A method for predicting athletes' physical condition based on machine learning

Abstract

The present invention relates to the field of specific computational models in computer systems, and in particular to a method for predicting an athlete's physical state based on machine learning. The method comprises the following steps: collecting and preprocessing the athlete's historical training data and physiological state data to obtain a physical state vector; based on the physical state vector, introducing a hierarchical structure, constructing a hierarchical state dependency matrix, and calculating a hierarchical physical state estimate; and predicting the athlete's future physical state based on the hierarchical physical state estimate to obtain a final physical state prediction result. This method addresses the problems of traditional athlete physical state prediction methods, which are unable to accurately model the multi-level dependency relationships of physical state, and the prediction results are susceptible to noise and error propagation, resulting in large long-term prediction deviations; and most technologies rely on fixed model parameters, have limited generalization capabilities, and generally rely on external sensing equipment, resulting in data loss and signal interference problems, which affect prediction reliability.

Description

Technical Field

[0001] The present invention relates to the field of specific computing models in computer systems, and in particular to a method for predicting the physical condition of athletes based on machine learning. Background Art

[0002] In modern competitive sports and sports science, monitoring and predicting athletes' physical condition is crucial for optimizing training plans, improving performance, and preventing sports injuries. With the increasing adoption of scientific training concepts, athlete physical condition assessment has gradually shifted from traditional empirical judgment to data-driven analysis. Changes in physical condition are influenced by a variety of factors, including training load, recovery time, physiological indicators, and environmental conditions. Therefore, accurate prediction of physical condition not only assists coaches in arranging training sessions but also helps athletes adjust their rhythm to maintain optimal competitive form.

[0003] Recent advances in data acquisition technology have enabled the storage and analysis of athletes' physiological and training data across various timescales. From short-term heart rate fluctuations to long-term endurance trends, these data types provide a foundation for modeling and predicting physical performance. However, extracting effective information from complex time series data and constructing reasonable computational models for high-precision prediction remain key areas of research at the intersection of sports science and intelligent computing.

[0004] Traditional methods for predicting athlete physical fitness have the following technical problems: they are unable to accurately model the multi-level dependencies of physical status, and the prediction results are easily affected by noise and error propagation, resulting in large long-term prediction deviations; most technologies rely on fixed model parameters, making it difficult to optimize for different athletes, with limited generalization capabilities, and usually rely on external sensing equipment, which is prone to data loss and signal interference problems, affecting prediction reliability. Summary of the Invention

[0005] The present invention provides a method for predicting the physical status of athletes based on machine learning to solve the problem that traditional methods for predicting the physical status of athletes cannot accurately model the multi-level dependencies of physical status, and the prediction results are easily affected by noise and error propagation, resulting in large long-term prediction deviations; most technologies rely on fixed model parameters, are difficult to optimize for different athletes, have limited generalization capabilities, and usually rely on external sensing equipment, which may cause data loss and signal interference problems, affecting the reliability of prediction.

[0006] The present invention provides a method for predicting an athlete's physical condition based on machine learning, which specifically includes the following technical solutions:

[0007] A method for predicting an athlete's physical condition based on machine learning comprises the following steps:

[0008] S1. After collecting and preprocessing the athlete's historical training data and physiological state data, a physical state vector is obtained; based on the physical state vector, a hierarchical structure is introduced to construct a hierarchical state dependency matrix, and a hierarchical physical state estimation value is calculated;

[0009] S2. Based on the hierarchical physical state estimation value, a dynamic nonlinear state evolution model is constructed to predict the athlete's future physical state and obtain the final physical state prediction result.

[0010] Preferably, the S1 specifically includes:

[0011] The physical state vector is decomposed into multiple layers of intrinsic modal components through the empirical mode decomposition technique, and the entropy values ​​of the intrinsic modal components are calculated; based on the entropy values ​​of the intrinsic modal components, the hierarchical weight factors are defined.

[0012] Preferably, the S1 specifically includes:

[0013] Based on the hierarchical weight factors and combined with the physical state vector, a hierarchical state dependency matrix is ​​constructed.

[0014] Preferably, the S1 specifically includes:

[0015] Based on the hierarchical state dependency matrix and physical state vector, a hierarchical smoothing factor and a local time adjustment term are introduced to evaluate the athlete's physical state at different time scales and calculate the hierarchical physical state estimate. The calculation formula of the hierarchical physical state estimate is as follows:

[0016] ,

[0017] in, It is in Tier The estimated value of physical state at each time step; is the time step; It is used to adjust the The physical state vector of the time step is The adjustment weight of the influence degree of the physical state vector of the time step; It is In the layer state dependency matrix, The physical state vector of time steps Hedi The physical state vector of time steps The degree of state dependence between them; It is The layer-level smoothing factor; It is Tier A local time adjustment term for each time step.

