Intelligent sports data analysis method
By leveraging advanced models like Transformer, BERT, and LSTM for sports data analysis, the method addresses multi-dimensional data associations and real-time updates, improving training strategy accuracy and safety with personalized and adaptive training plans.
Patent Information
- Application Number
- CN202510303276.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-03-14
- Publication Date
- 2025-07-15
AI Technical Summary
The existing sports data analysis methods have shortcomings in dealing with multi-dimensional data correlation, real-time model updates and personalized adaptation, which leads to insufficient comprehensive and in-depth analysis results, making it difficult to meet the diverse needs of different athletes, and lack effective risk warning and training safety guarantees.
The Transformer model, BERT model and generative pre-training model are adopted, and a variety of anomaly detection algorithms and particle swarm optimization algorithms are combined to build a training strategy prediction model. Through the multi-head attention mechanism and regularization term optimization model, physiological and biochemical indicators and psychological state data are integrated to realize personalized training strategy generation and real-time risk warning.
It significantly improves the accuracy of training strategy prediction and the fit of personalized training suggestions, ensures that the training plan meets the needs of athletes, provides real-time risk warning and safety guarantee, and improves the intelligence level of the training process.
Smart Images
Figure CN120316663A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of data analysis, and in particular to an intelligent analysis method for sports data. Background Art
[0002] In today's sports field, with the continuous progress of technology and the increasing richness of data acquisition means, sports data analysis has moved from traditional simple statistics to a new stage of intelligence and personalization. However, existing sports data analysis methods still have some limitations.
[0003] First, current traditional training data analysis often focuses on single - dimensional data processing, such as only paying attention to training intensity or training time, while ignoring the correlation between multi - dimensional data, resulting in incomplete and in - depth analysis results.
[0004] Second, although some existing intelligent analysis methods can perform a certain degree of prediction and optimization on training data, they still have deficiencies in real - time update and personalized adaptation of the model, and it is difficult to meet the diverse needs of different athletes.
[0005] Third, for risk warning and timely adjustment that may occur during the training process, existing technologies also lack effective solutions and cannot ensure the safety of athletes during training and the sustainability of training effects. Summary of the Invention
[0006] In view of the above problems that current traditional training data analysis often focuses on single - dimensional data processing, such as only paying attention to training intensity or training time, while ignoring the correlation between multi - dimensional data, resulting in incomplete and in - depth analysis results, and still having deficiencies in real - time update and personalized adaptation of the model, and it is difficult to meet the diverse needs of different athletes, the present invention is proposed.
[0007] Therefore, the purpose of the present invention is to provide an intelligent analysis method for sports data, aiming to: by introducing the Transformer model, BERT model, and generative pre - training model, it can more accurately capture the complex temporal features and multi - dimensional correlations in training data, thereby significantly improving the accuracy of training strategy prediction.
[0008] To solve the above - mentioned technical problems, the present invention provides the following technical solution: an intelligent analysis method for sports data, including Step 1: Data acquisition and pre - processing
[0009] S1, Collect the training data of athletes;
[0010] S2, Clean the data;
[0011] S3, Normalize the data;
[0012] Step 2: Construction and Training of Training Strategy Prediction Model
[0013] S1. Build a training strategy prediction model based on the Transformer model;
[0014] S2. Introduce the BERT model to encode the training strategy data;
[0015] S3. Use the generative pre-training model to generate personalized training suggestions;
[0016] Step 3: Construction and Training of Effect Evaluation Model
[0017] S1. The effect evaluation model includes a feature extraction layer, a physiological parameter change layer, a training effect change layer, and an effect evaluation value output layer;
[0018] S2. Introduce the LSTM network to process time series data;
[0019] S3. Use the Transformer model to encode the training strategy data;
[0020] Step 4: Group Optimization Simulation Calculation
[0021] S1. Build a simulated individual population, and each individual contains training strategy data items;
[0022] S2. Adopt an improved particle swarm optimization algorithm for iterative update;
[0023] S3. Calculate the fitness of the simulated individual, splice the training strategy data at the current monitoring time point with the simulated individual, and construct effect evaluation data;
[0024] Step 5: Real-time Training and Update of the Model
[0025] S1. At the monitoring time point, splice the current training strategy data with the optimal training strategy data;
[0026] S2. Input the effect evaluation data into the effect evaluation model;
[0027] S3. Calculate the gradient of the effect evaluation value with respect to the optimal training strategy data.
