Optimization of training load and recovery strategies for track and field sprinters by genetic algorithm

As an ancient and passionate sport, track and field has always been loved and concerned by people. The physical training of track and field athletes is particularly important, which is not only related to the athletes’ competitive performance, but also related to the athletes’ physical quality and health development [1-2]. Scientific training in athletics can improve the physical quality of athletes, and the quality level of athletes affects the development level of athletics. Athletes are at the age of adolescence, and coaches need to combine the physical and mental characteristics and character traits of adolescents, focus on stimulating the athletes’ enthusiasm for sports in the training process, enhance the athletes’ special skills through training, and be able to win in track and field competitions and promote the development of competitive sports [3-5].

In track and field sports, coaches need to strengthen the training of athletes’ physical quality, which mainly refers to the training of athletes’ physical fitness. Physical training is of great significance in athletics [6-7]. Through scientific and reasonable physical training programs, athletes’ endurance, speed, strength, and coordination can be effectively enhanced, thus improving game performance. Personalized training programs can stimulate the enthusiasm and participation of athletes and help them achieve better training results. Secondly, physical training also helps athletes to release negative emotions, develop strong character and teamwork spirit, and improve overall quality. Thirdly, physical training also plays an important role in preventing injuries and diseases in athletics. Through regular physical training, athletes’ muscle strength, flexibility, and body balance can be enhanced [8-10]. Finally, physical training also helps to improve athletes’ body posture and movement techniques and reduce the risk of injury caused by incorrect movement postures. By improving the body ’s adaptability and stress resistance, physical training can help athletes better adapt to the intensity of training and competition and reduce the risk of injury [11-12].

In the training process, coaches give athletes demonstrations, and athletes imitate them, constantly digging out their potential in training and finding suitable training methods on the basis of self-innovation and breakthrough. Physical quality training for athletes in track and field includes local training and overall training. Local training includes the training of speed, strength, lung capacity, agility, static stretching, and coordination, and the barbell press training helps to improve the athletes’ leg muscle strength, and the overall physical quality of the athletes can be improved through scientific training to enhance the athletes’ adaptability in track and field competitions so that the athletes can win the sports events [13-16]. To enhance the fun of training in the training process, the training enthusiasm of athletes will be higher, but overloaded physical training will frustrate the enthusiasm of athletes to train, so choosing a scientific training method can enhance the efficiency of physical training and enhance the physical quality of athletes. In the training process, coaches should strengthen the guidance of athletes to do a good job in the coordination of sports training and cultural learning time so as to cultivate comprehensive development of comprehensive talents [17-19].

In this paper, in order to realize the effective prediction of the training load of track and field sprinters and formulate the recovery strategy accordingly, the IAGABP algorithm is designed by combining the genetic algorithm and the BP neural network, and the training load prediction method of athletes based on the IAGABP algorithm is constructed. Firstly, the selection operator of the genetic algorithm is improved to ensure the convergence rate of the genetic algorithm and the diversity of the population. Then, the improved genetic algorithm is used to optimize the initial weights and thresholds randomly generated by the BP neural network, and the optimal weights and thresholds are selected to greatly improve the accuracy of the neural network. The method is applied to the actual training of athletes after the dataset is tested, and finally, six training recovery strategies are proposed, such as formulating a reasonable training program.

2

Method

2.1

Principle of BP Neural Network

2.1.1

BP Neural Network

BP neural network is a kind of feedback neural network that passes backward on the basis of the multi-layer feed-forward neural network, and its main training algorithm is an algorithm that continuously adjusts the weights and thresholds according to the error reversal [20]. It is widely used in many fields, such as classification in data-supervised learning, image compression, pattern recognition, and function approximation. It is mainly composed of 3 parts, i.e., input layer, hidden layer, and output layer. Its network structure is shown in Fig. 1. BP neural network has good self-learning and self-adaptive ability, strong fault tolerance, and is easy to construct.

The BP neural network algorithm operation consists of two parts: signal forward propagation and backpropagation. Firstly, the input feature signal is input, and the output is obtained through the implicit layer operation processing. Then the weight threshold is continuously adjusted according to the error backward until the set termination condition is reached. The operation process is as follows:

1)

Signal forward propagation process

The input calculation of the i st node of the implicit layer is shown in equation (1): (1) $n e t_{i} = \sum_{j = 1}^{M} w_{i j} x_{j} + θ_{i}$

Where W_ij denotes the weights between the input layer of the j nd node to the hidden layer of the i rd node, x_j denotes the eigenvalues of the input layer of the j th node, and θ_i denotes the threshold of the i th hidden layer node.

