A sports data intelligent analysis method

Abstract

The present invention relates to the field of data analysis, and specifically to a method for intelligent analysis of sports data. A method for intelligent analysis of sports data includes the steps of training strategy prediction, training strategy simulation iteration, and real-time training of a training strategy prediction model. The present invention predicts subsequent training strategies through time series prediction, and adjusts the predicted training strategy data through group optimization simulation calculation, outputs subsequent training strategies that are more suitable for athletes, and realizes intelligent analysis of sports data such as training strategy data; at the same time, at the monitoring time point, the training strategy prediction model is trained in real time through the training strategy data at the current monitoring time point and the optimal training strategy data, so that the predicted training data output by the training strategy prediction model is more suitable for the actual situation of the athletes.

Description

Technical Field

[0001] The present invention relates to the field of data analysis, and in particular to a sports data intelligent analysis method. Background Art

[0002] In today's sports field, with the continuous improvement of competitive level, traditional training methods and strategies can no longer meet the needs of athletes and coaches for high efficiency and high results. Traditional methods usually rely on the experience of coaches to formulate training plans. Although this method is effective to a certain extent, it lacks accurate data support and is difficult to achieve personalized and refined management. In addition, due to the lack of in-depth analysis of training data, problems may not be discovered and adjusted in time during training, which affects the training effect and even increases the risk of injury to athletes. Summary of the invention

[0003] The present invention predicts subsequent training strategies through time series prediction, and adjusts the predicted training strategy data through group optimization simulation calculation, outputs subsequent training strategies that are more suitable for athletes, and realizes intelligent analysis of sports data such as training strategy data; at the same time, at the monitoring time point, the training strategy prediction model is trained in real time through the training strategy data at the current monitoring time point and the optimal training strategy data, so that the predicted training data output by the training strategy prediction model is more suitable for the actual situation of the athletes.

[0004] A sports data intelligent analysis method, comprising:

[0005] Obtain the training strategy data at the current monitoring time point; combine the training strategy data obtained at the current monitoring time point with the training strategy data obtained N times before to form a training strategy prediction data set, and then send the training strategy prediction data set to the training strategy prediction model for processing, and output the predicted training strategy data. The training strategy prediction model is established based on the Transformer model;

[0006] Perform group optimization simulation calculation based on the predicted training strategy data and output the optimal training strategy data;

[0007] At the same time, the training strategy prediction model is trained in real time through the optimal training strategy data, which specifically includes the following contents: at the monitoring time point, the training strategy data at the current monitoring time point is spliced ​​with the optimal training strategy data to construct effect evaluation data, and then the effect evaluation data is sent to the effect evaluation model for processing to output the effect evaluation value; the training strategy gradient value of the effect evaluation value to the optimal training strategy data is calculated, and then the parameters of the training strategy prediction model are adjusted based on the training strategy gradient value using the gradient ascent method to realize real-time training of the training strategy prediction model; and the effect evaluation model is trained in real time at the monitoring time point.

[0008] Preferably, the effect evaluation model includes a feature extraction layer, a physiological parameter change layer, a training effect change layer and an effect evaluation value output layer, the feature extraction layer is used to perform feature extraction operations on the effect evaluation data to construct an effect evaluation vector; the physiological parameter change layer is used to predict physiological parameter changes to construct a physiological parameter change feature vector; the training effect change layer is used to predict training effect changes to construct a training effect change feature vector; the effect evaluation value output layer is used to output the effect evaluation value; wherein the feature extraction layer, the physiological parameter change layer, the training effect change layer and the effect evaluation value output layer are all established based on the BP neural network.

