Athlete physical training management method and system based on machine learning

Abstract

The invention provides an athlete physical training management method and system based on machine learning, and relates to the technical field of physical training optimization. The method comprises the steps of collecting historical physical training time sequence record data of multiple athletes, performing time sequence decomposition to obtain multiple groups of physical training trend feature information of the athletes, and performing physical training and recovery association analysis on the athletes to construct a load recovery adaptability curve of the athletes; carrying out joint clustering on the plurality of athletes to obtain a plurality of training mode groups; a mode physical training scheme of the training mode group is generated, the current physical training scheme of the athletes is optimized based on the load recovery adaptability curve and the mode physical training scheme of the athletes, and a personalized physical training scheme of the athletes is generated. According to the method, personalized optimization of the training scheme of the athletes is realized, and the risks of overtraining and insufficient recovery are reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of physical training optimization, and particularly to a method and system for managing athletes' physical training based on machine learning. Background Art

[0002] In modern management of athletes' physical training, scientifically formulating training programs is an important means to improve athletes' competitive levels and avoid overtraining. Some traditional training programs are usually formulated based on data such as physiological indicators, sports performance, and recovery status, with more emphasis on static evaluation, resulting in certain limitations in dynamic adjustment and personalized optimization during the training management process.

[0003] The physiological responses, recovery abilities, and training load adaptabilities of different athletes vary. A single training plan cannot effectively meet the personalized needs of athletes. Especially in the management of training load, if the relationship between the athlete's recovery status and training load cannot be accurately evaluated, it is easy to cause overtraining or insufficient recovery, thereby increasing the risk of athletes getting injured and affecting the training effect. Summary of the Invention

[0004] To solve the above technical problems, the present invention proposes a method and system for managing athletes' physical training based on machine learning. By extracting feature information such as long-term trends and short-term fluctuations contained in historical training data, comprehensively analyzing the correlation between athletes' training loads and recovery statuses, accurately evaluating athletes' training statuses and recovery abilities, and thus combining the individual differences of athletes, personalized training program optimization is realized to finely adjust the balance between training load and recovery and improve the effect of athletes' physical training.

[0005] To achieve the above object, the first aspect of the present invention provides a method for managing athletes' physical training based on machine learning, including:

[0006] Collecting historical physical training time series record data of multiple athletes, where the historical physical training time series record data includes physiological index record data, sports performance record data, and recovery status record data, performing time series decomposition on the historical physical training time series record data to obtain multiple groups of physical training trend feature information of each athlete, and performing an association analysis of physical training and recovery on each athlete according to the multiple groups of physical training trend feature information to construct a load recovery adaptability curve for each athlete;

[0007] Respectively constructing a physical training trend feature sequence and a load recovery adaptability feature sequence of the athlete according to the multiple groups of physical training trend feature information and the load recovery adaptability curve of the athlete, and performing joint clustering on multiple athletes according to the physical training trend feature sequence and the load recovery adaptability feature sequence to obtain multiple training mode groups;

[0008] Extract the physical fitness training mode features of each training mode group, generate a mode physical fitness training plan for the training mode group according to the physical fitness training mode features, and optimize the current physical fitness training plan of each athlete based on the load recovery adaptability curve of each athlete and the mode physical fitness training plan of the training mode group to which the athlete belongs, so as to generate a personalized physical fitness training plan for each athlete.

[0009] Preferably, perform an association analysis of physical fitness training and recovery for each athlete according to multiple sets of physical fitness training trend feature information, and construct a load recovery adaptability curve for each athlete, including:

[0010] Extract the long-term association trend data of load recovery, short-term association trend data of load recovery, and periodic fluctuation association trend data of load recovery for each athlete from the physical fitness training trend feature information, and perform association modeling on the long-term association trend data of load recovery, short-term association trend data of load recovery, and periodic fluctuation association trend data of load recovery for each athlete respectively;

[0011] Including performing long-term association modeling on the long-term association trend data of load recovery to construct a long-term trend association model of load recovery, performing short-term association modeling on the short-term association trend data of load recovery to construct a short-term fluctuation association model of load recovery, and performing periodic association modeling on the periodic fluctuation association trend data of load recovery to construct a periodic fluctuation association model of load recovery;

[0012] Fuse the long-term trend association model of load recovery, the short-term fluctuation association model of load recovery, and the periodic fluctuation association model of load recovery to construct a load recovery adaptability curve for each athlete, including determining the long-term trend association curve of load recovery, the short-term fluctuation association curve of load recovery, and the periodic fluctuation association curve of load recovery of the athlete according to the long-term trend association model of load recovery, the short-term fluctuation association model of load recovery, and the periodic fluctuation association model of load recovery, performing weighted fusion on the long-term trend association curve of load recovery, the short-term fluctuation association curve of load recovery, and the periodic fluctuation association curve of load recovery to construct an initial load recovery adaptability curve of the athlete, constructing a target optimization function corresponding to the initial load recovery adaptability curve, and based on the historical physical fitness training time series record data of the athlete, using a target optimization algorithm to optimize the target optimization function to obtain the target fusion weights corresponding to the long-term trend association curve of load recovery, the short-term fluctuation association curve of load recovery, and the periodic fluctuation association curve of load recovery in the initial load recovery adaptability curve, and optimizing the initial load recovery adaptability curve according to the target fusion weights to obtain the load recovery adaptability curve of the athlete.