The output of the i th hidden layer node is calculated as shown in equation (2): (2) $y_{i} = ϕ (\sum_{j = 1}^{M} w_{i j} x_{j} + θ_{i})$

The input of the k st output layer node is calculated as shown in equation (3): (3) $n e t_{k} = \sum_{i = 1}^{q} w_{k i} ϕ (\sum_{j = 1}^{M} w_{i j} x_{j} + θ_{i}) + a_{k}$

w_i denotes the weights between the k nd output layer node to the i rd hidden layer node, a_k denotes the threshold of the k th output layer node, ϕ(x) denotes the hidden layer activation function, ψ(x) denotes the output layer activation function.

The output of the k th output layer node is shown in equation (4): (4) $o_{k} = ψ (\sum_{i = 1}^{q} w_{k i} ϕ (\sum_{j = 1}^{M} w_{i j} x_{j} + θ_{i}) + a_{k})$

The activation function used is the Sigmoid function, as shown in equation (5): (5) $f (x) = \frac{1}{1 + e^{- x}}$

2)

Error back propagation process

Firstly, the error between the results of each output layer and the target value is calculated. Then, the gradient descent method is used to continuously optimize the weights and thresholds between the connections of each layer according to the back propagation of the mean square error in order to achieve the optimal model. The cycle is repeated until the pre-set termination conditions are adjusted. The final optimal model trained can predict values close to the expected values through the input data.

Each sample p error judgment criterion function is Ep, as shown in equation (6): (6) $E_{p} = \frac{1}{2} \sum_{k = 1}^{L} {(T_{k} - o_{k})}^{2}$

Then, the total error function for a total of P training samples is shown in equation (7): (7) $E = \frac{1}{2} \sum_{p = 1}^{P} \sum_{k = 1}^{L} {(T_{k}^{p} - o_{k}^{p})}^{2}$

2.1.2

Levenberg-Marquardt Optimization of BP

Aiming at the shortcomings of traditional BP neural networks using the gradient descent method, various optimization methods are proposed, such as the adaptive learning algorithm, particle swarm optimization algorithm, and error function correction algorithm. After several literature studies and comparisons, the BP neural network can be better optimized by using the L-M learning algorithm, which is essentially a combination of the gradient descent method and the Gauss-Newton method and can be automatically adjusted. In the process of network training and learning, the gradient descent method descends rapidly in the previous steps, and its gradient tends to zero when it is close to the optimal value. At this time, the objective function declines slowly when close to the optimal value of the Newton method, which can generate a better exploration of the direction can adaptively adjust the gradient descent process.

2.1.3

Normalization

Before training, neural networks need to preprocess the training data and test data, i.e., normalization. The main purpose of normalization is to eliminate the differences between data. In practice, due to the different data sizes and dimensions, high-dimensional data will weaken the low-dimensional data, resulting in bad final classification results. If the data is too big will lead to long training time and slow convergence, so it must be normalized.

Normalization is mainly to map the original data between [0, 1] or [-1, 1].

In this thesis, the most value normalization method is used as shown in equation (8): (8) $X = \frac{X - \min}{\max - \min}$

2.1.4

Cross-validation

The main purpose of cross-validation is to enable a small amount of data set to train the model effectively and improve the model’s generalization ability. The basic idea is to divide the data into two parts: one part is the training set, and the other part is the test set. The training set is used to train the model and the test set is used to evaluate the goodness of the trained model.

The main cross-validation methods are:

Holdout validation is one of the simplest methods. The idea is to slice the dataset once. One part is used as training data, and the other part is used as test data.

Leave-one-out is the most common method in cross-validation. The basic idea is to take out one sample as model validation data and the remaining samples as training data, and this is repeated until each sample is validated once.

K-fold cross-validation: this method divides the dataset into K parts on average, each time from the K parts of the dataset to take out 1 part as the test data and the rest of the K-1 parts as the training data, repeated K times.

