Research on sports simulation and fatigue characteristics of athletes based on machine learning

Abstract

The research on the fatigue characteristics of athletes has a certain role in promoting the development of sports. In order to detect fatigue more accurately in the state of human fatigue, this article uses a method of fusing characteristic information of many physiological parameters related to fatigue to design a multi-physical parameter-based exercise fatigue recognition method with high research value and significance. Moreover, this study combines machine learning technology to construct a dynamic fatigue detection system based on BP neural network and multiple physiological parameters. In addition, this study uses samples to construct a BP neural network and achieves dynamic detection of fatigue through multiple physiological parameters. Finally, by constructing controlled trials, fatigue is predicted. The results show that the predicted output of the fatigue value is in good agreement with the expected output, and the research method has certain practical effects.

Keywords

Machine learning athletes fatigue characteristics exercise simulation

1 Introduction

Athlete’s performance and ability will determine the performance and achievement of the athlete. One of the most important theories and realities is that athletes continue to exert influence on physical activity through continuous exercise training, and the increasing load of sports training is transforming the athlete’s body and function. In the process, the athlete’s athletic ability was improved and developed [1]. Athletes often need directional transformation to obtain the ability required for competition, which can only be achieved with the participation and influence of a large number of high-intensity sports training loads. Therefore, the athlete’s training process is to generate a certain amount of exercise fatigue through the use of appropriate methods and exercise training intensity, and then use appropriate methods to eliminate fatigue and achieve a state of physical fitness. This process goes back and forth, and finally the goal of improving the level of competition is achieved. This process involves how to resolve the contradiction between sports training recovery and fatigue adjustment [2]. Obviously, continuously increasing the exercise load during the training process will cause obvious changes in the athlete’s body, and gradually cause the athlete’s body function to adapt to changes, which will promote the athlete’s different competitive abilities. However, if there is no scientific and reasonable control of the training load during the training process, which exceed the athlete’s ability to bear it, it will have a negative impact on the physical function of the athlete. Moreover, some may also ruin the athlete’s sports career and cause irreparable consequences and bring serious consequences to the sports career and individual athletes. Therefore, sports training is implemented in a long-term continuous and phased implementation process. In this process, in addition to periodic training with varying time spans, a well-planned and effective exercise training cycle and adjustment and recovery are required. This article intends to explore a scientific and effective new way of recovering from sports fatigue by investigating the effects of outdoor marathon participation in adjusting the fatigue of sports training after athletes’ medium and high-intensity training.

It is a physiological phenomenon [4] that the body produces fatigue after a large amount of physical exertion, and it is also a protection mechanism, which prompts the body to enter a state of protective rest and energy replenishment to avoid further damage to the body. Under normal circumstances, when the human body is experiencing sports fatigue, general weakness, depressed mood, and lack of concentration will occur. At this time, the human body can usually recover after a period of rest. If a person frequently performs excessive physical activity, the body will be added with new fatigue when it has not recovered from the previous fatigue. At this time, fatigue may accumulate, and the body will be damaged by excessive fatigue over time. If the body is incompletely compensated and the cardiac output of oxygen can only meet the metabolism of the human body at rest, symptoms of hypoxia will occur after exercise. In more serious cases, it may even cause a sudden outbreak of a certain disease in the human body. Of course, things are two-sided, and exercise fatigue in a short period of time will not only cause harm to the human body. On the contrary, the human body must achieve a certain amount of fatigue before the body can obtain excessive recovery. Only when the two situations of exercise fatigue and excessive recovery of the body continue to alternate and reciprocate can the person’s physique gradually increase [5]. Therefore, real-time understanding of the degree of fatigue of the human body and timely measures to eliminate fatigue to restore the function of various organs to the original level can not only prevent the body from being fatigued, but also contribute to the continuous improvement of the body’s athletic ability.