[0028] S4. Regularly update the effect evaluation model by collecting new training data and evaluation results.
[0029] As a preferred solution of the sports data intelligent analysis method described in the present invention, in the data acquisition and preprocessing of step one, a combination of multiple anomaly detection algorithms is adopted in the data cleaning process, including identifying outliers based on statistical methods, detecting abnormal patterns in data based on machine learning models, and filtering out illogical data points based on domain knowledge rules to ensure the accuracy and reliability of the data.
[0030] As a preferred solution of the sports data intelligent analysis method described in the present invention, in the construction of the training strategy prediction model in step two, the calculation formula of the attention mechanism of the Transformer model is:
[0031] Calculation of attention scores:
[0032] where Q is the query vector; K is the key vector; V is the value vector; d k is the dimension of the key vector;
[0033] Through the multi-head attention mechanism, the outputs of multiple attention heads are concatenated and linearly transformed to obtain the final attention output, enhancing the model's ability to capture different features in the training data.
[0034] As a preferred solution of the sports data intelligent analysis method described in the present invention, in the model construction and training of step two, a regularization term is introduced to prevent the model from overfitting. The formula of the regularization term is:
[0035] L2 regularization term:
[0036] where w i is the weight parameter of the model; n is the total number of weight parameters
[0037] Adding the regularization term to the loss function to obtain the final optimization objective function:
[0038] L final = L original + λ·R L2
[0039] where L original is the original loss function; λ is the regularization coefficient, which controls the proportion of the regularization term in the total loss. By adjusting the regularization coefficient, the fitting ability and generalization ability of the model are balanced, and the robustness and reliability of the model are improved.
[0040] As a preferred solution of the sports data intelligent analysis method described in the present invention, in the training of the effect evaluation model in step three, the following formula is used to calculate the loss value:
[0041] Loss value of physiological parameter change:
[0042] Training effect change loss value:
[0043]
[0044] Total loss value: L total＝ α·L ph ysiology +β·L performance
[0045] Where N is the number of training samples; y physiology,i and y performance,i are respectively the true physiological parameter change and training effect change of the i-th sample; and are respectively the physiological parameter change and training effect change predicted by the model; α and β are loss weight coefficients used to balance the contributions of physiological parameter change and training effect change in the total loss.
[0046] As a preferred solution of the sports data intelligent analysis method described in the present invention, wherein: in the construction and training of the effect evaluation model in step three, a multi-source data fusion technology is introduced. In addition to training strategy data, various information sources such as athletes' physiological and biochemical indexes, psychological state evaluations, and sports performance data are integrated.
[0047] As a preferred solution of the sports data intelligent analysis method described in the present invention, wherein: in the group optimization simulation calculation in step four, the following formula is used for iterative update of the particle swarm optimization algorithm:
[0048] Velocity update formula:
[0049] Position update formula:
[0050] Where is the velocity value, and the initial velocity is determined by the developer; is the inertia factor, which is generally determined in advance by the developer or can be linearly changed according to the number of iterations; and are respectively the first learning factor and the second learning factor, both of which are generally 2; η is a random number between 0 and 1; is the historical optimal position of the simulated individual itself; is the historical optimal position of the population; is the current position of the simulated individual, that is, the training strategy data item.