[0013] Preferably, for the long-term trend correlation model, short-term fluctuation correlation model, and periodic fluctuation correlation model of load recovery, it further includes:

[0014] For the long-term correlation trend data of load recovery, perform the following long-term trend correlation modeling. Through the long-term correlation trend data of load recovery, perform model parameter fitting to construct the long-term trend correlation model of load recovery:

[0015]

[0016] In the formula, R l (x) is the recovery state parameter under the training load parameter x, x0 is the inflection point of the curve, which is the extreme point representing the degree of change in the recovery state, k is the slope of the curve, and L is the upper limit value of the recovery state parameter in the long-term trend;

[0017] For the short-term correlation trend data of load recovery, perform the following short-term fluctuation correlation modeling. Through the short-term correlation trend data of load recovery, perform model parameter fitting to construct the short-term fluctuation correlation model of load recovery:

[0018] V s (t) = A·e -bt + C

[0019] In the formula, V s (t) is the recovery state parameter at time t, A is the amplitude value of the short-term fluctuation, b is the decay rate value, and C is the lower limit value of the recovery state parameter in the short-term fluctuation;

[0020] For the periodic fluctuation correlation trend data of load recovery, perform the following periodic fluctuation correlation modeling. Through the periodic fluctuation correlation trend data of load recovery, perform model parameter fitting to construct the periodic fluctuation correlation model of load recovery:

[0021]

[0022] In the formula, P c (t) is the recovery state parameter at time t, D is the amplitude of the periodic fluctuation, f is the frequency of the fluctuation, is the phase of the fluctuation, and H is the central value of the recovery state parameter in the periodic fluctuation.

[0023] Preferably, perform joint clustering on multiple athletes according to the physical training trend feature sequence and the load recovery adaptability feature sequence to obtain multiple training mode groups, including:

[0024] The DBSCAN clustering algorithm is used to jointly cluster multiple athletes. For the target distance between any two athletes, the initial distance between the physical training trend feature sequences corresponding to the two athletes is calculated, and the correlation parameter between the load recovery adaptability feature sequences corresponding to the two athletes is calculated. Based on the correlation parameter, the initial distance between the physical training trend feature sequences is corrected to obtain the target distance between any two athletes. After calculating the target distance between any two athletes, the DBSCAN clustering algorithm is used to cluster multiple athletes based on the target distance to generate multiple training mode groups.

[0025] Preferably, the physical training mode features of each training mode group are extracted, including:

[0026] A training data set is constructed according to the physical training trend feature sequences and load recovery adaptability feature sequences of multiple athletes in the training mode group, and the group mode feature recognition model corresponding to the training mode group is self-supervised trained through the training data set. The group mode feature recognition model is an MLP model;

[0027] Among them, the physical training trend feature sequences and load recovery adaptability feature sequences of multiple athletes in the training data set are used as the input of the group mode feature recognition model, and the weight coefficient of each athlete is output through the group mode feature recognition model. The training objective is to minimize the total difference between the group features generated by the model prediction and the individual features of multiple athletes. The group features generated by the model prediction are weighted and generated for all athletes based on the output weight coefficients of each athlete. After training the group mode feature recognition model based on the training data set, a target weight set is generated through the group mode feature recognition model, and the physical training trend feature sequences and load recovery adaptability feature sequences of multiple athletes in the training mode group are processed through the target weight set to generate the physical training mode features of the training mode group.

[0028] Preferably, based on the load recovery adaptability curve of each athlete and the mode physical training plan of the training mode group to which the athlete belongs, the current physical training plan of each athlete is optimized to generate the personalized physical training plan of each athlete, including:

[0029] The individual training load data of multiple athletes is determined according to the physical training trend feature sequences, the group training load data is extracted from the mode physical training plan of the training mode group to which the athlete belongs, and the training load difference data of each athlete is generated based on the group training load data;

[0030] Determine the individual training load recovery adaptation data of multiple athletes according to the load recovery adaptation curve, extract the group load recovery adaptation data from the model physical training plan of the training mode group to which the athletes belong, generate the training load recovery adaptation difference data for each athlete based on the group load recovery adaptation data, and optimize the current physical training plan of multiple athletes based on the training load difference data and the training load recovery adaptation difference data to generate the personalized physical training plan for each athlete.

[0031] Preferably, use the STL decomposition algorithm to perform time series decomposition on the historical physical training time series record data to obtain multiple sets of physical training trend feature information for each athlete.

[0032] The second aspect of the present invention provides a machine learning-based athlete physical training management system for executing the above-mentioned machine learning-based athlete physical training management method, including:

[0033] A physical training data acquisition module for collecting the historical physical training time series record data of multiple athletes, where the historical physical training time series record data includes physiological index record data, sports performance record data, and recovery status record data;

[0034] A physical training data analysis module for performing time series decomposition on the historical physical training time series record data to obtain multiple sets of physical training trend feature information for each athlete, performing an association analysis of physical training and recovery for each athlete according to the multiple sets of physical training trend feature information, and constructing the load recovery adaptation curve for each athlete;

[0035] A joint clustering module for constructing the physical training trend feature sequence and the load recovery adaptation feature sequence of the athlete respectively according to the multiple sets of physical training trend feature information and the load recovery adaptation curve of the athlete, performing joint clustering on multiple athletes according to the physical training trend feature sequence and the load recovery adaptation feature sequence to obtain multiple training mode groups;

[0036] A group mode recognition module for extracting the physical training mode features of each training mode group and generating the model physical training plan of the training mode group according to the physical training mode features;

[0037] A personalized plan optimization module for optimizing the current physical training plan of each athlete based on the load recovery adaptation curve of each athlete and the model physical training plan of the training mode group to which the athlete belongs, and generating the personalized physical training plan for each athlete.

[0038] The present invention has the following beneficial effects:

[0039] The present invention performs time series decomposition on the historical physical fitness training data of athletes to extract multiple physical fitness training trend feature information. Based on the physical fitness training trend feature information, an adaptive modeling related to load recovery is performed on different athletes to quantify the dynamic relationship between training load and recovery status. Further, combined with the adaptive modeling information and the physical fitness training trend feature information, joint clustering processing is performed on multiple athletes. The training mode groups are divided by the joint clustering method, and the training mode group pattern features are extracted to generate a group training plan. Finally, combined with the load recovery adaptability curve and training mode features of individual athletes, the training plan for each athlete is optimized individually, which can dynamically and scientifically formulate personalized physical fitness training plans, significantly improve the training effect and safety, reduce the risks of overtraining and insufficient recovery, and provide precise and personalized technical support for the training management of athletes. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 It is a schematic flow chart of a method for managing athletes' physical fitness training based on machine learning provided in one embodiment of the present invention.