2 Related work

The academic term muscle fatigue may not be common in people’s daily life, but the phenomenon of muscle fatigue is very common. For example, the soreness in the thighs when running on the playground, and the stiffness of the neck when sitting in front of the computer for a long time are common muscle fatigue phenomenon. In terms of human perception, muscle fatigue is a phenomenon of muscle soreness or loss of strength due to prolonged muscle exertion [6]. The physiological definition of muscle fatigue is a physiological phenomenon in which the temporary decline in the function of the neuromuscular system causes the muscle to fail to maintain the expected action [7]. The generation of muscle fatigue is a complex process with many influencing factors. Moreover, muscle cell metabolites, structure, and energy all change with the decrease of intracellular oxygen content and nutrients. When these biochemical environments change, the control strategy of the neuromuscular control system is automatically adjusted [8]. In summary, the causes of muscle fatigue can be simply summarized into two types: peripheral causes and central causes. The former believes that during muscle fatigue, changes in biochemical factors such as the accumulation of metabolites such as lactic acid and the decrease in intracellular energy substances will lead to de-creased muscle contraction capacity. The latter believes that in order to protect the body from injury, the central nervous system automatically adjusts the recruitment strategy of motor units, resulting in decreased muscle motor function [9]. The process of producing muscle fatigue is so complicated that it is difficult to find an effective method to assess the degree of muscle fatigue. Historically, scholars have proposed many muscle fatigue assessment methods in response to this problem. The literature [10] proposed a method for grading exercise intensity and fatigue based on subjective feelings of subjects, that is, the PRE scale. Afterwards, the literature [11] used ultra-sound to evaluate muscle activity and used ultra-sound to detect structural changes in muscles. In addition, because the near-infrared electromagnetic spectrum can measure changes in the amount of oxygenated hemoglobin in the blood, some scholars use near-infrared to observe changes in muscle oxygenation index, and some scholars have con-firmed that changes in muscle oxygenation index have a certain relationship with muscle fatigue [12]. After years of in-depth research, the muscle fatigue analysis technology based on surface electromyography has made great breakthroughs and formed many consistent conclusions [13]. The literature [14] found that the amplitude parameters of surface EMG signals will gradually increase during muscle fatigue. Later, the literature [15] found that the spectrum of surface EMG signals gradually shifted to the left during muscle fatigue. After that, a series of surface EMG parameters were proposed and used for muscle fatigue analysis, such as Mean Power Frequency (MPF), Median Frequency (MDF), The first Coefficient of Auto Regression Model (ARCl), etc. The proposal of these parameters in turn has greatly promoted the application of surface electromyography in related fields.

The field of application of muscle fatigue research is very wide, including biomechanics, kinematics, rehabilitation medicine, etc. In biomechanical research, muscle fatigue is considered to be the main factor affecting the skeletal muscle biomechanical model [16]. Literature [17] explored the mechanical modeling method of the brachioradialis muscle under dynamic periodic force and found that muscle fatigue has a certain effect on the biomechanical model. In kinematics research, muscle fatigue research results are often used for analysis and guidance of athlete training. For example, the literature [18] applied surface electromyography to study the characteristics of upper limb muscle fatigue of table tennis players to analyze their training effects. In the field of rehabilitation medicine, many neuromuscular system related diseases, such as cerebral palsy and stroke, can cause abnormal functional states such as muscle stiffness or rigidity, and this abnormal muscle function is also reflected in muscle fatigue characteristics. The literature [19] found that muscle fatigue characteristics of children with cerebral palsy and young healthy adults are different, and muscle fatigue analysis can be used to assist the diagnosis of cerebral palsy. In addition, muscle fatigue has important research value for clinical rehabilitation training of patients with neuromuscular system diseases. The literature [24] talks about the construction of directed acyclic graph for video coding algorithms for motion estimation in parallel reconfigurable computing systems. The partitioning algorithm also plays a key role in optimizing the encoding of images. The literature [25] dealt with the exploitation of IoT and BigData Analytics using the Hadoop ecosystem in real-time environments. The implementation of IoT-based Smart City is accomplished through the above-mentioned processes. The article [26] centers around IoT and its noteworthy work in sophisticating the human hones and endeavors. This paper moreover overseen the combination of diverse data from distinctive resources that are related with the web. The article [27] talks approximately the different issues within the vehicular communication field with the proposition of agreeable centralized and distributed spectrum detecting model. Due to the execution of the agreeable cognitive model, obstructions and different hidden issues are minimized. The article [28] discusses the problem, such as the tremendous amount of big data, and introduces the SmartBuddy idea of a smart and intelligent world using individual activities and human resources [29, 30].