[0051] As a preferred solution of the sports data intelligent analysis method described in the present invention, in the real-time training and updating of the model in step five, the update of the effect evaluation model is not only based on new training data, but also combines expert experience and actual training feedback information, and corrects the biases and errors in the model through manual intervention, so that the model can better adapt to the complex situations in the actual training scenario.
[0052] Advantages of the present invention:
[0053] 1. In the present invention, by introducing the Transformer model, BERT model, and generative pre-training model, it is possible to more accurately capture the complex temporal features and multi-dimensional correlations in the training data, thereby significantly improving the accuracy of training strategy prediction. These advanced model architectures can better process long-sequence data, understand the semantic information in the training strategy, and generate personalized training suggestions that are more in line with the actual situation of the athletes.
[0054] 2. At the same time, the present invention fully considers the personal characteristics and historical training data of the athletes, and can customize exclusive training plans for each athlete. By real-time monitoring the physiological parameters and training status of the athletes, the training strategy is dynamically adjusted to ensure that the training plan always meets the current needs of the athletes, realizing true personalized training guidance.
[0055] 3. By introducing a comprehensive risk warning mechanism in the present invention, based on the real-time output of the effect evaluation model, potential training risks such as the possibility of injury or poor training effect can be detected in advance. Once a risk signal is detected, the system will immediately adjust the training strategy, such as reducing the training intensity, changing the training type, or increasing the rest time, thereby effectively protecting the health and safety of the athletes and avoiding injuries caused by improper training. And by performing real-time training and updating on the training strategy prediction model and the effect evaluation model at the monitoring time points, the present invention can continuously optimize the model parameters to keep them in the best state all the time. This real-time feedback and adjustment mechanism improves the prediction and evaluation accuracy of the model, and also enhances the adaptability of the model to new data and new situations, ensuring that the system can continuously provide high-quality training guidance.
[0056] 4. By introducing the generative pre-training model into the system, the present invention injects stronger intelligent and automated capabilities into the system. This model can automatically generate reasonable and innovative training suggestions based on the understanding and learning of a large amount of training data, providing more training ideas and scheme options for the athletes and coaches, and further improving the intelligent level of the entire training process. It not only improves the accuracy of training strategy prediction and optimization, but also realizes the generation of personalized training plans, risk warning, and real-time update and optimization of the model, providing more scientific, intelligent, and safe training guidance and decision-making support for the athletes and coaches, and effectively promoting the development of sports data analysis towards a more efficient and intelligent direction. Brief Description of the Drawings
[0057] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can be obtained based on these drawings. Among them:
[0058] Figure 1 It is a schematic diagram of the overall structure of the intelligent sports data analysis method of the present invention. Detailed Embodiments
[0059] To make the above objects, features, and advantages of the present invention more obvious and understandable, the detailed embodiments of the present invention will be described in detail below with reference to the accompanying drawings of the specification.
[0060] Many specific details are set forth in the following description in order to provide a thorough understanding of the present invention. However, the present invention may be implemented in other ways different from those described herein. Those skilled in the art can make similar extensions without departing from the spirit of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.