[0041] Figure 2 It is a schematic structural diagram of a system for managing athletes' physical fitness training based on machine learning provided in one embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0042] In order to enable those skilled in the art to better understand the technical solutions in the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0043] Please refer to Figure 1 , one implementation of the present invention provides a method for managing athletes' physical fitness training based on machine learning, including the following steps:

[0044] Step S1: Collect the historical physical fitness training time series record data of multiple athletes, perform time series decomposition on the historical physical fitness training time series record data to obtain multiple groups of physical fitness training trend feature information of each athlete, and perform physical fitness training and recovery correlation analysis on each athlete according to the multiple groups of physical fitness training trend feature information to construct the load recovery adaptability curve of each athlete.

[0045] In this embodiment, the historical physical fitness training time series record data of athletes includes at least the recorded data of physiological indicators, sports performance record data, and recovery status record data in different periods. For example, within a historical month or three months, during each training process, through data acquisition devices such as heart rate monitors, positioning devices, accelerometers, etc., the physical fitness training data of multiple athletes is obtained, including the recorded data of physiological indicators such as heart rate, blood pressure, respiratory rate, body temperature, blood oxygen saturation, etc., which reflects the physiological state of athletes during training. For example, during long-term endurance running training, the heart rate data of athletes shows a continuous increase, then stability, and then a continuous decrease; the sports performance record data including training load, training intensity, maximum training output, etc., which reflects the sports performance of athletes. For example, the weightlifting load of an athlete during strength training is recorded as 100 kg for 10 bench presses; the recovery status record data including rest heart rate Hoff duration, RPE score regarding fatigue, etc. For example, the heart rate recovery time recorded by an athlete after high-intensity training and the subjective evaluation of the fatigue after training based on the RPE score of 0-10.

[0046] For the collected historical physical fitness training time series record data, based on time series analysis techniques, methods such as the STL decomposition algorithm or wavelet transform are used to decompose the historical data to obtain multiple sets of physical fitness training trend feature information, such as multi-level trends, seasonal variations, and residual terms. In this embodiment, taking the STL decomposition algorithm as an example, the time series data of each athlete is finally decomposed into the following main features: Long-term trend feature: It represents the long-term development trend of the athlete's training performance, such as the gradual decrease of the resting heart rate, the gradual increase of the maximum heart rate, and the gradual increase of the maximum strength such as the maximum weight that can be lifted with a barbell during long-term training. Short-term fluctuation feature: It represents the short-term fluctuations in the athlete's training, usually related to the short-term changes in training intensity or recovery time, such as the increase and decrease amplitudes of the heart rate during different training cycles, the average speed during different running trainings, etc. Periodic fluctuation feature: It reflects the periodic fluctuations in the athlete's training, such as the change fluctuations of lactic acid concentration, heart rate, etc. during high-intensity load periods and low-intensity recovery periods.

[0047] For the multiple sets of physical fitness training trend feature information obtained from the above decomposition, the correlation analysis is carried out on the physical fitness training load data and the corresponding training recovery data of the athletes included therein, and the load recovery adaptability curve of each athlete is constructed. The core purpose is to analyze the impact of training load on the athlete's recovery status, so as to reflect the dynamic relationship between training load such as training intensity or duration and recovery status such as recovery heart rate or muscle soreness degree.

[0048] Step S2: constructing the athletes' physical training trend feature sequences and load recovery adaptability feature sequences respectively according to the athletes' multiple groups of physical training trend feature information and load recovery adaptability curves, and jointly clustering multiple athletes according to the physical training trend feature sequences and load recovery adaptability feature sequences to obtain multiple training pattern groups.

[0049] In this embodiment, according to the physical training trend characteristic information such as the long-term trend characteristics, short-term fluctuation characteristics and periodic fluctuation characteristics extracted in the above steps, and the load recovery adaptability curve of each athlete, the physical training trend characteristic sequence and the load recovery adaptability characteristic sequence of each athlete are constructed. These characteristic sequences will contain the dynamic changes of the training load and recovery state of the athletes in different time periods. For the physical training trend characteristic sequence, it includes data such as the long-term training trend, short-term fluctuation, and periodic fluctuation of the athletes, the characteristics such as the change slope in the long-term trend, such as the maximum force output growth rate, the resting heart rate decline rate, the characteristics such as the fluctuation amplitude in the short-term fluctuation, such as the change amplitude of the heart rate in a single training, and the characteristics such as the cycle length in the periodic fluctuation, such as the duration of the low-intensity recovery period, so as to reflect the physical training information such as the long-term upward trend, short-term violent fluctuation, and load changes in different training cycles of the athletes in the past period of time. For the load recovery adaptability characteristic sequence, it includes information such as the coordinated change parameters of some variables in the curve, which is generated by fitting the relationship data between the training load and the recovery state, and characterizes the state information such as the recovery speed and recovery potential of the athletes under different recovery intensities.

[0050] After generating the physical training trend feature sequence and load recovery adaptability feature sequence of the athletes, the clustering algorithm is used to cluster the training patterns of multiple athletes to identify the multiple training pattern groups contained therein. In the clustering process, the physical training trend feature sequence representing the training state and the load recovery adaptability feature sequence representing the adaptive relationship between training and load recovery during the training process are comprehensively considered, and multiple athletes are jointly clustered to finally generate multiple training pattern groups. For example, the clustering results divide the athletes into the following training pattern groups: Group 1 corresponds to the training pattern that adapts to high intensity and rapid recovery; Group 2 corresponds to the pattern that adapts to low intensity and long-term endurance training; Group 3 corresponds to the high-load training pattern that requires a longer recovery period.

[0051] Step S3, extract the physical training pattern characteristics of each training pattern group, generate a model physical training plan for the training pattern group according to the physical training pattern characteristics, optimize the current physical training plan of each athlete based on the load recovery adaptability curve of each athlete and the model physical training plan of the training pattern group to which they belong, and generate a personalized physical training plan for each athlete.