3 Blood oxygen pulse wave detection module

Blood oxygen saturation (SpO₂) is the percentage of oxygenated hemoglobin (HbO₂) in total hemoglobin in human blood, and it is an important parameter reflecting the oxygen carrying capacity of human blood. The second chapter has discussed the relationship between blood oxygen saturation and fatigue. The following mainly introduces the measurement method and principle of blood oxygen saturation [20].

Nowadays, the most commonly used formula for calculating blood oxygen saturation (SpO₂) is as follows: ${SpO}_{2} = \frac{C_{{HbO}_{2}}}{C_{{HbO}_{2}} + C_{{Hb}_{2}}} \times 100$ (1)

In the formula, C_{HbO
₂} is the oxygenated hemoglobin concentration, and C_{Hb
₂} is the reduced hemoglobin concentration [21].

Based on Lambert-Beer’s law, the degree of attenuation of light after a certain distance in the medium can be used to describe the concentration C of the medium and the absorption coefficient μ_a. $μ_{a} = - \frac{1}{l} In \frac{I}{I_{0}} = ɛ C$ (2)

In the formula, l is the length of the path through which light passes, I and I₀ are the intensity of incident light and reflected light, ɛ is the absorption coefficient, and C is the medium concentration.

Their relationship is as follows: $W = - μ_{a} k$ (3)

In the formula, k is a constant.

When red light (660 nm) and infrared light (940 nm) are used to detect tissues, the absorption coefficients of Hb and HbO₂ are as follows: $μ_{a}^{λ_{1}} = ɛ_{Hb}^{λ_{1}} C_{Hb} + ɛ_{{HbO}_{2}}^{λ_{1}} C_{{HbO}_{2}}$ (4) $μ_{a}^{λ_{2}} = ɛ_{Hb}^{λ_{2}} C_{Hb} + ɛ_{{HbO}_{2}}^{λ_{2}} C_{{HbO}_{2}}$ (5)

By substituting formulas (4) and (5) into formula (1), the following formula can be obtained: ${SpO}_{2} = A + B \frac{W λ_{1}}{W_{λ_{2}}}$ (6)

In the formula, both A and B are constant.

Because blood and other physiological tissues absorb light at different amplitudes, the reflected light contains both DC and AC quantities, which can be expressed as: W = I_AC/I_DC. The formula (6) can be written as [22]: ${SpO}_{2} = A + B \frac{I_{AC}^{λ_{1}} / I_{DC}^{λ_{1}}}{I_{AC}^{λ_{2}} / I_{DC}^{λ_{2}}}$ (7)

In the formula, λ₁ is the wavelength of red light, λ₂ is the wavelength of infrared light, and I_AC and I_DC represent the AC and DC quantities, respectively.

For the convenience of expression, we set $R = \frac{I_{AC}^{λ_{1}} / I_{DC}^{λ_{1}}}{I_{AC}^{λ_{2}} / I_{DC}^{λ_{2}}}$ , then formula (7) can be transformed into: ${SpO}_{2} = A + BR$ (8)

The common oximeters use LEDs and phototransistors to clamp thinner parts of human tissues during measurement, and the blood oxygen value is measured by analyzing the characteristics of transmitted light. This measurement method will compress the measurement point, resulting in poor blood flow. This system uses a reflection type photoelectric sensor NJL5501 R produced by New Japan Wireless to collect the blood oxygen pulse wave of the human body. The sensor encapsulates two LEDs, infrared and red, and a phototransistor. When measuring close to the skin surface, the purpose of measuring reflected light can be achieved. In addition, the size of the sensor package is only 1.9 × 2.6 × 0.8mm, which meets the small volume design requirements of the multi-physiological parameter detection device of this subject [23].

The LED driving circuit of the blood oxygen collection module is mainly composed of NJL5501 and a triode I/O □ (PA6, PA7) is used to control the gating of the red-light emitting tube D1 and the infrared light emitting tube D2. The phototransistor Q3 generates a corresponding voltage signal U₀ according to the intensity of the reflected light from the tissue, and this signal is the starting signal of the blood oxygen pulse wave. Due to the interference of ambient light and the difference in individual light absorption rates, the voltage signal U₀ may be too large or too small, resulting in the phenomenon that the amplitude of the blood oxygen pulse wave signal is not within the amplification range of the op amp chip in the subsequent amplification circuit. In view of this problem, this system adjusts the driving current of the two triodes D1 and D2 by regulating the voltage of the DAC1 terminal in the circuit, and controls the light emitting intensity of the LED, thereby adjusting the voltage U₀. The driving circuit diagram of LED is shown as in Fig. 1.