[0009] Preferably, training the effect evaluation model specifically includes the following steps:

[0010] Obtain positive effect evaluation samples and negative effect evaluation samples, both of which include training strategy data at two adjacent monitoring time points and corresponding physiological parameter change data and training effect change data, the difference being that the positive effect evaluation samples are the training strategy data at two adjacent monitoring time points when no training risk occurs, and the negative effect evaluation samples are the training strategy data at two adjacent monitoring time points when training risk occurs, annotate the positive effect evaluation samples with positive labels, and annotate the negative effect evaluation samples with negative labels; form a first training set with all annotated positive effect evaluation samples and negative effect evaluation samples;

[0011] The feature extraction layer and the physiological parameter change layer are trained by the first training set. During the training, the physiological parameter change data is used as a target, the physiological parameter change loss value is calculated, and it is determined whether the physiological parameter change loss value is within the preset range of the physiological parameter change. If the physiological parameter change loss value is within the preset range of the physiological parameter change, the parameters in the physiological parameter change layer are fixed; otherwise, the feature extraction layer and the physiological parameter change layer are continuously trained by the first training set;

[0012] The feature extraction layer and the training effect change layer are trained by the first training set. During the training, the training effect change data is used as a target, the training effect change loss value is calculated, and it is determined whether the training effect change loss value is within a preset range of the training effect change. If the training effect change loss value is within the preset range of the training effect change, the parameters in the training effect change layer are fixed; otherwise, the feature extraction layer and the training effect change layer are continuously trained by the first training set;

[0013] The parameters in the physiological parameter change layer and the training effect change layer are fixed, and the effect evaluation model is trained through the first training set. During the training, the corresponding positive labels and negative labels are used as targets, and the first loss value is calculated to determine whether the first loss value is within the first preset range. If the first loss value is within the first preset range, the trained effect evaluation model is output; otherwise, the effect evaluation model is continued to be trained through the first training set.

[0014] Preferably, real-time training is performed on the effect evaluation model, specifically including the following steps:

[0015] Obtain positive real-time samples and negative real-time samples of effect evaluation for all athletes in the same area. Both the positive real-time samples and the negative real-time samples include the training strategy data at the current monitoring time point and the previous monitoring time point, as well as the corresponding physiological parameter change data and training effect change data. The difference is that the positive real-time samples for effect evaluation are the training strategy data at the current monitoring time point and the previous monitoring time point when no training risk occurs, and the negative real-time samples for effect evaluation are the training strategy data at the current monitoring time point and the previous monitoring time point when training risk occurs. All the positive real-time samples and the negative real-time samples for effect evaluation are combined into a second training set; obtain the positive individual samples and the negative individual samples of effect evaluation corresponding to the athletes to be analyzed. Athletes refer to athletes to be analyzed. Both the positive individual samples for effect evaluation and the negative individual samples for effect evaluation include the training strategy data of two adjacent monitoring time points corresponding to the athletes to be analyzed and the corresponding physiological parameter change data and training effect change data. The difference is that the positive individual samples for effect evaluation are the training strategy data of two adjacent monitoring time points corresponding to the athletes to be analyzed who have not incurred training risks, and the negative individual samples for effect evaluation are the training strategy data of two adjacent monitoring time points corresponding to the athletes to be analyzed who have incurred training risks. The positive individual samples for effect evaluation are annotated with positive labels, and the negative individual samples for effect evaluation are annotated with negative labels, and all the annotated positive individual samples for effect evaluation and negative individual samples for effect evaluation form a third training set;

[0016] The feature extraction layer and the physiological parameter change layer are trained by the second training set. During the training, the physiological parameter change data is used as the target, the real-time loss value of the physiological parameter change is calculated, and it is determined whether the real-time loss value of the physiological parameter change is within the preset range of the physiological parameter change. If the real-time loss value of the physiological parameter change is within the preset range of the physiological parameter change, the parameters in the physiological parameter change layer are fixed; otherwise, the feature extraction layer and the physiological parameter change layer are continued to be trained by the second training set;

[0017] The feature extraction layer and the training effect change layer are trained by the second training set. During the training, the training effect change data is used as the target, the real-time loss value of the training effect change is calculated, and it is determined whether the real-time loss value of the training effect change is within the preset range of the training effect change. If the real-time loss value of the training effect change is within the preset range of the training effect change, the parameters in the training effect change layer are fixed; otherwise, the feature extraction layer and the training effect change layer are continued to be trained by the second training set;

[0018] The parameters in the physiological parameter change layer and the training effect change layer are fixed, and the effect evaluation model is trained through the third training set. During the training, the corresponding positive labels and negative labels are used as targets, and the second loss value is calculated to determine whether the second loss value is within the second preset range. If the second loss value is within the second preset range, the real-time trained effect evaluation model is output; otherwise, the effect evaluation model is continued to be trained in real time through the third training set.