[0052] In this embodiment, based on clustering to obtain multiple training mode groups, representative features of each training mode group are extracted. These features will include a comprehensive physical fitness training state representation obtained by fusing multiple trend features and adaptive curve features of members within each group. Representing the changing trends of the training states of multiple athletes over time, as well as information such as the state recovery rules under different training intensities, since the members within the group have similar training state changing trends and state recovery rules, a pattern physical fitness training plan for the training mode group can be formulated based on the physical fitness training mode features of the group as a reference standard. This plan will combine the features of the group, synthesize the training state changing rules of different athletes within the group, and combine relevant state recovery rules, optimize individuals with relatively high or low training intensities, and consider the corresponding training state recovery situations, and plan information such as the training load target and recovery cycle target of the group. For example, if the group is adapted to high intensity and has a relatively fast recovery speed, a training plan including high-intensity interval training, a shorter recovery cycle, and frequent training is formulated according to the overall historical sports situations of different members as a reference target to guide the training of different athletes.

[0053] Based on the group mode plan, personalized optimization is carried out according to the load recovery adaptive curve and unique training characteristics of each athlete. The specific method is to adjust training targets such as training load, training intensity, and recovery time in the pattern physical fitness training plan of the group according to the individual's adaptive curve. For example, if an athlete's individual adaptive curve indicates that their current recovery level is higher than the recovery state level used as a guide for the group, a part of the training load can be appropriately increased, or the recovery time after training can be appropriately shortened, etc. Personalized adjustment is made according to the individual's adaptability on the basis of the group reference state, so as to maximize the training effect of each athlete. Through the above method, a personalized physical fitness training plan is dynamically and scientifically formulated according to the individual differences of athletes and the group mode features, significantly improving the training effect of athletes, reducing the risk of sports injuries, and optimizing the physical fitness training process of athletes.

[0054] In one of the implementation processes, for step S1, correlation analysis of physical fitness training and recovery is performed on each athlete according to multiple sets of physical fitness training trend feature information, and a load recovery adaptive curve of each athlete is constructed, specifically including:

[0055] Extract the long-term correlation trend data of load recovery, short-term correlation trend data of load recovery, and periodic fluctuation correlation trend data of load recovery of each athlete from the physical fitness training trend feature information, and perform correlation modeling on the long-term correlation trend data of load recovery, short-term correlation trend data of load recovery, and periodic fluctuation correlation trend data of load recovery of each athlete respectively.

[0056] In this embodiment, the long-term trend of load recovery during the training process of athletes reflects the influence trend of training load on the long-term recovery ability of athletes. For example, as the training load continues to increase, the recovery ability of athletes may gradually improve, but this improvement will gradually tend to be stable. Therefore, for the multi-group physical fitness training trend characteristic information of athletes, the long-term correlation trend data of the long-term change trend of training load and the corresponding recovery state level are extracted, such as the change data of the corresponding recovery duration as the training load of the athlete gradually increases. Long-term correlation modeling is carried out on the long-term correlation trend data of load recovery, and a basic long-term trend correlation model of load recovery is constructed. Long-term trend correlation modeling is carried out based on the following formula, and the model parameters are fitted through the long-term correlation trend data of load recovery to construct a long-term trend correlation model of load recovery:

[0057]

[0058] In the formula, R l (x) is the recovery state parameter under the training load parameter x, representing the recovery ability of the athlete in the long-term trend. x0 is the inflection point of the curve, which is used to represent the extreme point of the change degree of the recovery state, such as the load value with the fastest change speed of the recovery state. k is the slope of the curve, reflecting the sensitivity of the load to the recovery state. L is the upper limit value of the recovery state parameter in the long-term trend, representing the theoretical maximum value of the recovery ability of the athlete after high-load training. This model can capture the overall influence trend of training load on the recovery ability during the long-term training process of athletes. For example, when the training load gradually increases, the recovery ability of athletes will improve with the adaptation training, but when the training load exceeds a certain value, the growth of the recovery ability will gradually slow down.

[0059] The short-term fluctuation characteristics during the training process of athletes reflect the short-term fluctuation of the training load on the recovery state of athletes, which is usually related to high-intensity interval training or short-term training adjustment. These short-term fluctuations reflect the change of the recovery ability of athletes for single or short-term high-load training. Therefore, from the multi-group physical fitness training trend characteristic information, the short-term correlation trend data of the short-term fluctuation trend of training load and the corresponding recovery state level are extracted. Short-term correlation modeling is carried out on the short-term correlation trend data of load recovery, and a basic short-term fluctuation correlation model of load recovery is constructed. Specifically, an exponential decay model is used to model the short-term fluctuation, and the model parameters are fitted through the short-term correlation trend data of load recovery to construct a short-term fluctuation correlation model of load recovery:

[0060] V s (t) = A·e -bt +C

[0061] In the formula, Vs (t) is the recovery state parameter at time t, representing the athlete's recovery ability within a short period. A is the amplitude value of short-term fluctuations, representing the starting value of the recovery state after high-intensity training. b is the decay rate value, reflecting the speed at which the recovery ability gradually decreases from the high-intensity training state to the stable state. C is the lower limit value of the recovery state parameter in short-term fluctuations, representing the stable level of the recovery state. This model is applicable to describe the short-term recovery process of athletes after a high-intensity training session. For example, the gradual recovery process of an athlete's heart rate or the rate of muscle fatigue relief in the short term can be quantified for its dynamic changes through this model to assist in the short-term adjustment of the training plan.