Fig. 1

LED driving circuit.

Figure 1 LED driving circuit U₀ is usually composed of two parts: AC component and DC component. However, the amplitude of the AC component is much smaller than the amplitude of the DC component, which is a signal of the millivolt level. As a result, the built-in analog-to-digital converter (ADC) of the single-chip microcomputer cannot sample and identify the AC component. Therefore, the AC component needs to be amplified. However, if the pulse wave signal is directly amplified, two situations will occur: the amplification factor is too large, which causes the signal amplitude to exceed the amplifier chip’s amplification range; the amplification factor is too small, which causes the signal to be too small and makes the characteristics of the AC component difficult to be identified. In order to avoid the above two situations and to fully amplify the AC component part, this system uses a secondary amplifier circuit to amplify U₀. This system uses a TLC2262 op amp chip, which has the advantages of low track output performance and low power consumption and is suitable for portable equipment. The secondary amplifier circuit is shown in Fig. 2:

Fig. 2

Secondary amplifier circuit.

In the amplification circuit, the same comparison operation circuit is used for one amplification. The following formula is solved by using the concept of virtual short and virtual break: $V = (1 + \frac{R_{12}}{R_{4}}) U_{0}$ (9) In the formula, U₀ is the output voltage of NJL5501 R, V is the pulse wave voltage signal after one amplification, and $1 + \frac{R_{12}}{R_{4}}$ is the voltage gain of one amplification.

The blood oxygen pulse wave after one amplification contains a large DC signal, which is not conducive to the secondary amplification of the AC component. In order to remove the influence of the DC signal, the secondary amplifier circuit uses a differential input subtraction circuit, as shown in Fig. 2.

Amongthem, R₁₃ = R₁₄, R₁₄ = R₁₆. Combining equation (9), it can be known that the relationship between the voltages U₀V_DACIV_out is: $V_{out} = \frac{R_{16}}{R_{13}} ((1 + \frac{R_{12}}{R_{4}}) U_{0} - V_{DAC 2})$ (10)

In the formula, V_out is the blood oxygen pulse wave signal after the second amplification, and V_DAC2 is the subtraction voltage.

4 Body temperature detection module

This subject uses an electronic thermometer thermistor. Its temperature measurement accuracy is higher than that of a general thermistor, which is suitable for the measurement of human body temperature. The relationship between its resistance and temperature is shown in Fig. 3:

Fig. 3

Corresponding relationship between thermistor resistance and temperature.

The body temperature measurement circuit consists of a simple voltage divider circuit and a subtraction amplifier circuit, as shown in Fig. 4:

Fig. 4

Body temperature measurement circuit.

When the human body temperature changes, the resistance value R_t of the thermistor changes accordingly, so that the voltage U_t also changes accordingly. In order to make the measurement more accurate, the subject magnifies U_t. Thermistor resistance R_t can be calculated by measuring the voltage U_t, and the temperature of the human body can be obtained by referring to Fig. 3. The formula for calculating thermistor resistance R_t and voltage U_t is as follows: $R_{t} = \frac{3.3 R_{18}}{U_{t}} - R_{18}$ (11)

5 BP neural network

In this study, a fatigue prediction algorithm model based on BP neural network is established. Moreover, this study takes a variety of physiological parameters during human movement as input and completes the intelligent assessment of fatigue value during human movement through a BP neural network mathematical model.

BP neural network generally consists of an input layer, one or more hidden layers, and an output layer. The neurons in each layer only accept the input of the neurons in the front layer, and the neurons in the back layer have no signal feedback to the neurons in the front layer. Finally, the output layer signals are fed back to the input layer, and each neuron in the same layer does not affect each other. Figure 5 is a topological diagram of a BP neural network with only one hidden layer, which reflects the mapping relationship between n input quantities and m output quantities.

Fig. 5

Topological structure of BP neural network.

Among them, X₁, X₂, ⋯ , X_n is the input value of the BP neural network, Y₁, Y₂, ⋯ , Y_m is the output value of the neural network, and w_ij and w_jk are the weights of the neural network.

When constructing a BP neural network model, the structure of the network model needs to be determined first, which is considered from the input layer, the hidden layer, and the output layer.