[0019] Preferably, training is performed on the training strategy prediction model, specifically including the following steps:

[0020] Obtain training strategy prediction training samples, the training strategy prediction training samples include N+1 training strategy data sorted by time without any training risk, all the training strategy prediction training samples are combined into a fourth training set, and the training strategy prediction model with initialized parameters is trained through the fourth training set. During the training, the last training strategy prediction training sample in the training strategy prediction training samples is used as the target, and the third loss value is calculated to determine whether the third loss value is within the third preset range. If the third loss value is within the third preset range, the real-time trained training strategy prediction model is output; otherwise, the real-time training of the training strategy prediction model continues to be performed through the fourth training set.

[0021] Preferably, performing a population optimization simulation calculation based on the predicted training strategy data and outputting the optimal training strategy data specifically includes the following steps:

[0022] Construct several simulated individuals F, including training strategy data items f i , i=1, 2, 3, …, I, I is the total number of training strategy data items; all simulated individuals F form a population set, and set the maximum number of iterations;

[0023] Calculate the fitness of simulated individual F, and record the simulated individual with the largest fitness as simulated individual B;

[0024] The simulated individuals F in the population set are iteratively updated through the particle swarm optimization algorithm. The specific iterative formula is:

[0025] v i =w·vi +c1·η(p i -f i )+c2·η(b i -f i );

[0026] f i =f i +v i ;

[0027] where v i is the speed value; w is the inertia factor, c1 and c2 are the first learning factor and the second learning factor respectively, b i is the i-th training strategy data item in the simulated individual B;

[0028] When the number of iterations reaches the maximum number of iterations, the simulated individual F with the largest fitness value is output as the optimal training strategy data.

[0029] Preferably, the specific construction method of the simulated individual F is as follows: the predicted training strategy data is recorded as P, and the training strategy data item in the predicted training strategy data P is recorded as p i , generating f i =p i +η(f_max-f_min), and then all training strategy data items f i The simulated individual F is composed, where η is a random number between 0 and 1, f_max is the maximum value corresponding to the i-th training strategy data item, and f_min is the minimum value corresponding to the i-th training strategy data item.

[0030] Preferably, the fitness corresponding to the simulated individual F is calculated, and the specific calculation method is as follows: the training strategy data at the current monitoring time point is spliced ​​with the simulated individual F to construct effect evaluation data, and then the effect evaluation data is sent to the effect evaluation model for processing to output the effect evaluation value, and the effect evaluation value is used as the fitness corresponding to the simulated individual F.

[0031] The present invention has the following advantages:

[0032] The present invention predicts subsequent training strategies through time series prediction, and adjusts the predicted training strategy data through group optimization simulation calculation, outputs subsequent training strategies that are more suitable for athletes, and realizes intelligent analysis of sports data such as training strategy data; at the same time, at the monitoring time point, the training strategy prediction model is trained in real time through the training strategy data at the current monitoring time point and the optimal training strategy data, so that the predicted training data output by the training strategy prediction model is more suitable for the actual situation of the athletes. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1A schematic diagram of a flow chart of a sports data intelligent analysis method adopted in an embodiment of the present invention. DETAILED DESCRIPTION

[0034] In order to enable persons 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.

[0035] A sports data intelligent analysis method, such as Figure 1 As shown, including:

[0036] Get the training strategy data at the current monitoring time point. The training strategy data specifically includes training type, training duration and training intensity. The training type refers to the items performed during training, such as sprinting, stretching and swimming. Different athletes have different corresponding training types. The training duration refers to the length of time for each training session. The training intensity refers to the intensity when the training is performed, including low intensity, medium intensity, high intensity and very high intensity. The training intensity is generally controlled by a heart rate monitor. The heart rate is maintained at 50%-60% of the maximum heart rate for low intensity, 60%-75% of the maximum heart rate for medium intensity, 75%-95% of the maximum heart rate for high intensity, and more than 95% of the maximum heart rate for very high intensity. The acquired training strategy data and the training strategy data acquired N times before form a training strategy prediction data set, and then the training strategy prediction data set is sent to the training strategy prediction model for processing, and the predicted training strategy data is output. The training strategy prediction model is established based on the Transformer model; as the training progresses, the training strategy of each type of athlete will show regular changes, such as changes in training type, changes in training intensity, etc., so the training strategy can be predicted by time series prediction; it should be noted that the training strategy data acquired at the current monitoring time point can be the training strategy data being executed on the day, and the monitoring time point is divided into days, that is, the monitoring time point is not a moment, but a time period;