In this paper, 4-fold cross-validation is mainly used to divide the dataset.

2.2

Genetic algorithms

2.2.1

Overview of genetic algorithms

A genetic algorithm is an optimization method that simulates natural selection and genetic mechanisms and is commonly used to solve complex optimization problems. Its basic idea is to evolve individuals that are more and more adapted to the environment generation by generation by simulating the process of biological evolution [21]. The genetic algorithm is generalized and summarized to form the basic framework of genetic algorithm genetic algorithm since its proposal, scholars in various countries have paid extensive attention, and after decades of research and development, significant progress has been made in theory and application. Nowadays, genetic algorithms have been widely used in many fields, including production scheduling, data mining, machine learning, etc., because of their strong global search and parallel processing ability. Its powerful search capability and global optimization characteristics make the genetic algorithm a powerful tool for solving practical problems. In genetic algorithms, each potential solution in the solution space is called an individual, usually represented by a chromosome. The chromosome consists of genes that correspond to the characteristics or parameters of the problem. Overall, genetic algorithms have become a mature and effective optimization algorithm that provides an efficient way to solve a variety of complex problems. With the continuous development of artificial intelligence and computer science, genetic algorithms will continue to play an important role in the future and combine with other optimization techniques to provide more possibilities for solving complex problems.

2.2.2

Genetic Algorithm Solution Flow

The flow of the genetic algorithm is shown in Figure 2.

1) Chromosome coding: coding is the solution to the problem expressed as a string of gene sequences, generating a language that can be recognized by the computer according to different laws and will produce different chromosomes. These gene strings are equivalent to chromosomes in nature, according to the basic data for chromosome coding.

2) Generation of initialization population: generate a set of initial solutions as the starting point of the algorithm. Each individual in the population corresponds to a candidate solution, which can be designed according to the characteristics and requirements of the problem.

3) Fitness evaluation: calculate the fitness of each individual in the population and review their problem-solving ability. The goal is to make individuals with high fitness more likely to be selected in the evolutionary process because individuals with high fitness are usually more helpful in problem-solving.

4) Execute the selection, crossover, and mutation operators: select a certain number of individuals as parents based on the fitness value, randomly select two individuals from the selected parents, perform a crossover operation to generate a new individual based on the selected crossover method, and then perform a random mutation operation on the chromosomes of the new individual to change the chromosome value of the individual.

5) Evaluate the resulting new population for fitness: evaluating the new population can determine which individuals have higher chances of survival and reproduction in the next generation of the population, and if the condition is satisfied, it is the optimal solution, and if it is not satisfied, then repeat 4).

2.3

Athlete training load prediction based on the IAGABP algorithm

2.3.1

Selection of Indicators

The training load prediction model for track and field sprinters needs to take the athlete’s physiological and biochemical data, athletic level indicators, and training content as the basis for predicting the extent of the athlete’s training load and judging whether the currently scheduled training volume is reasonable.

The required data are shown in Figure 3. The data needed for the experiment are divided into basic data of athletes, training data and physiological and biochemical data.

2.3.2

Factors Affecting Training Loads

In the process of predicting the training load of athletes, the selection of factors that have an impact on the training load should not only be reasonable but also representative [22]. Combined with the characteristics of the experimental data, this paper selects the factors that have an impact on the training load according to the recommendations of the coach of the national team of track and field sprints:

1) Athletes’ height, weight, age, gender, and body mass index.

2) Functional threshold power (FTP). FTP in track and field sprint training is defined as the maximum average power obtained during a 1-hour ride at full and steady effort.

3) Maximum oxygen uptake. Maximum oxygen uptake is an important indicator of an athlete’s aerobic capacity.

4) Average and maximum pedaling frequency. Pedaling frequency is a very critical but easily ignored indicator in the field of track and field sprinting, which refers to the number of revolutions per minute of the unilateral crank. It is the basic guarantee of riding efficiency. Reasonable pedaling frequency can effectively use the body functions, prevent muscle and knee injuries, and improve the efficiency of riding.

5) Average power and maximum power. Power for track and field sprint training is a very clear indicator of intensity, which, since it reflects the athlete’s current output of power, can effectively monitor the intensity of the movement.