[0061] Referring to Figure 1 , an embodiment of the present invention provides an intelligent sports data analysis method. This intelligent sports data analysis method includes Step 1: Data acquisition and preprocessing
[0062] S1. Comprehensively collect the training data of athletes, including but not limited to training types (such as sprinting, swimming, strength training, etc.), training duration (in minutes), training intensity (classified into four levels: low, medium, high, and extremely high through heart rate monitoring), physiological parameters (such as heart rate, blood pressure, HRV, etc.), and other relevant data (such as personal characteristics of athletes, such as age, gender, physical fitness, etc.). These data are obtained through a variety of sensors and monitoring devices to ensure the comprehensiveness and accuracy of the data;
[0063] S2. Use a combination of multiple anomaly detection algorithms to clean the collected data. Specifically, it includes using statistical methods to identify and remove outliers, such as using the 3σ principle to eliminate data points outside the normal range; using machine learning-based models to detect abnormal patterns in the data, such as using the Isolation Forest algorithm to identify abnormal data samples; and using domain knowledge-based rules to filter out data points that do not conform to logic, such as excluding data where the training intensity does not match the training type. Through these methods, the accuracy and reliability of the data are ensured, laying a solid foundation for subsequent analysis;
[0064] S3. Normalize the data to unify data of different dimensions into the same dimension and range, which is convenient for subsequent model processing. Through normalization, the differences between different data dimensions are eliminated, the processing efficiency and accuracy of the model for data are improved, and it is ensured that the subsequent model can better capture the features and laws in the data;
[0065] Step 2: Construction and Training of Training Strategy Prediction Model
[0066] S1. Construct a training strategy prediction model based on the Transformer model, and use the multi-head attention mechanism of the Transformer model to deeply mine the temporal features and long-term dependencies in the training data. Specifically, input the collected training data into the Transformer model according to the time series, and encode the training data at different times through the multi-head attention layer to capture the changing trends and internal connections of the training strategy in the time dimension. For example, when analyzing the training data of a week, the model can identify the reasonable distribution of different training types within a week and the progressive changes in training intensity, thus providing strong support for subsequent training strategy prediction;
[0067] S2. Introduce the BERT model to encode the training strategy data to enhance the model's understanding ability of the training strategy and improve the prediction accuracy. The BERT model has learned a large amount of language knowledge and semantic information through pre-training. In the present invention, the training strategy data is regarded as a special "language", and the BERT model is used to encode data such as training types and training intensities, and convert them into vector representations rich in semantic features. For example, different training types are mapped to different word vectors, enabling the model to better understand the semantic associations and differences between different training types, thereby improving the prediction accuracy;
[0068] S3. Use a generative pre-training model to generate personalized training suggestions, and combine the personal characteristics and historical training data of the athlete to provide a training strategy that better suits the actual situation of the athlete. Using the athlete's historical training data and personal characteristics as input, the generative pre-training model can generate a text description of the training strategy that meets the personal needs of the athlete according to the actual situation of the athlete, such as the physical recovery situation and training goals. For example, for an athlete recovering from an injury, the model may generate suggestions mainly focusing on low-intensity and high-frequency rehabilitation training, and gradually increase the training intensity in combination with the athlete's physical condition to ensure the scientific nature and personalization of the training plan;
[0069] Step 3: Construction and Training of Effect Evaluation Model
[0070] S1. The effect evaluation model includes a feature extraction layer, a physiological parameter change layer, a training effect change layer, and an effect evaluation value output layer. It is established based on the BP neural network and processes and analyzes the effect evaluation data through a multi-layer neural network structure. The feature extraction layer uses a convolutional neural network (CNN) to extract local features in the data, combines pooling operations to reduce the data dimension, reduces the computational complexity, and at the same time retains key feature information. For example, for the input training strategy data and physiological parameter data, the convolutional layer extracts the time series features and spatial features in the data, such as the change trend of training intensity, the fluctuation pattern of physiological parameters, etc., providing basic feature support for subsequent evaluation;
[0071] S2. Introduce the LSTM network to process time series data, capture the change trends of physiological parameters and training effects over time, and improve the model's processing ability for time series data. Input the physiological parameter and training effect data of athletes at multiple monitoring time points into the LSTM network, and the model can learn the dynamic change laws of physiological parameters and training effects during the training process. For example, it can predict the change trends of physiological parameters such as the heart rate recovery situation and fatigue accumulation degree of athletes under different training strategies, as well as the change of training effects such as the improvement trend of sports performance, thus providing more accurate time series information for effect evaluation;
