Research on the New Mode of Integration and Development of College Students’ Physical Education Teaching and Athletic Training Empowered by Cloud Computing Technology in Colleges and Universities

The cloud computing deployment structure consists of a base platform layer, a data resource layer, a functional application layer and a display performance layer. The basic platform layer is the infrastructure of the cloud computing platform; the data resource layer provides all the data management for the cloud computing platform; the functional application layer has seven subsystems developed according to the needs of sports in universities; the display performance layer is the system interface that interacts with users. The interface is provided by the unified identity authentication system developed by the university itself, and all the subsystems can be logged in and accessed after authentication through the server [18]. The structure of the cloud computing platform is shown in Figure 1.

Figure 1.

Cloud computing platform structure

Figure 2 shows that college sports big data is mainly realized through the framework of data collection, storage, service, and application. In the process of sports activities in colleges and universities, administrators, teachers, students and other groups are always generating massive amounts of data between the three need to establish a closed loop of information flow, much of which is the real and effective information on the performance of the functions of colleges and universities, and it is an important reference for decision makers in the development of policies and systems.

Figure 2.

Large number according to the service architecture

Colleges and universities have three traditional functions, namely, talent cultivation, scientific research and social service. Talent cultivation refers to the cultivation of professional talents through education and teaching; scientific research is to utilize knowledge for academic research and innovation, which is a way to cultivate talents; social service is to use knowledge and research results to serve economic development and social progress, which is an extension of talent cultivation and scientific research. The functions of the university sports teaching and training platform system are realized by the subsystems of office system, sports teaching, extracurricular exercise, physical fitness test, sports competition, stadium and sports research. It is realized by the sub-systems of sports competitions, stadiums and sports research. The functions of the above systems are shown in Figure 3.

Figure 3.

System structure function

In the process of dividing the sample set, groups of centers of mass are close to each other, which indicates that the density of the divided sample set is large, so in the case of neighboring centers of mass are far apart, the minimization of the rate of change of the distance within the group is used as the termination condition for the determination of the number of clusters, and the number of output centers corresponds to the number of clusters [19].

From the principle of the improved algorithm, it can be seen that the selection of the initial center of mass is of great significance to the determination of the number of clusters and the clustering results. For the selection of pairs of centers of mass which are far away from each other, the following method is used: the first center of mass is the closest point of distance s, which is also the point with higher density; the other center of mass is k times of the farthest point of distance s, and k is the quartile of the divided area. The farthest distance from s as a reference point, if the reference point as the center of gravity after the division, the mean value of the retained sample set is greater than 1, indicating that the reference point is an outlier, and at this time the sample corresponding to the k quartile of the retained sample set is the actual farther center of gravity. The farther point selected in this way is not an outlier as far as possible, and the division of the retained area includes the outlier point, so that the density of the area to be divided next time is relatively large, which is conducive to the selection of the initial center of gravity.

The principle of SA is to simulate the physical annealing of solids, where the system is in a random certain state at an initial high temperature T. Due to the motion of the particles, the state of the system changes and it leads to a change in the energy of the system. If the change is in the direction of energy decrease, the change is accepted, otherwise the change is accepted with a certain probability according to the Metropolis criterion. The algorithm is divided into an inner and outer loop, the outer loop is controlled by decreasing the temperature by αT(t − 1), the inner loop is updating the state according to the Metropolis criterion while the particles are equilibrating at temperature T with probability p: (1)p=exp(−ΔE/KBT)

where E denotes the internal energy at the moment of temperature T, ΔE denotes the amount of internal energy change, and K_B denotes the Bolitzmann constant: (2)M={exp(E(Xn)≤E(X0)T),E(Xn)>E(X0)1,E(Xn)≤E(X0)

where X_n denotes the latter moment, X₀ denotes the current moment, and E denotes the internal energy [20].

The steps of the algorithm are as follows:

Step 1: Initialize the parameters, set the initial temperature T₀, the annealing speed, the number of iterations at each temperature, the cease-fire temperature, the random perturbation function on the parameters, randomly generate a set of solutions w as the current optimal solution, and take the distance between the center of mass and the sample as the objective function f(w).