6) Physiological and biochemical indicators. There are many physiological and biochemical data in the data, which contains several major items such as blood routine, blood glucose muscle enzymes, liver function, urine routine, testosterone measurement, etc., and each of which contains dozens of test indicators respectively.

7) Participation in training programs and training volume.

2.3.3

BP neural network structure and parameter design

In order to establish the athlete training load prediction model, the structure and parameter design of the BP neural network should be determined first. Its design mainly includes the number of network layers, the number of neurons, data set processing and the selection of initial weights and thresholds, etc. The network structure of the prediction model in this paper is shown in Fig. 4.

1)

Number of network layers

The structure of the BP neural network consists of an input layer, a hidden layer, and an output layer, where the hidden layer can be one or multiple layers. The three-layer BP neural network can approximate the continuous function in any closed interval, and at the same time, because the prediction network needs to be applied in the system, the training time can not be too long, so this paper chooses the basic three-layer structure.

2)

Determination of the number of neurons in the input and output layers

The prediction target of the model is the loading result of the training volume, which is used to describe the degree of the athlete’s load on the training, so the number of neurons in the output layer is 3, and the loading result is categorized to represent the training volume is too small, the training volume is moderate, and the training volume is too large.

3)

Determination of the number of hidden layer neurons

The number of hidden layer neurons can be calculated based on empirical formulas, and then the trial method is used to determine the optimal number of hidden layer neurons. The empirical formula for determining the number of hidden layer neurons is shown below: (9) $n = \sqrt{m l}$ (10) $n = \log_{2} m$ (11) $n = \sqrt{m + l} + a$

4)

Data set pre-processing

Due to the large variability in the range of values of the screened data, the data of different metrics are not comparable and cannot be directly used as input data for BP neural networks. The data need to be normalized before use, and this paper uses normalization. The purpose of normalization is to map the data into the same interval after the algorithm operation, such as [0, 1]. The formula of the Min-max normalization method is as follows: (12) $x = \frac{x_{i} - x_{m i n}}{x_{m a x} - x_{m i n}}$

5)

Selection of initial weights and thresholds

Due to the nonlinear characteristics of the BP neural network itself, whether the model can converge, whether it falls into a local minimum, and the length of the training time are all closely related to the selection of the initial weights and thresholds. In this paper, the initial selection of weights and thresholds is optimized using an improved adaptive genetic algorithm, the details of which are described below.

2.3.4

Optimization of BP Neural Networks

In this chapter, the specific methods of optimizing the BP neural network using an improved adaptive genetic algorithm are as follows: the selection operation of the adaptive genetic algorithm is generally chosen in the way of roulette, while this paper proposes a simple and effective selection operator.

1)

Coding and initial population generation

Since the initial weights and thresholds of the BP neural network are numbers in the range of (0, 1), the use of binary encoding methods will make the chromosome very long, and the efficiency of the algorithm will be greatly reduced. This paper chooses to use real number encoding. Real number coding not only eliminates the decoding work but also facilitates the calculation of genetic operations.

2)

Adaptation function

BP neural network to minimize the prediction error of the athlete training load for the purpose of regulating the connection weights and thresholds, and adaptive genetic algorithm applied in this paper aims to find out a set of weights and thresholds to minimize the prediction error, so the fitness function should be inversely proportional to the sum of the squares of the output error of the network, the smaller the error, the greater the fitness, and the formula for the calculation of the degree of individual fitness is shown in (13): (13) $F (x) = \frac{1}{E (x) + 1}$

The formula x denotes the current chromosome and E(x) denotes the sum of error squares of the network output when the BP neural network uses a set of weights and thresholds represented by the current chromosome.

3)

Improve the selection operator

This paper introduces the optimal preservation strategy to improve the selection operator on the basis of the sorting selection method, and the operation steps of the improved selection operator are as follows:

(1) Initialize a population and sort the individuals in the population according to the order of live responsiveness from smallest to largest:

(2) Divide the ranked individuals equally into 4 segments, and the mass of the population in the 4 segments becomes higher from front to back, and select each segment according to the ratio of 0.4, 0.6, 0.8, and 1.

(3) The individuals selected in the previous step are formed into a new population. The number of individuals in the new population is reduced due to the proportional selection. The number of individuals in the new population is kept constant by randomly selecting individuals from the optimal individuals in the 4th segment to be added to the new population.