[0037] Perform group optimization simulation calculation based on the predicted training strategy data and output the optimal training strategy data. It should be noted that the predicted training strategy data can be used as a priori data to guide the group optimization simulation iterative calculation, thereby improving the convergence speed; athletes can be trained directly through the optimal training strategy data, and the optimal training strategy data can also be used to provide a reference for the training of athletes, so as to facilitate the coach to arrange training plans;

[0038] At the same time, the training strategy prediction model is trained in real time through the optimal training strategy data, which specifically includes the following contents: at the monitoring time point, the training strategy data at the current monitoring time point is spliced ​​with the optimal training strategy data to construct effect evaluation data, and then the effect evaluation data is sent to the effect evaluation model for processing to output the effect evaluation value. By processing the training strategy data and the optimal training strategy data through the effect evaluation model, the impact of the training strategy change on the sport can be analyzed. The impact on athletes includes two aspects: training effect and training risk. Based on these impacts, the parameters inside the training strategy prediction model can be adjusted; the training strategy gradient value of the effect evaluation value to the optimal training strategy data is calculated. The calculation method is generally to calculate the derivative of the effect evaluation value to the optimal training strategy data, and use the calculated derivative as the training strategy gradient value, and then use the gradient ascent method based on the training strategy gradient value to adjust the parameters of the training strategy prediction model to achieve real-time training of the training strategy prediction model; and the effect evaluation model is trained in real time at the monitoring time point;

[0039] The present application predicts subsequent training strategies through time series prediction, and adjusts the predicted training strategy data through group optimization simulation calculation, outputs subsequent training strategies that are more suitable for athletes, and realizes intelligent analysis of sports data such as training strategy data; at the same time, at the monitoring time point, the training strategy prediction model is trained in real time through the training strategy data at the current monitoring time point and the optimal training strategy data, so that the predicted training data output by the training strategy prediction model is more in line with the actual situation of the athletes.

[0040] The effect evaluation model includes a feature extraction layer, a physiological parameter change layer, a training effect change layer and an effect evaluation value output layer, wherein the physiological parameter change layer and the training effect change layer are parallel tasks, the feature extraction layer is used to perform feature extraction operations on the effect evaluation data to construct an effect evaluation vector; the physiological parameter change layer is used to predict physiological parameter changes to construct a physiological parameter change feature vector, the physiological parameters may specifically include heart rate, blood pressure and HRV, etc., these parameters can reflect the training risks of athletes; the training effect change layer is used to predict training effect changes to construct a training effect change feature vector, the training effect is generally described by test results, such as time data for running events; the effect evaluation value output layer is used to output the effect evaluation value; wherein the feature extraction layer, the physiological parameter change layer, the training effect change layer and the effect evaluation value output layer are all established based on the BP neural network.

[0041] Training the effect evaluation model includes the following steps:

[0042] Obtain positive samples and negative samples for effect evaluation. Both the positive samples and the negative samples include training strategy data at two adjacent monitoring time points and corresponding physiological parameter change data and training effect change data. The difference is that the positive samples for effect evaluation are training strategy data at two adjacent monitoring time points when no training risk occurs, and the negative samples for effect evaluation are training strategy data at two adjacent monitoring time points when training risk occurs. Training risks generally include injuries and mental health problems during training. Both physiological parameter change data and training effect change data are obtained based on actual records. The positive samples for effect evaluation are labeled with positive labels, which are training effect change data. The negative samples for effect evaluation are labeled with negative labels, which are generally 0. All labeled positive samples for effect evaluation and negative samples for effect evaluation form the first training set.

