, the vector feature vector and the scalar similarity threshold are compared to filter out deviating feature terms, and then weighted and summed with the scalar influence weights to generate a quantitative result. This design not only retains the multidimensional feature information of vector parameters, but also utilizes the adjustment effect of scalar parameters to optimize calculation accuracy. At the same time, by using activation functions, feature fusion operators, etc., the calculation results are mapped to a reasonable range, ensuring that the calculation results of different types of parameters can be effectively integrated to form output data that is both comprehensive and accurate, adapting to the needs of complex parameter processing in multimodal adaptive student physical fitness training.

[0048] A multimodal adaptive student physical fitness training method and system overcomes the shortcomings of existing technologies in data collection and analysis by synchronously capturing and deeply integrating multi-dimensional data. It goes beyond the collection or simple integration of single-type data; instead, it simultaneously acquires various data related to students' physiological state and athletic performance. Through systematic analysis, it uncovers the intrinsic relationships between different data points, achieving a comprehensive and accurate assessment of students' physical fitness, clearly defining specific weaknesses in each dimension, and providing a solid and reliable basis for training program development. Simultaneously, by combining students' historical training data with physical development patterns for training planning and constructing a real-time feedback and adjustment chain during training, it effectively solves the problem of existing training programs lacking personalization and dynamic adjustment capabilities. This allows key elements such as training content and intensity to be flexibly adapted and optimized according to real-time changes in students' physical fitness.

[0049] This method and system excel in the comprehensiveness and accuracy of data processing and physical fitness assessment. Through in-depth fusion and analysis of multi-dimensional data, it ensures accurate judgment of students' physical fitness status, providing scientific support for subsequent training. The training program is highly personalized and adaptable, fully adapting to the physiological conditions and physical fitness differences of different students and meeting their individual training needs. The dynamic adjustment mechanism of the training process is efficient and flexible, responding in real time to changes in students' physical fitness status and ensuring continuous optimization of training effects. Fourth, the overall training process forms a closed-loop link, seamlessly connecting data collection, assessment, program formulation to execution and adjustment, greatly improving training efficiency and overall quality, and helping students steadily improve their physical fitness.

[0050] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "set," "install," "connect," "link," and "fix" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal communication between two components. Those skilled in the art will understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0051] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various equivalent changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A multimodal adaptive student physical fitness training method, characterized in that, Includes the following steps: S1. Simultaneously capture heart rate variability, electromyography, respiratory rate, blood oxygen saturation, and motion posture biomechanics signals during student physical training using a multimodal physiological signal acquisition module to construct a multi-dimensional physiological-motor fusion signal set. S2. Input the fusion signal set into a multimodal physiological signal parsing convolutional network. Deep feature extraction links are used to mine temporal, frequency, and spatial domain correlation features, generating a high-dimensional feature vector matrix. S3. Based on the high-dimensional feature vector matrix, a student physical fitness weakness diagnosis and assessment model is invoked. Through multi-level feature matching and weight allocation mechanisms, the student's weaknesses in endurance, strength, speed, agility, and flexibility are identified. S4. Utilizing a student physical fitness growth trajectory prediction algorithm, combined with historical training data and current physical fitness weakness features, a time-series prediction model is constructed to output data on physical fitness development trends within a preset future period. S5 inputs data on physical weaknesses and development trends into the adaptive physical training intelligent optimization analysis platform, and generates training intensity, training duration, movement combinations and rest interval parameter configurations that are suitable for the individual through a multi-objective constraint optimization mechanism; S6 initiates a multimodal adaptive student physical training execution process based on parameter configuration, provides real-time feedback on signal changes during the training process, and dynamically adjusts training parameter configuration to form a closed-loop training link.