[0062] The periodic fluctuation characteristics of athletes during the training process are the periodic patterns shown in the training and recovery processes of athletes, such as the peaks and troughs of training loads in different training cycles. By capturing these periodic fluctuation patterns, the dynamic relationship between the load and recovery state of athletes under a periodic training plan can be better understood. Therefore, from multiple sets of physical training trend characteristic information, the load recovery cycle fluctuation correlation trend data containing the periodic fluctuation trend of the training load and the corresponding recovery state level is extracted. Based on the sine function, a periodic fluctuation correlation modeling is performed on the periodic fluctuation to construct a basic load recovery cycle fluctuation correlation model, and then the model parameters are fitted through the load recovery cycle fluctuation correlation trend data to construct the load recovery cycle fluctuation correlation model:

[0063]

[0064] In the formula, P c (t) is the recovery state parameter at time t, representing the athlete's recovery ability under periodic training. D is the amplitude of the periodic fluctuation, representing the intensity of the recovery state fluctuation. f is the frequency of the fluctuation, reflecting the frequency of the periodic fluctuation, is the phase of the fluctuation, representing the starting time point of the fluctuation cycle. H is the central value of the recovery state parameter in the periodic fluctuation, representing the average recovery ability of the athlete in the periodic fluctuation. This model is applicable to analyze the recovery pattern of athletes under a periodic training plan. For example, the alternating arrangement of training days and recovery days in an athlete's weekly training plan may lead to periodic changes in the recovery state. Through this model, the rationality of the periodic training plan can be evaluated, and the training and recovery arrangements can be optimized.

[0065] The above several correlation models respectively quantify the relationship between the training load and recovery state of athletes from the perspectives of long-term trends, short-term fluctuations, and periodic fluctuations. The comprehensive application of these models can help to more comprehensively understand the characteristics of athletes' recovery ability and provide support for the formulation of personalized training programs.

[0066] In this embodiment, in order to comprehensively characterize the dynamic relationship between the training load and the recovery state of athletes, by integrating the long-term trend correlation model of load recovery, the short-term fluctuation correlation model of load recovery, and the periodic fluctuation correlation model of load recovery, the load recovery adaptability curve of athletes is constructed. To comprehensively reflect the long-term recovery trend, short-term fluctuations, and periodic changes of athletes, so as to provide data support and decision-making basis for personalized training management.

[0067] In this process, according to the modeling processes of the above-mentioned long-term trend correlation model of load recovery, short-term fluctuation correlation model of load recovery, and periodic fluctuation correlation model of load recovery, based on the correlation curves obtained after fitting the model parameters through data, the long-term trend correlation curve of load recovery, the short-term fluctuation correlation curve of load recovery, and the periodic fluctuation correlation curve of load recovery of athletes are finally determined.

[0068] In order to comprehensively consider the load recovery ability of athletes on long-term, short-term, and periodic time scales, the long-term trend correlation curve of load recovery, the short-term fluctuation correlation curve, and the periodic fluctuation correlation curve are weighted and integrated to preliminarily construct the load recovery adaptability curve of athletes, obtaining the initial load recovery adaptability curve of athletes. And construct the target optimization function corresponding to the initial load recovery adaptability curve. The design of the target optimization function aims to minimize the error between the recovery state parameters output by the load recovery adaptability curve and the true recovery state parameters in the sample data. Specifically, the mean square error loss function can be used to measure the error to construct the target optimization function. Based on the historical physical training time series record data of multiple athletes as sample data, the target optimization algorithm is used to optimize the target optimization function. In this process, the gradient descent method can be used to solve the target optimization function to iteratively optimize the fusion weights of each curve in the initial load recovery adaptability curve, and finally obtain the optimal values of the fusion weights corresponding to each curve, outputting the target fusion weights corresponding to the long-term trend correlation curve of load recovery, the short-term fluctuation correlation curve of load recovery, and the periodic fluctuation correlation curve of load recovery. Through the optimized weights, the contributions of different time scales to the load recovery adaptability curve can be dynamically adjusted, making the final curve more in line with the actual recovery state of athletes.

[0069] Finally, according to the multiple target fusion weights generated by optimization, the long-term trend correlation curve of load recovery, the short-term fluctuation correlation curve, and the periodic fluctuation correlation curve are re-integrated to optimize the initial load recovery adaptability curve, generating the load recovery adaptability curve of athletes. Some curve parameters obtained by fitting the sample data included therein can characterize the state information such as the recovery speed and recovery potential of athletes under different recovery intensities. Through the load recovery adaptability curve, the recovery ability of athletes on long-term, short-term, and periodic time scales can be comprehensively reflected.

[0070] In one of the implementation processes, for step S2, in order to group multiple athletes and extract the characteristics of different training mode groups, a joint clustering method based on the physical training trend feature sequence and the load recovery adaptability feature sequence is adopted. The correlation between the training trend characteristics and the load recovery characteristics of athletes is comprehensively considered, and the accuracy of the clustering result is ensured by a distance correction method. Specifically, multiple athletes are jointly clustered according to the physical training trend feature sequence and the load recovery adaptability feature sequence to obtain multiple training mode groups, including:

[0071] The DBSCAN clustering algorithm is used to jointly cluster multiple athletes. In this process, the target distance between multiple athletes is first calculated, that is, the target distance between any two athletes. The calculation method of the target distance includes the following steps:

[0072] The initial distance between the physical training trend feature sequences corresponding to two athletes is calculated. For example, the Euclidean distance between the physical training trend feature sequences is used to obtain the initial distance between two athletes. Then, the correlation parameter between the load recovery adaptability feature sequences corresponding to two athletes is further calculated to represent the similarity degree between the two in terms of recovery characteristics. For example, the Euclidean distance between the feature sequences is calculated as the correlation parameter in the same way as the initial distance. And the initial distance between the physical training trend feature sequences is corrected based on the correlation parameter. For example, the correlation parameter is used as a weight for weighted correction to obtain the target distance between any two athletes, so that the similarity of the physical training trend characteristics and the load recovery adaptability characteristics of the two athletes can be comprehensively considered in the target distance, thereby making the final clustering result more comprehensive and accurate.

[0073] After calculating the target distance between any two athletes, multiple target distances are used as the input of the DBSCAN clustering algorithm. Based on the target distance, the DBSCAN clustering algorithm is used to cluster multiple athletes. Finally, the athletes are divided into multiple training mode groups to generate multiple training mode groups. Through the joint clustering method based on the physical training trend feature sequence and the load recovery adaptability feature sequence, the training mode groups of athletes can be accurately divided. The distance correction method of the target distance effectively integrates the correlation between the training trend characteristics and the recovery adaptation characteristics, providing key data support for the subsequent extraction of group mode characteristics and the optimization of personalized training programs.