Step 2: Randomly perturb the current w to produce a new solution w′ and a new objective function f(w′) , calculate the increment Δf, the increment is shown in the following equation: (3)Δf=f(w′)−f(w)

Step 3: If the increment is less than 0, accept the new solution and function, otherwise accept the new solution according to Metropolis criterion.

Step 4: Judge whether the iteration number and termination condition are met. If the number of iterations is not met, repeat step 2; if the iteration condition is met and the termination condition is not met, slowly reduce the temperature, reset the number of iterations, and repeat step 2; if the above two conditions are met at the same time, output the center of mass.

When the data have time characteristics, the distribution of the data for each time period is different, only the distance on the similarity can not represent the morphology of the similarity, so for the time characteristics of the data measure similarity, need to be considered in the distance on the basis of the similarity of the morphology of the similarity, at this time, with the Frechette Euclidean distance measure of similarity between the samples. The expression is as follows: (4)dis(A,B)=ED∂(A,B)+FD∂N(A,B) (5)ED∂(A,B)=∑i=0n−1‖Ai−Bi‖2 (6)FDaN(A,B)=infα,β{maxt∈{tk}k=0k=n−1d(A(α(t)),B(β(t)))}

where A and B denote two consecutive curves in space, n denotes the number of features, i denotes the ith feature, ED(A,B) and FD∂N(A,B) denote the Euclidean distance and the discretized Frechet distance, respectively, ‖*‖2 denotes an arbitrary distance measure, α and β denote the two reparameterized functions in the unit interval, and d denotes the distance measure function on space.

To verify the validity of the cluster number determination, it was verified by the elbow method. The optimal degree of distortion is as follows: (7)F(X)=1n∑i=1Kfmin((xi−pi)2)

where x_i denotes the sample set of class i, n denotes the total number of all sample sets, p_i denotes the clustering center of class i samples, and K denotes the number of clusters.

The purpose of clustering is to group users with the same sports training behavior into one category, and the number of users in each category is large, so it is necessary to screen out typical users in each category for sports training behavior analysis. Many scholars have studied it, mainly using the following methods:

Method 1: Take the clustered center of mass as a typical user. Method 2: Take each class of users as a unit, find out the center of gravity of each class of users, and screen out the closest user to the center as a typical user from the users that are larger than the mean. Method 3:

Score the users according to their behaviors, and take the user with the top score as the typical user.

The model for typical user screening is: (8)f(x)=fatt(h(xK,cK))

where x_K denotes the sample set for category K, c_K denotes the clustering center for category K, and att denotes the attributes of the data with the minimum or mean point as the typical user.

The clustered data were analyzed from the questionnaire and the combined data after 496 athletes were assessed to derive relevant clustering information. As the original assessment level of the questionnaire is generally divided into five subdivided levels (fully compliant, relatively compliant, average, relatively non-compliant, and completely non-compliant), the W, to get the raw score of the data, each question score of 5 points, and the cumulative score for the current subdivided dimensions of the total score. As the number of questions corresponding to each subdivided dimension is not the same, so the clustered data should be normalized to prevent the impact of the weight judgment.

According to the improved algorithm, after cleaning and transforming the data, the clustering analysis should be carried out, and the approximate number of clusters should first be calculated using the profile coefficient, K. Figure 5 shows the graph of the value of the profile coefficient (re) for values ranging from 2 to 100. As can be seen from the figure, the contour coefficient value is the largest when the number of clusters is 2, which means that the number of clusters is better when it is 2. However, this is not the final result, and we can first use the hierarchical clustering algorithm to perform clustering, and according to the improved algorithm, the initial clustering center of the next step of the K-Means ++ is calculated, which is convenient for the subsequent calculations.

Figure 5.

Contour coefficient diagram

Further, accordingly, it is determined to use hierarchical clustering algorithm on the experimental data to cluster 2 classes, and then according to the improved algorithm to divide the average clustering in the two classes is also used as the initial clustering center of K-Means++ clustering. After calculation, the initial clustering center can be derived as shown in Table 1.

The improved algorithm was applied to cluster the dataset on 22 dimensions such as competition stressors, social support, and athlete burnout, and the final clustering centers were obtained as shown in Table 2.