2. The multimodal adaptive student physical fitness training method according to claim 1, characterized in that, The feature extraction expression of the multimodal physiological signal parsing convolutional network is as follows: ,in, It is a high-dimensional eigenvector matrix. For improved activation functions, These are the one-dimensional, two-dimensional, and three-dimensional convolution kernel weight matrices, respectively. These are the bias vectors for each convolutional layer. It is a collection of multimodal physiological-motor fusion signals. represents the weight coefficients of the attention mechanism, Conv1D, Conv2D, and Conv3D represent one-dimensional, two-dimensional, and three-dimensional convolution operations, respectively, and Attention(⋅) is the cross-modal attention calculation function.

3. The multimodal adaptive student physical fitness training method according to claim 1, characterized in that, The expression for locating weaknesses in the student physical fitness deficiency diagnostic assessment model is as follows: ,in, Quantify the assessment vector for physical fitness weaknesses. The number of feature dimensions, For the first Importance weights of class features For the first High-dimensional feature vectors This is a standard physical fitness feature template vector. This is the function for calculating feature similarity. For feature matching coefficients, The number of dimensions for physical fitness assessment. For the first Diagnostic coefficient of individual capacity dimension For the first Feature components of the individual energy dimension For the first Threshold vector of individual energy dimension, This is a logical activation function.

4. The multimodal adaptive student physical fitness training method according to claim 1, characterized in that, The trend output expression of the student physical fitness growth trajectory prediction algorithm is as follows: for Predicted physical growth at any given time For Long Short-Term Memory (LSTM) network computation functions, This is the high-dimensional feature vector at the current time. for Predicted physical fitness values ​​at any given time. For time-series dependency coefficients, For the weights of the gated recurrent units, The function is used to calculate the gated loop unit. Let this be the vector of the weakest physical condition at the current moment. for The hidden state vector at time step 1. For dynamic adjustment coefficients, The number of historical data windows. For the first The weighting coefficients of each historical window, For the first Training data vectors for each historical window, It is the hyperbolic tangent activation function.

5. The multimodal adaptive student physical fitness training method according to claim 1, characterized in that, The parameter configuration generation expression of the adaptive physical training intelligent optimization analysis platform is as follows: ,in, Configure vectors for training parameters. This is the computation function for the non-dominated sorting genetic algorithm III. To train the set of constraints, For feature fusion operators, Principal component analysis function, For element-wise product operators, To optimize the coefficients for the parameters, For training intensity-duration adaptation matrix, For the weighting coefficients of the action combination, Configure vectors for intermittent periods.

6. The multimodal adaptive student physical fitness training method according to claim 1, characterized in that, The closed-loop adjustment expression for the multimodal adaptive student physical fitness training is: ,in, This is a dynamically adjusted set of training parameters. This is a function for calculating real-time signal feedback. For real-time multimodal signals during the training process, Configure vectors for initial training parameters. For feedback adjustment coefficients, A function for calculating feature differences. For real-time high-dimensional feature vectors, As the initial high-dimensional feature vector, Update the operator for the parameter. Update function for parameters, To adjust the step size coefficient.

7. The multimodal adaptive student physical fitness training method according to claim 1, characterized in that, S3 includes the following steps: S31, splitting the high-dimensional feature vector matrix according to the feature dimensions corresponding to endurance, strength, speed, agility, and flexibility to form a unique feature subset for each physical fitness dimension. Each subset includes the temporal correlation feature, frequency response feature, and spatial distribution feature under that dimension; S32, calling the feature matching layer of the student physical fitness weakness diagnosis and assessment model, comparing the unique feature subsets for each physical fitness dimension with the preset standard physical fitness feature template library layer by layer, and filtering out feature items that deviate from the standard range by calculating feature distance and similarity value; S33, based on the filtered deviating feature items, assigning corresponding influence weights to each deviating feature item through the model's weight allocation mechanism. The weight allocation is dynamically allocated according to the correlation degree of the feature with physical fitness performance and the reliability of the signal; S34, integrating the number, degree of deviation, and corresponding influence weights of the deviating feature items to form a quantitative assessment result of the weaknesses in each physical fitness dimension, and clarifying the priority ranking of each weakness in the overall physical fitness system.