[0043] The feature extraction layer and the physiological parameter change layer are trained by the first training set. During the training, the physiological parameter change data is used as a target, the physiological parameter change loss value is calculated, and it is determined whether the physiological parameter change loss value is within the preset range of the physiological parameter change. If the physiological parameter change loss value is within the preset range of the physiological parameter change, the parameters in the physiological parameter change layer are fixed; otherwise, the feature extraction layer and the physiological parameter change layer are continuously trained by the first training set;

[0044] The feature extraction layer and the training effect change layer are trained by the first training set. During the training, the training effect change data is used as a target, the training effect change loss value is calculated, and it is determined whether the training effect change loss value is within a preset range of the training effect change. If the training effect change loss value is within the preset range of the training effect change, the parameters in the training effect change layer are fixed; otherwise, the feature extraction layer and the training effect change layer are continuously trained by the first training set;

[0045] The parameters in the physiological parameter change layer and the training effect change layer are fixed, and the effect evaluation model is trained through the first training set. During the training, the corresponding positive label and negative label are used as targets, and a first loss value is calculated to determine whether the first loss value is within a first preset range. If the first loss value is within the first preset range, the trained effect evaluation model is output; otherwise, the effect evaluation model is continuously trained through the first training set;

[0046] The above preset ranges of physiological parameter changes, training effect changes and the first preset range are all determined by developers; through the above training method, it is possible to learn the impact of changes in training strategies on exercise, that is, changes in physiological parameters and changes in training effects.

[0047] Real-time training of the effect evaluation model includes the following steps:

[0048] Obtain positive real-time samples and negative real-time samples of effect evaluation for all athletes in the same area. Both the positive real-time samples and the negative real-time samples include the training strategy data at the current monitoring time point and the previous monitoring time point, as well as the corresponding physiological parameter change data and training effect change data. The difference is that the positive real-time samples for effect evaluation are the training strategy data at the current monitoring time point and the previous monitoring time point when no training risk occurs, and the negative real-time samples for effect evaluation are the training strategy data at the current monitoring time point and the previous monitoring time point when a training risk occurs. All the positive real-time samples and the negative real-time samples for effect evaluation form a second training set. It should be noted that the same area can be the same training venue. The same training venue has similar environmental conditions and has a certain impact on training. The training strategy data of athletes in the same area will be uploaded for subsequent use. Apply; obtain positive individual samples and negative individual samples of effect evaluation corresponding to the athletes to be analyzed, the athletes to be analyzed refer to the athletes to be analyzed, and both the positive individual samples and negative individual samples include the training strategy data of two adjacent monitoring time points corresponding to the athletes to be analyzed and the corresponding physiological parameter change data and training effect change data, the difference being that the positive individual samples of effect evaluation are the training strategy data of two adjacent monitoring time points corresponding to the athletes to be analyzed who have not incurred training risks, and the negative individual samples of effect evaluation are the training strategy data of two adjacent monitoring time points corresponding to the athletes to be analyzed who have incurred training risks, the positive individual samples of effect evaluation are labeled with positive labels, the negative individual samples of effect evaluation are labeled with negative labels, and all labeled positive individual samples and negative individual samples of effect evaluation form a third training set;

[0049] The feature extraction layer and the physiological parameter change layer are trained by the second training set. During the training, the physiological parameter change data is used as the target, the real-time loss value of the physiological parameter change is calculated, and it is determined whether the real-time loss value of the physiological parameter change is within the preset range of the physiological parameter change. If the real-time loss value of the physiological parameter change is within the preset range of the physiological parameter change, the parameters in the physiological parameter change layer are fixed; otherwise, the feature extraction layer and the physiological parameter change layer are continued to be trained by the second training set;

[0050] The feature extraction layer and the training effect change layer are trained by the second training set. During the training, the training effect change data is used as the target, the real-time loss value of the training effect change is calculated, and it is determined whether the real-time loss value of the training effect change is within the preset range of the training effect change. If the real-time loss value of the training effect change is within the preset range of the training effect change, the parameters in the training effect change layer are fixed; otherwise, the feature extraction layer and the training effect change layer are continued to be trained by the second training set;

[0051] The parameters in the physiological parameter change layer and the training effect change layer are fixed, and the effect evaluation model is trained through the third training set. During the training, the corresponding positive labels and negative labels are used as targets, and the second loss value is calculated to determine whether the second loss value is within the second preset range. If the second loss value is within the second preset range, the second preset range is also determined by the developer, and the real-time trained effect evaluation model is output; otherwise, the effect evaluation model is continued to be trained in real time through the third training set.