[0018] Preferably, the S2 specifically includes:

[0019] The dynamic nonlinear state evolution model introduces a state evolution operator to simulate the temporal transformation law of physical state.

[0020] Preferably, the S2 specifically includes:

[0021] The calculation formula of the state evolution operator is as follows:

[0022] ,

[0023] in, It is in Tier The state evolution operator of time steps; It is The state evolution rate parameter of each time step; It is in Tier The estimated value of physical state at each time step; Representatives in the Tier The nonlinear state transition term of time steps, is the state transition function, is the weight factor for each nonlinear state transition term; It is The state perturbation coefficient of each time step; It is in Tier The estimated value of physical state at each time step.

[0024] Preferably, the S2 specifically includes:

[0025] Based on the state evolution operator, the hierarchical physical state estimation value is dynamically updated to obtain the updated physical state estimation value.

[0026] Preferably, the S2 specifically includes:

[0027] The updated physical fitness state estimate is adjusted to obtain the final physical fitness state prediction result.

[0028] The beneficial effects of the technical solution of the present invention are:

[0029] 1. By constructing a hierarchical state dependency matrix and calculating the degree of state dependency at different time scales, we ensure that the interaction between short-term physical fitness fluctuations and long-term physical fitness changes is fully modeled, and enhance the ability of the dynamic nonlinear state evolution model to capture key physical fitness change trends. By extracting the intrinsic modal components of the physical fitness state vector through empirical mode decomposition technology and combining it with entropy to calculate hierarchical weight factors, the main characteristics of the physical fitness state can be more effectively involved in prediction, thereby improving the accuracy of state estimation.

[0030] 2. By introducing a state evolution operator to describe the instantaneous change trend of physical state, the adaptability of the dynamic nonlinear state evolution model to complex physical states is enhanced. The state disturbance coefficient is used to simulate unpredictable external influences, making the dynamic nonlinear state evolution model more robust in practical applications. The logarithmic correction term in the state evolution operator is used to ensure the smoothness of physical state changes, avoid drastic fluctuations in the predicted value in a short period of time, and improve the continuity and credibility of the prediction results.

[0031] 3. A hierarchical structure is introduced to classify the physical state vectors according to different time scales, so that the dynamic nonlinear state evolution model can be personalized and optimized for the physical state characteristics of different athletes; by introducing a global adjustment coefficient, the personalized adaptability of the prediction results is improved, so that the dynamic nonlinear state evolution model can maintain a high generalization ability among different athletes. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 This is a flow chart of a method for predicting an athlete's physical condition based on machine learning according to the present invention. DETAILED DESCRIPTION

[0033] In order to further illustrate the technical means and effects adopted by the present invention to achieve the predetermined purpose of the invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0034] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0035] The following describes in detail a specific solution of a method for predicting an athlete's physical condition based on machine learning provided by the present invention with reference to the accompanying drawings.

[0036] Refer to the attached Figure 1, which shows a flow chart of a method for predicting an athlete's physical condition based on machine learning provided by one embodiment of the present invention, the method comprising the following steps:

[0037] S1. After collecting and preprocessing the athlete's historical training data and physiological state data, a physical state vector is obtained; based on the physical state vector, a hierarchical structure is introduced to construct a hierarchical state dependency matrix, and a hierarchical physical state estimation value is calculated;

[0038] Before physical state prediction, the athletes' historical training data and physiological state data are first collected and preprocessed in order to construct a hierarchical state dependency matrix and input it into the dynamic nonlinear state evolution model for prediction.

[0039] Since the collected historical training data and physiological status data of athletes will change over time and may contain noise, in order to ensure the stability of the data, the collected data needs to be preprocessed. The preprocessing process includes: standardizing the collected data so that it is mapped to the same scale; and using the mean filtering method to correct outliers and smooth short-term fluctuations to improve the calculation accuracy of the hierarchical state dependency matrix, making the subsequent state evolution calculation more accurate.

[0040] The dataset obtained after normalization and outlier correction is defined as , where each Representative athletes in The physical state vector of each time step includes key indicators that affect the evolution of physical state, such as heart rate, blood oxygen saturation, cadence, speed, lactate concentration, etc. is the time step.

[0041] An athlete's physical condition is influenced by multiple timescales, including short-term influences (seconds and minutes): such as heart rate fluctuations and rapid changes in lactate levels during exercise; medium-term influences (hours and days): such as post-training recovery and the gradual accumulation of physical energy expenditure; and long-term influences (weeks and months): such as an athlete's endurance adaptability and the long-term accumulation of training cycles. To accurately model the dynamic changes in physical condition, a hierarchical structure is introduced, where data at different timescales is divided into different levels, with each level capturing information at a specific timescale.