Due to the excessive number of dimensions, it is more intuitive to convert them into images, and Figure 6 shows the final clustering center. Observations can be made to know that Cluster-1 has the highest percentage, followed by Cluster-2, and Cluster-0 has the least. The overall similarity between Cluster-1 and Cluster-2 is relatively high, with some differences in some of the attributes, whereas Cluster-0 has a small percentage of the total number of scores, but it is significantly different from Cluster-1 and Cluster-2, implying that this part of the athlete is at a lower level (low overall score, indicating that the stressor does not match, suggests less stress).

Figure 6.

The final clustering is also schematic

Taken together there are the following conclusions: 1)

Cluster-1 athletes, in general, are better in terms of toughness. In the relationship with the coach, the overall relationship with the coach is better, but worse in terms of complementarity, which mainly refers to the athlete’s reaction and coping condition when the coach is instructing, indicating that the Cluster-1 athletes’ coping condition is worse when the coach is instructing them, and needs to be improved. In the stressor scoring items, the overall stress level of Cluster-1 athletes is lower, indicating that the athletes also have a better state quality; in the social support, in general, the emotional support, information support send some main factors have higher scores, indicating that Cluster-1 athletes have a higher degree of well-being, and get the highest degree of social support.

The algorithms involved in the comparison are item-based collaborative filtering algorithm, user-based collaborative filtering algorithm, sorting-based matrix decomposition model, and the algorithm designed in this paper. The experimental results are shown in Fig. 7.

Figure 7.

Recommended accuracy of different recommendation models

From the experimental results, it can be concluded that although the accuracy ranking of the remaining three baseline models fluctuates up and down, the simulated annealing optimized K-Means++ clustering algorithms designed in this paper are all much more accurate than the baseline models, with accuracies of 0.829, 0.907, and 0.916, respectively. And the baseline model accuracies are all below 0.8. The algorithm in this paper obtains the best recommendation performance, which can be verified as superior compared to traditional algorithms.

This set of experiments compares the effect of different number of neighbors on the recommendation results of the final improved model, and the results are shown in Figure 8. From the experimental results, it can be seen that when the number of neighbors is 150, the recommendation results are optimal, and when the number of neighbors increases the recommendation results show a decreasing trend.

Figure 8.

The impact of different recent neighbors on recommended results

The experimental results show that the number of nearest neighbors affects the accuracy of recommendation. This is because when the number of nearest neighbors is too small, the corresponding regularization term has less influence on the target user’s latent factor vector, and thus the accuracy improvement is smaller. When the number of nearest neighbors is too many, users with low similarity will interfere with the target user’s latent factor vectors, which will also affect the accuracy improvement.

Similarity calculation methods based on common friends are often used in cluster analysis, and these two common similarity methods are Jaccard and Cosine similarity. The similarity calculation method designed in this paper is compared with it, and the experimental results are shown in Figure 9.

Figure 9.

The effect of different similarity methods on the recommended results

The experimental results show that the similarity algorithm in this paper is better than the other two similarity algorithms. On MAP@1, MAP@3, and MAP@5, this paper’s model leads at least 0.026, 0.037, and 0.054, respectively.

Although the method based on common buddy seeking similarity is widely used, the user relationship on the cloud computing sports teaching and training platform is unidirectional. If common buddy similarity is used for different structures, some more useful information will be lost. We use different similarity methods for different structures in our model, which improves the accuracy of recommendation.

In order to verify the application effect of the college students’ sports teaching and training platform based on cloud computing designed in this paper, an experiment was carried out in the sports department of a university. The experimental subjects are 100 college students randomly selected from the 2021 class of students in the school, including 25 students in the male experimental group and 25 students in the control group, and 25 students in the female experimental group and 25 students in the control group, and it is ensured that there is no significant difference between the sports knowledge, sports ability, sports behavior and sports psychology of the male and female control group and the experimental group through reasonable division. The study was conducted for a period of 3 months in conjunction with the university physical education program.

The control group carried out the exercise prescription experiment under the coordination of the instructor. The exercise prescription was designed by the instructor according to the physical state of the students and in conjunction with the general physical education curriculum, including the purpose of exercise, health care knowledge, types and loads of exercise, frequency of exercise, and prevention of sports injuries. During the implementation process, the instructor will adjust the exercise prescription appropriately according to the students’ feedback.