8. The multimodal adaptive student physical fitness training method according to claim 1, characterized in that, S4 includes the following sub-steps: S41, collecting students' multimodal physiological signal data, physical fitness test results data, and training execution parameter data from previous preset periods, organizing them according to the time series, and constructing a historical training dataset, which includes feature vectors for each time node and corresponding physical fitness status labels; S42, fusing the historical training dataset with the quantitative assessment results of physical fitness weaknesses output in S3 to generate the input dataset for the time series prediction model, ensuring data consistency through feature splicing and dimension alignment during the fusion process; S43, Call the student physical fitness growth trajectory prediction algorithm, input the input dataset into the algorithm's time series feature extraction module, mine the long-term dependencies and short-term fluctuation patterns in the data, and construct a time series prediction model for physical fitness development; S44, Through the model's prediction output module, based on the constructed time series prediction model, output the predicted values ​​of physical fitness status at different time nodes within the future preset period, forming complete physical fitness growth trajectory data.

9. The multimodal adaptive student physical fitness training method according to claim 1, characterized in that, S5 includes the following steps: S51, inputting the quantitative assessment results of physical weaknesses and the predicted data of physical growth trajectory into the constraint setting module of the adaptive physical training intelligent optimization analysis platform, setting multi-dimensional constraint parameters such as the upper limit of training intensity, the range of movement difficulty levels, and the threshold of the interval cycle; S52, the platform's optimization objective construction module constructs a multi-objective optimization function based on the input data and constraint parameters, with the main objectives being to maximize the efficiency of improving weaknesses, balance the training load, and rationalize the training cycle; S53, calling the platform's optimization algorithm execution module, iteratively optimizing the training intensity, training duration, movement combination, and interval cycle parameters through the multi-objective constraint optimization mechanism, generating multiple sets of candidate parameter configuration schemes; S54, through the platform's scheme evaluation and screening module, comprehensively evaluating the candidate parameter configuration schemes according to preset evaluation indicators, and selecting the optimal training parameter configuration scheme as the output result.

10. A multimodal adaptive student physical fitness training system, characterized in that, This system is applied to a multimodal adaptive student physical training method as described in claim 1, comprising: a multimodal physiological-motor signal synchronous acquisition unit, a high-dimensional feature intelligent analysis unit, a physical fitness deficiency accurate diagnosis unit, a growth trajectory time-series prediction unit, a training parameter adaptive optimization unit, and a closed-loop training execution and adjustment unit. Each unit is bidirectionally connected to a control bus via a data bus. The multimodal physiological-motor signal synchronous acquisition unit captures heart rate variability, electromyography, respiratory rate, blood oxygen saturation, and motion posture mechanical signals during student physical training and transmits them to the high-dimensional feature intelligent analysis unit. The high-dimensional feature intelligent analysis unit extracts features from the received signals using a multimodal physiological signal analysis convolutional network, generates a high-dimensional feature vector matrix, and sends it to... The system comprises several modules: a precise physical fitness deficiency diagnosis unit, a growth trajectory prediction unit, and a training parameter adaptive optimization unit. The former uses a high-dimensional feature vector matrix to pinpoint physical fitness weaknesses using a student physical fitness deficiency diagnosis and assessment model. The latter uses a student physical fitness growth trajectory prediction algorithm to output physical fitness development trend data, which, along with the weakness data, is sent to the training parameter adaptive optimization unit. The latter generates the optimal training parameter configuration through an adaptive physical fitness training intelligent optimization analysis platform and transmits it to the closed-loop training execution and adjustment unit. The latter initiates the training process based on the parameter configuration, collects signal changes in real time during training, and feeds them back to the high-dimensional feature intelligent analysis unit, dynamically adjusting the training parameters for multimodal adaptive physical fitness training.

Images