[0072] S3. Use the Transformer model to encode the training strategy data, enhance the model's understanding ability of training strategies, and improve the accuracy of evaluation. By converting the training strategy data into high-dimensional semantic vectors, the model can better combine the complex relationships between training strategies, physiological parameters, and training effects, improving the accuracy of evaluation. For example, when evaluating the impact of a certain training strategy on athletes, the Transformer model can accurately capture the key features in the training strategy, such as the diversity of training types and the rationality of training intensity, and relate them to the changes in physiological parameters and training effects, thus more comprehensively and accurately evaluating the actual effect of the training strategy;
[0073] Step Four: Group Optimization Simulation Calculation
[0074] S1. Construct a simulated individual population, and each individual includes training strategy data items such as training type, duration, intensity, etc. During the population initialization process, fully consider the athlete's historical training strategies and current physical state, and generate a diverse initial population through intelligent algorithms. For example, refer to the athlete's training data in the past month, combine their current physical recovery situation and training goals, and generate multiple different training strategy combinations as simulated individuals to ensure the diversity and representativeness of the population, increase the coverage of the search space, and improve the probability of finding the global optimal solution;
[0075] S2. Use an improved particle swarm optimization algorithm for iterative update. Based on the particle swarm optimization algorithm, add the mutation operation of the genetic algorithm to avoid falling into local optima and improve the global search ability. During the iteration process, in addition to updating according to the velocity and position update formulas of the particle swarm optimization algorithm, randomly mutate the positions of the particles with a certain probability;
[0076] S3. Calculate the fitness of the simulated individual. Concatenate the training strategy data at the current monitoring time point with the simulated individual to construct the effect evaluation data, input it into the effect evaluation model, and output the effect evaluation value as the fitness of the simulated individual. The effect evaluation value comprehensively considers the influence of the training strategy on the athlete's physiological parameters and training effect. For example, an excellent training strategy should improve the training effect while not causing abnormal fluctuations or excessive fatigue in physiological parameters. In this way, ensure that the selected optimal training strategy data not only conforms to the current state of the athlete but also can effectively improve the training effect, realizing the optimization of the training strategy;
[0077] Step Five: Real-time Training and Update of the Model
[0078] S1. At the monitoring time point, concatenate the current training strategy data with the optimal training strategy data to construct the effect evaluation data. Integrate the currently actually executed training strategy data and the optimal training strategy data obtained through population optimization simulation to form a comprehensive data sample;
[0079] S2. Input the effect evaluation data into the effect evaluation model to calculate the effect evaluation value;
[0080] S3. Calculate the gradient of the effect evaluation value with respect to the optimal training strategy data, and use the gradient ascent method to adjust the parameters of the training strategy prediction model to achieve real-time training, enabling the model to self-optimize and update according to the latest training data. Calculate the gradient of the effect evaluation value with respect to the optimal training strategy data through the backpropagation algorithm. This gradient reflects the influence degree of the optimal training strategy data on the effect evaluation value. Use the gradient ascent method to update the parameters of the training strategy prediction model, enabling the model to self-optimize and update according to the latest training data and evaluation results, improving the prediction accuracy of the model, and ensuring that the subsequent generated training strategies are more in line with the actual situation of the athlete.
[0081] S4. Regularly update the effect evaluation model. By collecting new training data and evaluation results, retrain the effect evaluation model to ensure its evaluation accuracy and adaptability to the training strategy. For example, obtain the latest training data and corresponding evaluation results from the training database every week, and use these new data to fine-tune the effect evaluation model, adjusting the model's parameters and structure to adapt to the changes in the athlete's state, as well as the new training environment and conditions, maintaining the high performance and practicality of the model, and providing reliable training strategy evaluation and optimization support for athletes and coaches.
[0082] In the data acquisition and preprocessing of Step 1, the data cleaning process adopts a combination of multiple anomaly detection algorithms, including identifying outliers based on statistical methods, detecting abnormal patterns in data using machine learning-based models, and filtering out illogical data points based on domain knowledge rules, ensuring the accuracy and reliability of the data.