[0074] In one of the implementation processes, for step S3, based on generating multiple training mode groups through joint clustering, in order to further extract the physical training mode features of each training mode group, a self-supervised learning method based on a group mode feature recognition model is proposed. This method can comprehensively analyze the individual characteristics of multiple athletes within the group and generate an overall mode feature that can represent the training mode group through weighted fusion. The specific implementation steps are as follows:

[0075] Based on the physical training trend feature sequences and load recovery adaptability feature sequences of multiple athletes in the training mode group, as the basic data for constructing the training dataset, the self-supervised training of the group mode feature recognition model corresponding to the training mode group is carried out through the training dataset. The group mode feature recognition model uses a multi-layer perceptron (MLP) as the basic structure. Through a series of non-linear transformations, MLP learns the relationship between the individual characteristics of multiple athletes and the group mode features.

[0076] Among them, the physical training trend feature sequences and load recovery adaptability feature sequences of multiple athletes in the training dataset are used as the input of the group mode feature recognition model, and the weight coefficient of each athlete is output through the group mode feature recognition model. The training objective is to minimize the total difference between the group features generated by model prediction and the individual characteristics of multiple athletes. In this process, for the group features generated by model prediction, they are weighted and generated for all athletes based on the output weight coefficients of each athlete. The weight coefficient of each athlete represents the importance of the athlete in the group features. The total difference between the group features and the individual characteristics of multiple athletes can be comprehensively expressed by calculating the Euclidean distances between multiple individuals and the group features respectively, so as to construct the loss function of the model. In the model training process, a gradient descent algorithm such as the Adam optimizer is used to optimize the loss function, iteratively adjust the parameters of the model, and finally generate the weight coefficient of each athlete, obtaining the target weight set containing the weight coefficients corresponding to multiple members respectively. The physical training trend feature sequences and load recovery adaptability feature sequences of multiple athletes in the training mode group are processed through the target weight set, so that the physical training mode features of the training mode group generated by weighted generation in the target weight set, and this group feature includes the long-term trend feature, short-term fluctuation feature, periodic fluctuation feature of the group, as well as the dynamic relationship between load and recovery, which can comprehensively describe the physical training mode of the group and provide a basis for the subsequent optimization of personalized training programs.

[0077] In one of the implementation processes, for step S3, based on the physical training mode characteristics of the training mode group, a physical training plan for the training mode group can be generated, and a personalized physical training plan can be generated for each athlete based on the physical training plan of the training mode group. The purpose of this process is to comprehensively consider the individual characteristics of the athletes, the mode characteristics of the group they belong to, and the differences between individuals and the group. By reasonably designing the training load, recovery adaptation, and optimizing the process, the personalized needs of each athlete can be more accurately met, and the training effect and safety can be improved.

[0078] Specifically, based on the load recovery adaptability curve of each athlete and the physical training plan of the training mode group to which the athlete belongs, the current physical training plan of each athlete is optimized to generate a personalized physical training plan for each athlete, including the following content:

[0079] Determine the individual training load data of multiple athletes according to the physical training trend feature sequence, extract the group training load data from the physical training plan of the training mode group to which the athletes belong, and generate the training load difference data of each athlete based on the group training load data;

[0080] In this embodiment, multiple trend information included in the physical training trend feature sequence is used to extract the long-term trend, short-term fluctuation, and periodic fluctuation data related to the training load as the individual training load data. The long-term trend includes the overall change trend of the individual's training load, such as whether the load increases steadily. The short-term fluctuation includes the fluctuation of the individual's training load in the short term, such as the load fluctuations in high-intensity interval training. The periodic fluctuation includes the periodic law of the individual's training load, such as the training peaks and troughs per week. For example, for a certain athlete, the long-term trend shows that the load gradually increases, and the periodic fluctuation shows that there are two peak loads per week. For the physical training plan of the training mode group, it also includes the long-term trend characteristics, short-term fluctuation characteristics, and periodic fluctuation characteristics that conform to the training load of the group, which are used as the group training load data. This represents the overall law of the training load of the athletes in the group. For example, the long-term trend of the group shows that the load increases steadily as a whole, and the short-term fluctuation of the group is relatively small. The periodic fluctuation shows that there is one peak load per week, which can be used as a reference benchmark for optimizing the individual training load.

[0081] By performing a difference analysis on the group training load data and the individual training load data of the athletes, the differences between the specific compliance of the athlete individuals in the long term and the training cycle and the group comprehensive level can be determined. For example, whether the long-term load growth rate is higher or lower than the group comprehensive level, and the training load differences in high-intensity training weeks or low-intensity training weeks during the training cycle.

[0082] Then, based on the load recovery adaptability curve, the individual training load recovery adaptation data of multiple athletes are determined, which includes information such as recovery rate, recovery time, and recovery level, representing the speed at which the athlete recovers to a stable state after training, the time required to recover to a stable state, and the stable level data after recovery. For example, if an athlete has a slower recovery rate, it takes a longer time for the heart rate to recover to a stable state after training, and a longer recovery time indicates a lower stable level. Similarly, the group load recovery adaptation data is extracted from the pattern physical training plan of the training mode group to which the athlete belongs, reflecting the average recovery characteristics of the athletes within the group, and can be used as a reference for individual recovery adaptation optimization. Then, based on the group load recovery adaptation data, the training load recovery adaptation difference number of each athlete is generated, and the difference between its recovery characteristics and the group benchmark is characterized by the training load recovery adaptation difference number. For example, if an athlete's recovery rate is lower than the group benchmark, it means that he needs a longer recovery time, and the training arrangement needs to appropriately extend the rest period. Finally, based on the training load difference data and the training load recovery adaptation difference data, the current physical training plan of multiple athletes is optimized to generate the personalized physical training plan for each athlete.