A graphical representation of the above steps is used, where each grid represents a chromosome and the numbers within the grid represent fitness values:

4)

Crossover and variation operators

An adaptive genetic algorithm generates new individuals through crossover and mutation operations of the population and selects the individuals for the operations according to the crossover rate and mutation rate in the process of crossover and mutation in order to construct individuals with higher fitness values, the smaller the fitness of an individual is, the higher is the probability of crossover and mutation, and the opposite is true of an individual with higher fitness. In this paper, we calculate the crossover rate p_c and variation rate P_m as follows: (14) $p_{c} = {\begin{array}{l} p_{a} - \frac{(p_{a} - p_{b}) (f - f_{a v g})}{f_{\max} - f_{a v g}}, f \geq f_{a v g} \\ p_{a}, f < f_{a v g} \end{array}$ (15) $p_{m} = {\begin{array}{l} p_{i} - \frac{(p_{i} - p_{j}) (f_{\max} - f)}{f_{\max} - f_{a v g}}, f \geq f_{a v g} \\ p_{i}, f < f_{a v g} \end{array}$

In the above equation, p_a, p_b, p_i, p_j is a constant and takes the value of p_a = 0.9, p_b = 0.6, p_i = 0.1, p_j = 0.001. f represents the current individual fitness, and f_max, f_avg represents the maximum fitness and average fitness of all individuals in the population, respectively.

The BP neural network algorithm IAGABP optimized based on an improved adaptive genetic algorithm is divided into the genetic algorithm part and the BP neural network part. The flowchart of the algorithm is shown in Fig. 5.

3

Results and Discussion

3.1

Simulation experiment of the neural network model in this paper

Three kinds of networks were used to fit the function, and 4000 groups of data were randomly generated, of which 3900 groups were used as training samples and 100 groups were used as prediction samples. The parameters of the genetic algorithm are set as follows: the population size is 40, the number of evolutionary generations is 40, the crossover probability is 0.6, and the mutation probability is 0.1, and the best effect is proved through a large number of experiments when the parameters of the rank selection method P0=0.4, δ=0.2, at this time, the selection probability of four grades of excellent, good, medium and poor are 1, 0.8, 0.6 and 0.4, respectively. The parameters of the BP neural network model are set as follows: The number of nodes in the input layer is 2, the number of nodes in the hidden layer is 5, the number of nodes in the output layer is 1, the maximum number of training times is 100, the accuracy of the training target is 0.0001, and the learning rate is 0.1. The comparison of predicted and true values using the neural network in this paper is shown in Figure 6, where the predicted value is very close to the true value, which indicates that the parameters are set reasonably.

The change curves of the best individual fitness values with generations for this paper’s algorithm and GA-BP neural network are shown in Fig. 7 and Fig. 8, respectively. The neural network of this paper tends to stabilize in the 10th to 11th generation, and the GA-BP neural network determines whether it tends to stabilize or not after the 20th to 21st generation, which indicates that the improved genetic algorithm of this paper can search for more appropriate weights and thresholds.

The percentage of prediction errors obtained using this paper’s algorithm, GA-BP neural network, and BP neural network are shown in Figure 8. From the figure, it can be visualized that this paper’s algorithm has the smallest error and stable variation, and the BP neural network has the largest prediction error and large variation.

In order to verify the effectiveness of the improved algorithm, the method of this paper, GA-BP neural network, and BP neural network were run 40 times, respectively, and the corresponding mean-square errors were obtained, as shown in Table 1.

Table 1.