[0042] The physical state vector is decomposed into multiple layers of intrinsic modal components through empirical mode decomposition technology, and the entropy value of each layer of intrinsic modal components is calculated. The hierarchical weight factor is defined according to the entropy value. , Indicates the Layer weight factors, It is The entropy of the intrinsic modal components of the layer, is the number of levels, set based on expert experience; levels with greater information content are given higher weights to ensure that the main variation characteristics of the data are fully utilized. The levels are categorized according to the entropy value and time scale of the intrinsic modal components: high-frequency levels (short-term effects) are used to capture rapidly fluctuating physical changes, medium-frequency levels (medium-term effects) are used to model the recovery process after exercise, and low-frequency levels (long-term effects) are used to model the long-term effects of training cycles.

[0043] Because changes in an athlete's physical state are affected by complex factors and have dependencies on different time scales, a hierarchical state dependency matrix needs to be constructed to describe the state correlation between different time steps. The hierarchical state dependency matrix is ​​used to measure the state similarity between different time steps and is stored and calculated in a computer system to ensure that the dynamic nonlinear state evolution model can accurately capture the dynamic evolution characteristics of the physical state. The calculation method for any element in the hierarchical state dependency matrix is ​​as follows:

[0044] ,

[0045] in, It is In the layer state dependency matrix, The physical state vector of time steps Hedi The physical state vector of time steps The degree of state dependence between them; It is Layer weight factor, used to control the relative importance between different time steps. The physical state vector of the time step and the The Euclidean distance between the physical state vectors of time steps , and weights are assigned using an exponential decay function to ensure that time steps with similar states have high correlation.

[0046] Based on the hierarchical state dependency matrix, the estimated physical state value of each level is calculated to describe the athlete's physical state at different levels. It is a comprehensive assessment of the athlete's physical state at different time scales and can reflect the coupling relationship between its short-term fluctuations and long-term trends. The calculation process of the estimated physical state value integrates the influence of the hierarchical state dependency matrix and introduces a local time adjustment term to ensure the smoothness of state changes. The calculation formula of the hierarchical physical state estimate is as follows:

[0047] ,

[0048] ,

[0049] ,

[0050] in, It is in Tier The estimated value of the physical state of the time step represents the athlete's multi-index physical state (such as heart rate, lactate concentration, step frequency, etc.) at the time step. Estimates of layers; Is to adjust the weight, used to adjust the The physical state vector of the time step is The degree of influence of the physical state vector of each time step is obtained through experiments; It is The layer-level smoothing factor is used to control the smoothness of the changes in physical state; It is Tier The local time adjustment term of each time step is used to compensate for the short-term fluctuation of physical state, making the estimated value of physical state more stable; is the amplitude of the 𝑚th sinusoidal component, which is used to control the contribution of this sinusoidal component to the overall periodic variation and is obtained experimentally; is the angular frequency of the 𝑚th sinusoidal component, which is used to describe the time rate of change of this sinusoidal component and is obtained experimentally; is the phase offset of the 𝑚th sinusoidal component, which is used to adjust the time alignment of the sine function and is obtained experimentally; Indicates the number of sinusoidal components used to approximate the periodic changes in physical state, that is, how many sinusoidal components are selected to approximate the periodic changes, and is set based on expert experience.

[0051] S2. Based on the hierarchical physical state estimation value, a dynamic nonlinear state evolution model is constructed to predict the athlete's future physical state and obtain the final physical state prediction result.

[0052] Based on the hierarchical physical state estimates, a dynamic nonlinear state evolution model is constructed to predict the athlete's future physical state. Because physical state is affected by multiple nonlinear factors, the change process of physical state is usually not a simple linear change. Therefore, a state evolution operator is introduced into the dynamic nonlinear state evolution model to describe the time evolution characteristics of physical state. It is used to simulate the transformation law of physical state over time to ensure that the athlete's future physical state can be accurately predicted. The calculation formula of the state evolution operator is as follows:

[0053] ,

[0054] in, It is in Tier The state evolution operator of time steps represents the instantaneous change trend of physical state; It is The state evolution rate parameter of each time step is used to control the temporal variation of the current physical state and is obtained through experiments; Representatives in the Tier The nonlinear state transition term of time steps, is a state transition function, and the Bessel function in the prior art can be used to describe different nonlinear state change modes. is the weight factor for each nonlinear state transition term, which is used to control the relative contribution of different state transition functions and is obtained through experiments; It is The state perturbation coefficient of each time step is used to control the random perturbation part of the state update to simulate unpredictable changes in physical state, such as environmental factors and temporary fatigue of athletes, and is obtained through experiments; It is in Tier The estimated value of physical state at time steps; the state disturbance term It is used to ensure the stability of state changes, prevent drastic fluctuations in the predicted value of physical state, and balance the dynamic changes of physical state.