[0052] This application can perform real-time training on the effect evaluation model and can combine the sports information of all athletes in the area and the individual sports information to adaptively adjust the effect evaluation model, so that the effect evaluation model can more accurately evaluate the training strategy.

[0053] Training the training strategy prediction model includes the following steps:

[0054] Obtain training strategy prediction training samples, the training strategy prediction training samples include N+1 training strategy data sorted by time without training risks, all training strategy prediction training samples are combined into a fourth training set, and the training strategy prediction model with initialized parameters is trained through the fourth training set. During the training, the last training strategy prediction training sample in the training strategy prediction training samples is used as the target, and the third loss value is calculated to determine whether the third loss value is within a third preset range. If the third loss value is within the third preset range, the third preset range is also determined by the developer, and the real-time trained training strategy prediction model is output; otherwise, the training strategy prediction model is continued to be trained in real time through the fourth training set.

[0055] Perform group optimization simulation calculation based on the predicted training strategy data and output the optimal training strategy data, which specifically includes the following steps:

[0056] Construct several simulated individuals F, including training strategy data items f i , i=1, 2, 3, …, I, I is the total number of training strategy data items. The specific construction method of the simulated individual F is as follows: the predicted training strategy data is recorded as P, and the training strategy data item in the predicted training strategy data P is recorded as p i, generating f i =p i +η(f_max-f_min), and then all training strategy data items f i The simulated individuals F are formed, where η is a random number between 0 and 1, f_max is the maximum value corresponding to the i-th training strategy data item, and f_min is the minimum value corresponding to the i-th training strategy data item. It should be added that since the training type is generally one-hot encoding, it will be rounded after calculation, and the training intensity is divided into intervals; all simulated individuals F are formed into a population set, and the maximum number of iterations is set;

[0057] Calculate the fitness of simulated individual F, and record the simulated individual with the largest fitness as simulated individual B;

[0058] Calculate the fitness corresponding to the simulated individual F. The specific calculation method is as follows: splice the training strategy data at the current monitoring time point with the simulated individual F to construct effect evaluation data, and then send the effect evaluation data to the effect evaluation model for processing to output the effect evaluation value, and use the effect evaluation value as the fitness corresponding to the simulated individual F;

[0059] The simulated individuals F in the population set are iteratively updated through the particle swarm optimization algorithm. The specific iterative formula is:

[0060] v i =w·v i +c1·η(p i -f i )+c2·η(b i -f i );

[0061] f i =f i +v i ;

[0062] where v i is the velocity value, the initial v i Determined by the developer; w is the inertia factor, which is generally determined in advance by the developer and can also change linearly according to the number of iterations. c1 and c2 are the first learning factor and the second learning factor, respectively, and are generally both 2. b i is the i-th training strategy data item in the simulated individual B;

[0063] When the number of iterations reaches the maximum number of iterations, the simulated individual F with the largest fitness value is output as the optimal training strategy data.

[0064] It should be understood that those skilled in the art can make improvements or changes based on the above description, and all such improvements and changes should fall within the scope of protection of the appended claims of the present invention. Parts not described in detail in this specification belong to the prior art known to those skilled in the art.