[0083] In this embodiment, after calculating the training load difference data and the training load recovery adaptation difference data of the athlete, the current training plan of the athlete is optimized based on these two data to generate a personalized physical training plan. The following optimization principles can be used as a reference: For training load optimization, if the individual training load is higher than the group benchmark and the recovery adaptability is poor, the training intensity is reduced or the training cycle is extended. If the individual training load is lower than the group benchmark and the recovery adaptability is good, the training intensity is appropriately increased or the cycle is shortened. For recovery adaptation optimization, the training cycle and rest time are adjusted according to the difference in recovery adaptability characteristics. Athletes with poor recovery ability increase the recovery time, and athletes with good recovery ability shorten the recovery time. Finally, the optimized training load data and recovery adaptation data are integrated to generate the personalized physical training plan for each athlete. The personalized physical training plan can include information such as training intensity, training cycle, and training goal.

[0084] Exemplarily, for a certain athlete, the training intensity is a short-term load reduction to reduce load fluctuations, the training cycle is to reduce the weekly training peak to once, and the intermittent recovery time is extended, and the training goal is to improve the recovery rate and load stability within the next month. By calculating the training load difference and the recovery adaptation difference, a machine learning-based athlete physical training management method provided by the present invention can accurately identify the deviation between the individual and the group, and generate a targeted training plan in combination with the individualized optimization rules, which can significantly improve the scientificity and personalization level of training compared with the traditional standardized training plan, and ensure that each athlete achieves the best training effect on the premise of safety.

[0085] Please refer to Figure 2 , one implementation of the present invention also provides a machine learning-based physical training management system for athletes, specifically for implementing a machine learning-based physical training management method provided in the above implementation process. The system includes:

[0086] A physical training data acquisition module for collecting historical physical training time series record data of multiple athletes. The historical physical training time series record data includes physiological index record data, sports performance record data, and recovery status record data;

[0087] A physical training data analysis module for performing time series decomposition on the historical physical training time series record data to obtain multiple sets of physical training trend feature information for each athlete, performing physical training and recovery correlation analysis on each athlete according to the multiple sets of physical training trend feature information, and constructing a load recovery adaptability curve for each athlete;

[0088] A joint clustering module for constructing a physical training trend feature sequence and a load recovery adaptability feature sequence for athletes respectively according to the multiple sets of physical training trend feature information and the load recovery adaptability curve of the athletes, and performing joint clustering on multiple athletes according to the physical training trend feature sequence and the load recovery adaptability feature sequence to obtain multiple training mode groups;

[0089] A group mode recognition module for extracting the physical training mode features of each training mode group and generating a mode physical training plan for the training mode group according to the physical training mode features;

[0090] A personalized plan optimization module for optimizing the current physical training plan of each athlete based on the load recovery adaptability curve of each athlete and the mode physical training plan of the training mode group to which the athlete belongs, and generating a personalized physical training plan for each athlete.

[0091] The above is only the specific implementation manner of the present invention, enabling those skilled in the art to understand or implement the present invention. Various modifications to these embodiments will be obvious to those skilled in the art. The parts not described in detail in this specification belong to the prior art well-known to those skilled in the art.

Claims

1. A method for managing an athlete's physical fitness training based on machine learning, characterized in that, Including: Collecting the historical time-series record data of the physical fitness training of multiple athletes. The historical time-series record data of physical fitness training includes physiological index record data, sports performance record data, and recovery status record data. Decompose the historical time-series record data of physical fitness training by time series to obtain multiple sets of physical fitness training trend feature information for each athlete. Conduct an association analysis of physical fitness training and recovery for each athlete based on the multiple sets of physical fitness training trend feature information, and construct the load recovery adaptability curve for each athlete; Construct the physical fitness training trend feature sequence and the load recovery adaptability feature sequence of the athlete respectively according to the multiple sets of physical fitness training trend feature information and the load recovery adaptability curve of the athlete. Conduct joint clustering on multiple athletes according to the physical fitness training trend feature sequence and the load recovery adaptability feature sequence to obtain multiple training mode groups; Extract the physical fitness training mode features of each training mode group, generate the mode physical fitness training plan of the training mode group according to the physical fitness training mode features, and optimize the current physical fitness training plan of each athlete based on the load recovery adaptability curve of each athlete and the mode physical fitness training plan of the training mode group to which the athlete belongs, and generate the personalized physical fitness training plan for each athlete.

2. The method for managing the physical fitness training of athletes based on machine learning according to claim 1, wherein, Conduct an association analysis of physical fitness training and recovery for each athlete based on the multiple sets of physical fitness training trend feature information, and construct the load recovery adaptability curve for each athlete, including: Extract the long-term association trend data of load recovery, the short-term association trend data of load recovery, and the periodic fluctuation association trend data of load recovery for each athlete from the physical fitness training trend feature information, and conduct association modeling on the long-term association trend data of load recovery, the short-term association trend data of load recovery, and the periodic fluctuation association trend data of load recovery for each athlete respectively; Including conducting long-term association modeling on the long-term association trend data of load recovery and constructing the long-term trend association model of load recovery, conducting short-term association modeling on the short-term association trend data of load recovery and constructing the short-term fluctuation association model of load recovery, and conducting periodic association modeling on the periodic fluctuation association trend data of load recovery and constructing the periodic fluctuation association model of load recovery; Fuse the long-term trend correlation model of load recovery, the short-term fluctuation correlation model of load recovery, and the periodic fluctuation correlation model of load recovery to construct the load recovery adaptability curve for each athlete, including determining the long-term trend correlation curve of the athlete's load recovery, the short-term fluctuation correlation curve of the load recovery, and the periodic fluctuation correlation curve of the load recovery according to the long-term trend correlation model of load recovery, the short-term fluctuation correlation model of load recovery, and the periodic fluctuation correlation model of load recovery; performing weighted fusion on the long-term trend correlation curve of load recovery, the short-term fluctuation correlation curve of load recovery, and the periodic fluctuation correlation curve of load recovery to construct the initial load recovery adaptability curve of the athlete; constructing the target optimization function corresponding to the initial load recovery adaptability curve; based on the historical physical training time series record data of the athlete, using the target optimization algorithm to optimize the target optimization function to obtain the target fusion weights corresponding to the long-term trend correlation curve of load recovery, the short-term fluctuation correlation curve of load recovery, and the periodic fluctuation correlation curve of load recovery in the initial load recovery adaptability curve respectively; and optimizing the initial load recovery adaptability curve according to the target fusion weights to obtain the load recovery adaptability curve of the athlete.