[0055] In the state evolution stage, in order to accurately simulate the nonlinear change trend of physical state at different time steps, the state evolution operator is used for state transformation, and the fusion weight is introduced to dynamically update the layered physical state estimate to obtain the updated physical state estimate. This allows the state evolution to not only capture short-term change trends but also consider long-term dependencies. The state update formula is as follows:

[0056] ,

[0057] in, is the updated estimated value of physical state, that is, the physical state after calculation by the state evolution operator; is the step size factor of the state update, which is used to control the convergence and divergence of the state evolution process and is obtained through experiments; It is The fusion weight of the layer is used to control the contribution of the state evolution operators at different levels and is obtained through experiments; is the regularization coefficient, which ensures that the state change remains stable and is obtained through experiments.

[0058] Based on the updated physical state estimate, the prediction value is further calculated to obtain the final physical state prediction result. In the embodiment of the present application, the updated physical state estimate can be adjusted by existing technologies (such as automatic differentiation, regularization, attention mechanism and adaptive optimization algorithm), or by introducing the following formula to adjust the final physical state prediction result so that it conforms to the changing trend of historical physical state data and effectively reduces the accumulation of long-term prediction errors. The specific calculation formula is:

[0059] ,

[0060] in, It is in The final physical state prediction result of time steps; Represents the global adjustment coefficient, which is used to control the magnitude of the error adjustment; According to the loss function The calculated gradient information is used to adaptively adjust the prediction error. The loss function can be constructed using the mean square error. It is a state change smoothing term, which is used to ensure the stability of the final physical state prediction results. It is a state change smoothing factor, which is used to prevent the predicted value from changing drastically between consecutive time steps, making the final physical state prediction result more stable.

[0061] In summary, a method for predicting athletes' physical status based on machine learning has been completed.

[0062] The order in which the embodiments of the invention are presented is for illustrative purposes only and does not necessarily represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require the specific order or sequential order shown to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0063] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on the differences from other embodiments.

[0064] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or make equivalent replacements for some of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included in the scope of protection of the present invention.

Claims

1. A method for predicting athlete's physical condition based on machine learning, characterized in that: The following steps are involved: S1. After collecting and preprocessing the athlete's historical training data and physiological state data, a physical state vector is obtained. Based on the physical state vector, a hierarchical structure is introduced to construct a hierarchical state dependency matrix. Based on the hierarchical state dependency matrix and the physical state vector, the athlete's physical state at different time scales is evaluated and the hierarchical physical state estimate is calculated. The specific calculation formula is as follows: ， in, It is in Tier The estimated value of physical state at each time step; is the time step; It is used to adjust the The physical state vector of the time step is The adjustment weight of the influence degree of the physical state vector of the time step; It is In the layer state dependency matrix, The physical state vector of time steps Hedi The physical state vector of time steps The degree of state dependence between them; It is The layer-level smoothing factor; It is Tier A local time adjustment term for each time step; S2. Based on the hierarchical physical state estimation value, a state evolution operator is introduced to construct a dynamic nonlinear state evolution model to predict the athlete's future physical state and obtain the final physical state prediction result; the calculation formula of the state evolution operator is as follows: ， in, It is in Tier The state evolution operator of time steps; It is The state evolution rate parameter of each time step; Representatives in the Tier The nonlinear state transition term of time steps, is the state transition function, is the weight factor for each nonlinear state transition term; It is The state perturbation coefficient of each time step; It is in Tier The estimated value of physical state at each time step.

2. A method for predicting athlete's physical condition based on machine learning according to claim 1, characterized in that: Said S1 specifically includes: The physical state vector is decomposed into multiple layers of intrinsic modal components through the empirical mode decomposition technique, and the entropy values ​​of the intrinsic modal components are calculated; based on the entropy values ​​of the intrinsic modal components, the hierarchical weight factors are defined.

3. A method for predicting athlete's physical condition based on machine learning according to claim 2, characterized in that: Said S1 specifically includes: Based on the hierarchical weight factors and combined with the physical state vector, a hierarchical state dependency matrix is ​​constructed.

4. The method for predicting athlete's physical condition based on machine learning according to claim 1, characterized in that: Said S2 specifically includes: Based on the state evolution operator, the hierarchical physical state estimation value is dynamically updated to obtain the updated physical state estimation value.

5. A method for predicting athlete's physical condition based on machine learning according to claim 4, characterized in that: Said S2 specifically includes: The updated physical fitness state estimate is adjusted to obtain the final physical fitness state prediction result.

Images