[0083] In the construction of the training strategy prediction model in Step 2, the calculation formula for the attention mechanism of the Transformer model is:
[0084] Attention score calculation:
[0085] where Q is the query vector; K is the key vector; V is the value vector; d k is the dimension of the key vector;
[0086] Through the multi-head attention mechanism, the outputs of multiple attention heads are concatenated and linearly transformed to obtain the final attention output, enhancing the model's ability to capture different features in the training data.
[0087] In the model construction and training of Step 2, a regularization term is introduced to prevent the model from overfitting. The formula for the regularization term is:
[0088] L2 regularization term:
[0089] where w i is the weight parameter of the model; n is the total number of weight parameters
[0090] Add the regularization term to the loss function to obtain the final optimized objective function:
[0091] L final = L original + λ·R L2
[0092] where L originalis the original loss function; λ is the regularization coefficient, which controls the proportion of the regularization term in the total loss. By adjusting the regularization coefficient, the fitting ability and generalization ability of the model are balanced, and the robustness and reliability of the model are improved.
[0093] In the training of the effect evaluation model in Step 3, the following formula is used to calculate the loss value:
[0094] Loss value of physiological parameter change:
[0095] Loss value of training effect change:
[0096]
[0097] Total loss value: L total = α·L ph ysiology + β·L performance
[0098] where N is the number of training samples; y physiology,i and y performance,i are the true physiological parameter change and training effect change of the i-th sample respectively; and are the physiological parameter change and training effect change predicted by the model respectively; α and β are loss weight coefficients, which are used to balance the contributions of physiological parameter change and training effect change in the total loss.
[0099] In the construction and training of the effect evaluation model in Step 3, a multi-source data fusion technology is introduced. In addition to training strategy data, various information sources such as athletes' physiological and biochemical indexes, psychological state evaluation, and sports performance data are integrated.
[0100] In the group optimization simulation calculation in Step 4, the following formula is used for the iterative update of the particle swarm optimization algorithm:
[0101] Velocity update formula:
[0102] Position update formula:
[0103] where is the velocity value, and the initial velocity is determined by the developer; is the inertia factor, which is generally determined in advance by the developer, or can be linearly changed according to the number of iterations; and are the first learning factor and the second learning factor respectively, and are generally both 2; η is a random number between 0 and 1; is the historical optimal position of the simulated individual itself; is the historical optimal position of the population; To simulate the current position of an individual, that is, the training strategy data item.
[0104] In the real-time training and updating of the model in Step 5, the update of the effect evaluation model is not only based on new training data, but also combines expert experience and actual training feedback information. By manually intervening to correct the biases and errors in the model, the model can better adapt to the complex situations in the actual training scenario.
[0105] During use, by introducing the Transformer model, BERT model, and generative pre-trained model, it is possible to more accurately capture the complex temporal features and multi-dimensional correlations in the training data, thereby significantly improving the accuracy of training strategy prediction. These advanced model architectures can better process long sequence data, understand the semantic information in the training strategy, and generate personalized training suggestions that are more in line with the actual situation of the athletes. At the same time, the present invention fully considers the personal characteristics and historical training data of the athletes, can customize an exclusive training plan for each athlete, and dynamically adjusts the training strategy by real-time monitoring of the physiological parameters and training status of the athletes to ensure that the training plan always meets the current needs of the athletes and realizes true personalized training guidance.
[0106] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered within the scope of the claims of the present invention.