Claims

1. A sports data intelligent analysis method, characterized in that: include: Get the training strategy data at the current monitoring time point; The training strategy data obtained at the current monitoring time point and the training strategy data obtained at the previous N times are combined into a training strategy prediction data set, and then the training strategy prediction data set is sent to the training strategy prediction model for processing, and the predicted training strategy data is output. The training strategy prediction model is established based on the Transformer model; Perform group optimization simulation calculation based on the predicted training strategy data and output the optimal training strategy data; At the same time, the training strategy prediction model is trained in real time through the optimal training strategy data, which specifically includes the following contents: at the monitoring time point, the training strategy data at the current monitoring time point is spliced ​​with the optimal training strategy data to construct effect evaluation data, and then the effect evaluation data is sent to the effect evaluation model for processing to output the effect evaluation value; Calculate the training strategy gradient value of the effect evaluation value to the optimal training strategy data, and then adjust the parameters of the training strategy prediction model using the gradient ascent method based on the training strategy gradient value to achieve real-time training of the training strategy prediction model; and perform real-time training on the effect evaluation model at the monitoring time point; The effect evaluation model includes a feature extraction layer, a physiological parameter change layer, a training effect change layer and an effect evaluation value output layer, wherein the feature extraction layer is used to perform feature extraction operations on the effect evaluation data to construct an effect evaluation vector; the physiological parameter change layer is used to predict physiological parameter changes to construct a physiological parameter change feature vector; the training effect change layer is used to predict training effect changes to construct a training effect change feature vector; the effect evaluation value output layer is used to output an effect evaluation value; wherein the feature extraction layer, the physiological parameter change layer, the training effect change layer and the effect evaluation value output layer are all established based on a BP neural network; Training the effect evaluation model includes the following steps: Obtain positive effect evaluation samples and negative effect evaluation samples, both of which include training strategy data at two adjacent monitoring time points and corresponding physiological parameter change data and training effect change data, the difference being that the positive effect evaluation samples are the training strategy data at two adjacent monitoring time points when no training risk occurs, and the negative effect evaluation samples are the training strategy data at two adjacent monitoring time points when training risk occurs, annotate the positive effect evaluation samples with positive labels, and annotate the negative effect evaluation samples with negative labels; form a first training set with all annotated positive effect evaluation samples and negative effect evaluation samples; The feature extraction layer and the physiological parameter change layer are trained by the first training set. During the training, the physiological parameter change data is used as a target, the physiological parameter change loss value is calculated, and it is determined whether the physiological parameter change loss value is within the preset range of the physiological parameter change. If the physiological parameter change loss value is within the preset range of the physiological parameter change, the parameters in the physiological parameter change layer are fixed; otherwise, the feature extraction layer and the physiological parameter change layer are continuously trained by the first training set; The feature extraction layer and the training effect change layer are trained by the first training set. During the training, the training effect change data is used as a target, the training effect change loss value is calculated, and it is determined whether the training effect change loss value is within a preset range of the training effect change. If the training effect change loss value is within the preset range of the training effect change, the parameters in the training effect change layer are fixed; otherwise, the feature extraction layer and the training effect change layer are continuously trained by the first training set; The parameters in the physiological parameter change layer and the training effect change layer are fixed, and the effect evaluation model is trained through the first training set. During the training, the corresponding positive label and negative label are used as targets, and a first loss value is calculated to determine whether the first loss value is within a first preset range. If the first loss value is within the first preset range, the trained effect evaluation model is output; otherwise, the effect evaluation model is continuously trained through the first training set; Training the training strategy prediction model includes the following steps: Obtain training strategy prediction training samples, the training strategy prediction training samples include N+1 training strategy data sorted by time without any training risk, all the training strategy prediction training samples are combined into a fourth training set, and the training strategy prediction model with initialized parameters is trained through the fourth training set. During the training, the last training strategy prediction training sample in the training strategy prediction training samples is used as the target, and the third loss value is calculated to determine whether the third loss value is within the third preset range. If the third loss value is within the third preset range, the real-time trained training strategy prediction model is output; otherwise, the real-time training of the training strategy prediction model continues to be performed through the fourth training set.