The average square error comparison table of the neural network model

Run Frequency	Bp neural network	Ga-bp neural network	This model
1	0.00371	0.00304	0.00183
2	0.00491	0.00279	0.00143
3	0.0044	0.00326	0.0017
4	0.00456	0.00349	0.0017
5	0.00433	0.00334	0.00141
6	0.00372	0.00305	0.00151
7	0.00348	0.00252	0.0019
8	0.00327	0.0031	0.00156
9	0.00349	0.00251	0.00192
10	0.00402	0.00283	0.00115
11	0.00317	0.0033	0.00195
12	0.0047	0.00301	0.00101
13	0.00374	0.00282	0.00195
14	0.00325	0.00212	0.00155
15	0.00469	0.00263	0.00199
16	0.00397	0.00337	0.00169
17	0.00402	0.00282	0.00155
18	0.00331	0.00299	0.00158
19	0.00429	0.00268	0.00184
20	0.00321	0.00218	0.00171
21	0.00374	0.00201	0.00141
22	0.00416	0.00313	0.00187
23	0.0043	0.00261	0.00138
24	0.0043	0.00234	0.00161
25	0.00456	0.0034	0.00131
26	0.00429	0.00234	0.00127
27	0.00359	0.00231	0.00118
28	0.00362	0.00326	0.00122
29	0.00374	0.00294	0.00156
30	0.00485	0.00309	0.00161
31	0.00404	0.00305	0.00179
32	0.00383	0.00302	0.00146
33	0.00442	0.00294	0.00199
34	0.00374	0.00208	0.00184
35	0.00304	0.00337	0.00109
36	0.00438	0.00234	0.00163
37	0.00324	0.00282	0.00195
38	0.00422	0.00329	0.00161
39	0.0036	0.00318	0.00189
40	0.00414	0.00226	0.00188

After running the BP neural network model, the GA-BP neural network model, and the model in this paper independently 40 times, respectively, the mean square error averages were calculated. The mean square error averages of the BP neural network model, GA-BP model, and this paper’s neural network model are 0.003951, 0.00284075, and 0.001612, respectively. The mean square error average of this paper’s neural network model is smaller than that of the GA-BP neural network model by 0.00122875. Among the three kinds of neural networks, this paper’s neural network model prediction is the most accurate.

3.2

Application of the model to real measurements

3.2.1

Measured data

Test data for parameter estimation were obtained from six outstanding male track and field sprinters. The test was conducted every day, and the athlete’s heart rate reflection (heart rate recovery) to the quantitative load at the end of the training session was used as the comprehensive evaluation of the training load, and the test method of physical status was Illinois sensitive running. The test data were 45d of training data during the athlete’s regular training period. The data of the first 40d were used for fitting, and the data of the last 5d were used for prediction. Figure 9 shows the data of a randomly selected athlete.

3.2.2

Model parameter estimation

The inputs to the model were training loadings, and the outputs were quantified fitness states. A nonlinear least squares fit to the test data was performed using the LMA algorithm. There were five parameters in the model, and the parameter estimates are shown in Table 2.

Table 2.

Model parameter estimation

N	Duration of motion	Ka	Kf	Ta	Tf	G	R2	Prediction similarity
1	4	0.098	0.056	0.009	0.174	9.823	0.837	0.927
2	4	0.165	0.037	0.016	0.184	11.055	0.856	0.864
3	4.5	0.161	0.034	0.005	0.169	12.637	10.562	0.935
4	5	0.159	0.049	0.014	0.152	11.762	0.954	0.987
5	5	0.116	0.048	0.018	0.113	12.739	0.852	0.773
6	5	0.174	0.043	0.016	0.152	12.685	0.843	0.992

3.2.3

Model output response

With the assumption of no overtraining, typical idealized data and real model parameters for regular loadings were input to examine whether the response curves of the model were consistent. The results are shown in Figure 10. At lower loads in the first phase, the fitness state went down and then up and then maintained at a certain level. When the load was increased in the second stage, the physical fitness state first declined and then elevated to a higher level of adaptation after some time. When the load was reduced again in the third phase, the physical fitness state elevated, reached the highest level on day 8, and then gradually declined.

3.2.4

Model predictive capability

The first 40 d of the 45 d of training data for a particular subject was used as the fitting data, and the last 5 d as the prediction data. The prediction results are shown in Fig. 11, and the similarity between the actual and predicted values is 99.15%.

The predicted and actual values were analyzed for similarity of distance correlation, and the similarity between predicted and actual values for the six sets of data was 0.916+-0.053, with the minimum and maximum values of 0.837 and 0.996, respectively.

3.3

Training Recovery Strategies

3.3.1

Establishment of a rational training program

A reasonable sports training program has the characteristic of matching, which should be formulated according to the characteristics of the athletes, their physical qualities, training stages, and other factors, and gradually increase its intensity and difficulty. Therefore, the training objectives should be clearly defined first, such as increasing strength, improving endurance, increasing flexibility, etc. Before starting any sports training, proper warm-up activities must be carried out, which include some light stretching, cardiorespiratory training and physical preparation activities to reduce the risk of injury, and then appropriate training time should be scheduled every week according to the objectives.