3. The method for managing an athlete's physical fitness training based on machine learning according to claim 2, characterized in that For the long-term trend correlation model of load recovery, the short-term fluctuation correlation model of load recovery, and the periodic fluctuation correlation model of load recovery, it also includes: For the long-term correlation trend data of load recovery, perform the following long-term trend correlation modeling, and construct the long-term trend correlation model of load recovery by fitting the model parameters through the long-term correlation trend data of load recovery: where R l (x) is the recovery state parameter under the training load parameter x, x0 is the inflection point of the curve, which is used to represent the extreme point of the change degree of the recovery state, k is the slope of the curve, and L is the upper limit value of the recovery state parameter in the long-term trend; For the short-term correlation trend data of load recovery, perform the following short-term fluctuation correlation modeling, and construct the short-term fluctuation correlation model of load recovery by fitting the model parameters through the short-term correlation trend data of load recovery: V s (t) = A·e -bt + C where V s (t) is the recovery state parameter at time t, A is the amplitude value of the short-term fluctuation, b is the attenuation rate value, and C is the lower limit value of the recovery state parameter in the short-term fluctuation; For the periodic fluctuation correlation trend data of load recovery, perform the following periodic fluctuation correlation modeling, and construct the periodic fluctuation correlation model of load recovery by fitting the model parameters through the periodic fluctuation correlation trend data of load recovery: Where P c (t) is the recovery state parameter at time t, D is the amplitude of the periodic fluctuation, f is the frequency of the fluctuation, is the phase of the fluctuation, and H is the central value of the recovery state parameter in the periodic fluctuation.

4. A method for managing an athlete's physical fitness training based on machine learning according to claim 1, characterized in that, Perform joint clustering on multiple athletes according to the physical training trend feature sequence and the load recovery adaptability feature sequence to obtain multiple training mode groups, including: Use the DBSCAN clustering algorithm to perform joint clustering on multiple athletes. For the target distance between any two athletes, calculate the initial distance between the physical training trend feature sequences corresponding to the two athletes respectively, and calculate the correlation parameter between the load recovery adaptability feature sequences corresponding to the two athletes respectively. Based on the correlation parameter, correct the initial distance between the physical training trend feature sequences to obtain the target distance between any two athletes. After calculating the target distance between any two athletes, use the DBSCAN clustering algorithm to cluster multiple athletes based on the target distance to generate multiple training mode groups.

5. A method for managing an athlete's physical training based on machine learning according to claim 1, characterized in that, Extract the physical training mode features of each training mode group, including: Construct a training dataset based on the physical training trend feature sequences and load recovery adaptability feature sequences of multiple athletes in the training mode group, and perform self-supervised training on the group mode feature recognition model corresponding to the training mode group through the training dataset. The group mode feature recognition model is an MLP model; Among them, use the physical training trend feature sequences and load recovery adaptability feature sequences of multiple athletes in the training dataset as the input of the group mode feature recognition model, and output the weight coefficient of each athlete through the group mode feature recognition model. The training objective is to minimize the total difference between the group features generated by model prediction and the individual features of multiple athletes. The group features generated by model prediction are weighted and generated for all athletes based on the output weight coefficients of each athlete. After training the group mode feature recognition model based on the training dataset, generate a target weight set through the group mode feature recognition model, and process the physical training trend feature sequences and load recovery adaptability feature sequences of multiple athletes in the training mode group through the target weight set to generate the physical training mode features of the training mode group.

6. The method for managing an athlete's physical training based on machine learning according to claim 5, wherein, Optimize the current physical training plan of each athlete based on the load recovery adaptability curve of each athlete and the pattern physical training plan of the training mode group to which the athlete belongs, and generate a personalized physical training plan for each athlete, including: Determine the individual training load data of multiple athletes according to the physical training trend feature sequences, extract the group training load data from the pattern physical training plan of the training mode group to which the athletes belong, and generate the training load difference data of each athlete based on the group training load data; Determine the individual training load recovery adaptation data of multiple athletes according to the load recovery adaptability curve, extract the group load recovery adaptation data from the pattern physical training plan of the training mode group to which the athletes belong, generate the training load recovery adaptation difference data of each athlete based on the group load recovery adaptation data, and optimize the current physical training plan of multiple athletes based on the training load difference data and the training load recovery adaptation difference data to generate a personalized physical training plan for each athlete.

7. A method for managing the physical training of athletes based on machine learning according to claim 1, characterized in that, Use the STL decomposition algorithm to perform time series decomposition on the historical physical training time series record data to obtain multiple sets of physical training trend feature information of each athlete.

8. An athlete physical fitness training management system based on machine learning, characterized in that, A method for managing athletes' physical training based on machine learning according to any one of claims 1-7 above, including: A physical training data acquisition module for acquiring historical physical training time series record data of multiple athletes. The historical physical training time series record data includes physiological index record data, sports performance record data, and recovery status record data; A physical training data analysis module for performing time series decomposition on the historical physical training time series record data to obtain multiple sets of physical training trend feature information of each athlete, performing an association analysis of physical training and recovery on each athlete according to the multiple sets of physical training trend feature information, and constructing a load recovery adaptability curve for each athlete; A joint clustering module, which is used to respectively construct a physical training trend feature sequence and a load recovery adaptability feature sequence of an athlete according to multiple groups of physical training trend feature information and load recovery adaptability curves of the athlete, and perform joint clustering on multiple athletes according to the physical training trend feature sequence and the load recovery adaptability feature sequence to obtain multiple training mode groups; A group mode recognition module, which is used to extract the physical training mode features of each training mode group and generate a mode physical training plan for the training mode group according to the physical training mode features; A personalized plan optimization module, which is used to optimize the current physical training plan of each athlete based on the load recovery adaptability curve of each athlete and the mode physical training plan of the training mode group to which the athlete belongs, and generate a personalized physical training plan for each athlete.

Images