2. A sports data intelligent analysis method according to claim 1, characterized in that: Real-time training of the effect evaluation model includes the following steps: Obtain positive real-time samples and negative real-time samples of effect evaluation for all athletes in the same area. Both the positive real-time samples and the negative real-time samples include the training strategy data at the current monitoring time point and the previous monitoring time point, as well as the corresponding physiological parameter change data and training effect change data. The difference is that the positive real-time samples for effect evaluation are the training strategy data at the current monitoring time point and the previous monitoring time point when no training risk occurs, and the negative real-time samples for effect evaluation are the training strategy data at the current monitoring time point and the previous monitoring time point when training risk occurs. All the positive real-time samples and the negative real-time samples for effect evaluation are combined into a second training set; obtain the positive individual samples and the negative individual samples of effect evaluation corresponding to the athletes to be analyzed. Athletes refer to athletes to be analyzed. Both the positive individual samples for effect evaluation and the negative individual samples for effect evaluation include the training strategy data of two adjacent monitoring time points corresponding to the athletes to be analyzed and the corresponding physiological parameter change data and training effect change data. The difference is that the positive individual samples for effect evaluation are the training strategy data of two adjacent monitoring time points corresponding to the athletes to be analyzed who have not incurred training risks, and the negative individual samples for effect evaluation are the training strategy data of two adjacent monitoring time points corresponding to the athletes to be analyzed who have incurred training risks. The positive individual samples for effect evaluation are annotated with positive labels, and the negative individual samples for effect evaluation are annotated with negative labels, and all the annotated positive individual samples for effect evaluation and negative individual samples for effect evaluation form a third training set; The feature extraction layer and the physiological parameter change layer are trained by the second training set. During the training, the physiological parameter change data is used as the target, the real-time loss value of the physiological parameter change is calculated, and it is determined whether the real-time loss value of the physiological parameter change is within the preset range of the physiological parameter change. If the real-time loss value of the physiological parameter change is within the preset range of the physiological parameter change, the parameters in the physiological parameter change layer are fixed; otherwise, the feature extraction layer and the physiological parameter change layer are continued to be trained by the second training set; The feature extraction layer and the training effect change layer are trained by the second training set. During the training, the training effect change data is used as the target, the real-time loss value of the training effect change is calculated, and it is determined whether the real-time loss value of the training effect change is within the preset range of the training effect change. If the real-time loss value of the training effect change is within the preset range of the training effect change, the parameters in the training effect change layer are fixed; otherwise, the feature extraction layer and the training effect change layer are continued to be trained by the second training set; The parameters in the physiological parameter change layer and the training effect change layer are fixed, and the effect evaluation model is trained through the third training set. During the training, the corresponding positive labels and negative labels are used as targets, and the second loss value is calculated to determine whether the second loss value is within the second preset range. If the second loss value is within the second preset range, the real-time trained effect evaluation model is output; otherwise, the effect evaluation model is continued to be trained in real time through the third training set.

3. A sports data intelligent analysis method according to claim 1, characterized in that: Perform group optimization simulation calculation based on the predicted training strategy data and output the optimal training strategy data, which specifically includes the following steps: Construct several simulated individuals F, including training strategy data items f i , i=1, 2, 3, …, I, I is the total number of training strategy data items; all simulated individuals F form a population set, and set the maximum number of iterations; Calculate the fitness of simulated individual F, and record the simulated individual with the largest fitness as simulated individual B; The simulated individuals F in the population set are iteratively updated through the particle swarm optimization algorithm. The specific iterative formula is: v i =w·v i +c1·η(p i -f i )+c2·η(b i -f i ); f i =f i +v i ; where v i is the speed value; w is the inertia factor, c1 and c2 are the first learning factor and the second learning factor respectively, b i is the i-th training strategy data item in the simulated individual B; When the number of iterations reaches the maximum number of iterations, the simulated individual F with the largest fitness value is output as the optimal training strategy data.

4. A sports data intelligent analysis method according to claim 3, characterized in that: The specific construction method of the simulated individual F is as follows: the predicted training strategy data is recorded as P, and the training strategy data item in the predicted training strategy data P is recorded as p i , generating f i =p i +η(f_max-f_min), and then all training strategy data items f i The simulated individual F is composed, where η is a random number between 0 and 1, f_max is the maximum value corresponding to the i-th training strategy data item, and f_min is the minimum value corresponding to the i-th training strategy data item.

5. A sports data intelligent analysis method according to claim 3, characterized in that: Calculate the fitness corresponding to the simulated individual F. The specific calculation method is as follows: splice the training strategy data at the current monitoring time point with the simulated individual F to construct effect evaluation data, and then send the effect evaluation data to the effect evaluation model for processing to output the effect evaluation value, and use the effect evaluation value as the fitness corresponding to the simulated individual F.

Images