3.3.2

Strengthening physical fitness training

Injury risk can be comprehensively evaluated through a star balance test, single-leg jump test, hand-held measuring device, etc. Combined with the requirements of different programs on physical quality, athletes’ balance, strength, flexibility, coordination, etc., can be improved through different types of physical training so as to enhance the comprehensive quality of the body. The cardiorespiratory function is improved through aerobic exercise to enhance endurance and health.

3.3.3

Scientific setting of training intensity

In sports training, training intensity refers to the exercise load endured by athletes in a unit of time, which has an important impact on athletic ability and training effect. Research at home and abroad shows that the setting of intensity is related to the athlete’s aerobic capacity, not only lactate tolerance, and the energy supply of the exercise process is mainly realized from the three paths of aerobic, anaerobic lactate and anaerobic lactate-free, and the scientific setting of the training intensity target promotes the enhancement of the effect of the three paths of energy supply. Therefore, according to these three pathways, setting the training intensity for improving the effect of energy supply is conducive to improving training performance and promoting the recovery and improvement of the body.

3.3.4

Use of appropriate sports equipment

The right equipment is essential for sports training. Choosing the right sportswear, sports shoes, protective gear, and other equipment ensures that the body is adequately protected. Choosing the right equipment for one’s body can reduce injuries sustained from sports. For example, wearing helmets and knee pads can effectively prevent head and knee injuries. Choosing the right equipment can improve an athlete’s performance.

3.3.5

Recognize injuries and treat them promptly

In sports training, the occurrence of sports injuries is characterized by regularity and frequency, so it is very important to correctly recognize and deal with sports injuries. First, it is necessary to have a full understanding and risk assessment of the types, causes, prevention, and treatment of sports injuries and to be extra cautious when playing high-risk sports, as correct assessment is conducive to determining the optimal treatment, rehabilitation programs, and return to sports conditions. Secondly, timely and correct treatment and care measures should be taken while consulting with a medical professional and undergoing regular treatment. Timely management of sports injuries can effectively relieve pain and reduce the severity of the injury, speeding up the recovery process.

3.3.6

Standardized techniques for normative training

Incorrect movements go against the structural characteristics of the body and the principles of sports dynamics, which can also lead to an increase in sports injuries. Therefore, athletes should strive to improve their skills and make their movements more standardized. For example, when fighting for the rebound, they should use the power of the waist and abdomen to control their body balance in mid-air, and when landing, the forefoot of both feet will land first and bend the knees to reduce the cushioning force. When the body is out of balance, take appropriate postural maneuvers to protect it.

4

Conclusion

In this paper, the training load of track and field sprinters is evaluated by a genetic algorithm, and data set experiments and practical applications are carried out, on the basis of which a scientific and reasonable training recovery strategy is proposed. It is found in experiments and applications that the improved genetic algorithm can search for more appropriate weights and thresholds. Using this paper’s algorithm, GA-BP neural network, and BP neural network for comparison, this paper’s algorithm has the smallest error and stable change amplitude, and the BP neural network has the largest prediction error and large change amplitude. The neural network of this paper, GA-BP neural network, and BP neural network were run 40 times, respectively, to get the corresponding mean square error. The mean square error of the neural network model of this paper is 0.001612, which is smaller than that of the GA-BP neural network model by 0.00122875. The prediction of this paper’s neural network model is the most accurate among the three kinds of neural networks.

Idioma:: Inglés

Calendario de la edición:: 1 veces al año
Temas de la revista:: Ciencias de la vida, Ciencias de la vida, otros, Matemáticas, Matemáticas aplicadas, Matemáticas generales, Física, Física, otros

RSS Feed de revista

Optimization of training load and recovery strategies for track and field sprinters by genetic algorithm

Qinhai Wang

Publicado en línea: 19 mar 2025

Recibido: 09 nov 2024

Aceptado: 11 feb 2025

DOI: https://doi.org/10.2478/amns-2025-0543

Palabras claveBP neural network, Genetic algorithm, Training load, Sports recovery

© 2025 Qinhai Wang, published by Sciendo

This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Palabras clave
BP neural network, Genetic algorithm, Training load, Sports recovery