Claims
1. An intelligent analysis method for sports data, characterized in that, It includes the following steps: Step 1: Data acquisition and preprocessing S1. Collect the training data of athletes; S2. Clean the data; S3. Normalize the data; Step 2: Construction and training of the training strategy prediction model S1. Build a training strategy prediction model based on the Transformer model; S2. Introduce the BERT model to encode the training strategy data; S3. Use the generative pre-training model to generate personalized training suggestions; Step 3: Construction and training of the effect evaluation model S1. The effect evaluation model includes a feature extraction layer, a physiological parameter change layer, a training effect change layer, and an effect evaluation value output layer; S2. Introduce the LSTM network to process time series data; S3. Use the Transformer model to encode the training strategy data; Step 4: Swarm optimization simulation calculation S1. Construct a simulated individual population, and each individual contains training strategy data items; S2. Adopt an improved particle swarm optimization algorithm for iterative update; S3. Calculate the fitness of the simulated individuals, splice the training strategy data at the current monitoring time point with the simulated individuals, and construct the effect evaluation data; Step 5: Real-time training and update of the model S1. At the monitoring time point, splice the current training strategy data with the optimal training strategy data; S2. Input the effect evaluation data into the effect evaluation model; S3. Calculate the gradient of the effect evaluation value with respect to the optimal training strategy data. S4. Regularly update the effect evaluation model by collecting new training data and evaluation results.
2. The intelligent analysis method for sports data according to claim 1, wherein: In the data acquisition and preprocessing of Step 1, the data cleaning process adopts a combination of multiple anomaly detection algorithms, including identifying outliers based on statistical methods, detecting abnormal patterns in data using machine learning-based models, and filtering out illogical data points based on domain knowledge rules to ensure the accuracy and reliability of the data.
3. The intelligent analysis method for sports data according to claim 2, characterized in that: In the construction of the training strategy prediction model in Step 2, the calculation formula for the attention mechanism of the Transformer model is: Attention score calculation: Among them, Q is the query vector; K is the key vector; V is the value vector; d k is the dimension of the key vector; Through the multi-head attention mechanism, the outputs of multiple attention heads are spliced and linearly transformed to obtain the final attention output, enhancing the model's ability to capture different features in the training data.
4. The intelligent sports data analysis method according to claim 3, characterized in that: In the model construction and training of Step 2, a regularization term is introduced to prevent the model from overfitting. The formula for the regularization term is: L2 regularization term: where w i is the weight parameter of the model; n is the total number of weight parameters Add the regularization term to the loss function to obtain the final optimized objective function: L final = L original + λ·R L2 Among them, L original is the original loss function; λ is the regularization coefficient, which controls the proportion of the regularization term in the total loss. By adjusting the regularization coefficient, the fitting ability and generalization ability of the model are balanced, and the robustness and reliability of the model are improved.
5. The intelligent sports data analysis method according to claim 4, wherein: In the training of the effect evaluation model in Step 3, the following formula is used to calculate the loss value: Loss value of physiological parameter change: Loss value of training effect change: Total loss value: where N is the number of training samples; and y performance,i are the true physiological parameter change and training effect change of the i-th sample, respectively; and are the physiological parameter change and training effect change predicted by the model, respectively; α and β are loss weight coefficients used to balance the contributions of the physiological parameter change and training effect change to the total loss.
6. The intelligent analysis method of sports data according to claim 5, wherein: In the construction and training of the effect evaluation model in Step 3, a multi-source data fusion technology is introduced. In addition to the training strategy data, various information sources such as the physiological and biochemical indicators of athletes, psychological state evaluation, and sports performance data are also integrated.
7. The intelligent sports data analysis method according to claim 6, wherein: In the swarm optimization simulation calculation of Step 4, the following formula is used for the iterative update of the particle swarm optimization algorithm: Speed update formula: Position update formula: Among them, is the speed value, and the initial speed is determined by the developer; is the inertia factor, which is generally determined by the developer in advance or can vary linearly according to the number of iterations; c1 and c2 are the first learning factor and the second learning factor respectively, both of which are generally 2; η is a random number between 0 and 1; is the historical optimal position of the simulated individual itself; is the historical optimal position of the population; is the current position of the simulated individual, that is, the training strategy data item.
8. The intelligent sports data analysis method according to claim 7, characterized in that: In the real-time training and updating of the model in Step 5, the update of the effect evaluation model is not only based on new training data, but also combines expert experience and actual training feedback information. Deviations and errors in the model are corrected through manual intervention, enabling the model to better adapt to the complex situations in the actual training scenario.