The experimental group mainly relies on the designed sports teaching and training platform to carry out exercise prescription experiments. The platform enters the sports quality characteristics of students in the experimental group in advance, and then generates different personalized exercise prescriptions for each student based on the rich case base of exercise prescription and general constraints by using the improved K-Means++ clustering algorithm. During the implementation of the exercise prescription, the system will continuously monitor the changes in the physical fitness, motor skills, and fitness environment resources of the athletes, and precisely adjust the parameters of the exercise prescription according to the cycle data, and finally, the system will also judge and analyze the actual effect of the exercise prescription for each student.

The results of the experiment are quantitatively analyzed in terms of physical literacy and physical fitness of college students, in which the commonly used “Self-measurement Scale of College Students’ Physical Literacy” is used for the assessment of college students’ physical literacy. The scale consists of 4 primary indicators and 17 secondary indicators, including sports knowledge, sports ability, sports behavior and sports psychology, and each secondary indicator is assigned a corresponding weight. Students complete the 30 questions in the self-measurement scale and get the scores of the first level indicators after formula calculation, and the scores of the first level indicators are added together to get the total score of students’ physical fitness. The physical fitness assessment of college students adopts the evaluation standards of college students’ physical fitness monitoring, and the common assessment indicators for both male and female students are lung capacity, 50-meter run, standing long jump, and seated forward bending, in addition to 800-meter run and one-minute sit-up for female students, and 1,000-meter run and pull-ups for male students, with each indicator scored in accordance with the scoring standards of the college students’ physical fitness monitoring.

Table 3 shows the comparison of sports literacy between the control and experimental groups after the experiment.

There were significant differences (p < 0.05) in sports knowledge, athletic ability, sports behavior and sports psychology between male and female control and experimental groups after the experiment. Among female students, the experimental group scored 0.057, 0.081, 0.098, and 0.081 points higher than the control group on the four items, and among male students, 0.034, 0.098, 0.086, and 0.103 points higher, respectively.

Although artificial exercise prescription can increase college students’ sports knowledge and improve their exercise ability and sports behavior, there is still a gap in the improvement of sports literacy compared with intelligent exercise prescription. The design and implementation of artificial exercise prescription is constrained by the instructor’s level of experience in the field, and it is difficult to fully consider the effective cooperation and co-promotion of various factors of physical literacy.

Table 4 shows the comparison of physical fitness between control and experimental groups after the experiment. There were significant differences (p < 0.05) between the male and female control and experimental groups after the experiment in lung capacity, 50-meter run, standing long jump, seated forward bend, 800-meter/1000-meter run, and 1-minute sit-up/pull-up. The experimental group performed better than the control group in all categories except for the 50-meter run and the 800-meter/1000-meter run previously.

Although the teacher’s manually designed exercise prescription can improve the ability of college students in standing long jump, 800m/1000m run, and 1-minute sit-up/pull-up, there is still a certain gap in physical health improvement compared with the exercise prescription given in the platform of this paper. Physical fitness level depends on the comprehensive performance of various physical qualities, which needs to be promoted through systematic, continuous and targeted physical fitness activities. However, the design and implementation of artificial exercise prescription carries the more obvious tendency of instructors, the teaching content is more likely to dominate the fitness content of exercise prescription, and the teaching method is more likely to dominate the implementation method of exercise prescription, which is manifested in the faster improvement of physical fitness related to teaching activities and the slow improvement of other physical fitness.

In this paper, the sports teaching and training platform has developed a proprietary fitness program for each student, which takes into full consideration the students’ physical condition, fitness interest, course content, work and rest patterns, sports injury protection, and strictly controls the types of fitness programs, frequency of exercise, and exercise load, while the system will evaluate the effect of exercise at any time, helping students to more accurately understand and master their own fitness effect, and adjusting the exercise prescription program on a regular basis. At the same time, the system will evaluate the effect of exercise at any time to help students more accurately understand and grasp their own fitness effect, and regularly adjust the exercise prescription program to ensure that the exercise prescription is always compatible with the actual fitness of students.