The determination of the number of nodes in the hidden layer often refers to the following formula: $l < \sqrt{m + n} + a$ (12)

In the formula, l is the number of nodes in the hidden layer, m is the number of nodes in the input layer, n is the number of nodes in the output layer, and a is a constant between 0 and 10.

When setting the number of nodes in the hidden layer, the approximate range of l is generally determined first, and then the best value of l is determined by trial and error, which will be discussed in the following.

The training of BP neural network in this subject mainly includes the following seven steps:

Step 1: network initialization. The input amount of this subject is 5 physiological parameters, and the output amount is the fatigue value. Therefore, the number of nodes in the input layer of the network is set to 5, the number of hidden layers is set to 1, and the number of output layer nodes is set to 1. The connection weight w_ij, w_jk, the hidden layer threshold a and the output layer threshold b between the neurons in each layer are initialized, and the neuron excitation function and learning rate are set.

Step 2: Hidden layer output calculation. According to the input multiple physiological parameters, the weight w_ij between the input layer and the hidden layer and the threshold a of the hidden layer, the output H of the hidden layer can be obtained as: $H_{i} = f (\sum_{i = 1}^{n} w_{ij} x_{i} - a_{i}) j = 1, 2, \dots, l$ (13)

In the formula, f is the excitation function of the hidden layer, l is the number of nodes in the hidden layer, and x_i is the input amount.

Step 3: Output layer output calculation. From the hidden layer output H, the weight w_jk and the threshold b, the network prediction output O is: $O_{k} = \sum_{j = 1}^{l} H_{jw} w_{jk} - b_{k} k = 1, 2, \dots, m$ (14)

Step 4: Error calculation. From the predicted fatigue value and expected fatigue value of the network, the prediction error e of the network is obtained as: $e_{k} = Y_{k} - O_{k} k = 1, 2, \dots, m$ (15)

In the formula, Y_k is the expected fatigue value and O_k is the predicted fatigue value.

Step 5: Weight update. According to the prediction error e of the network, the weight w_ijandw_jk are corrected. $w_{ij} = w_{ij} + η H_{j} (1 - H_{j}) x (i) \sum_{k = 1}^{m} w_{jk} e_{k} j = 1, 2, \dots, l$ (16) $w_{jk} = w_{ijkj} + η H_{j} e_{k} j = 1, 2, \dots, l; k = 1, 2, \dots, m$ (17)

In the formula, η is the learning rate.

Step 6: Threshold update. Based on the prediction error e of the network, the thresholds a and b is modified. $a_{j} = a_{j} + η H_{j} (1 - H_{j}) \sum_{k = 1}^{m} w_{jk} e_{k} j = 1, 2, \dots, l$ (18) $b_{k} = b_{k} + e_{k} k = 1, 2, \dots, m$ (19)

Step 7: It is judged whether it reaches the set number of iterations. If the number of iterations is not reached, the algorithm returns to the second step.

The first three steps in the above steps are forward transfer processes, and the fourth to sixth steps are reverse error transfer processes. In the process of continuously adjusting the weights, the actual output of the network is constantly approaching the expected output. The specific training process is shown in Fig. 6:

Fig. 6

BP neural network training process.

At the same time, BP neural network needs to pay attention to the following three points during the training process:

(1) When training BP neural network, in order to avoid the situation that the weights increase or decrease at the same time, the initial value of the weights should be set to a random number. In addition, the initial value of the weight is too large or too small will affect the training speed, and its value is generally selected between -2.4F and 2.4F, where F is the number of input terminals of the connected neurons.

(2) In the process of BP neural network learning, the choice of learning rate is more important. If the learning rate is set too large, the amount of weight changes will be large each time. Although this speed up the convergence of the network, it will cause the network to be unstable. On the contrary, if the learning rate is set too small, it can avoid network instability, but the network convergence speed is relatively slow.

(3) If each layer of the BP neural network is only a linear transformation, the multi-layer input is still a linear transformation after superposition, and the linear model has insufficient expressive power. Therefore, the nonlinear function is introduced by using an activation function. Because the BP algorithm requires the activation function to be differentiable, a Sigmoid-type function is used in this study, as shown in the formula: $f (x) = \frac{1}{1 + e^{- x}}$ (20)

The BP neural network is combined with the human fatigue characteristic detection model in this study.

During the exercise, the mechanics of the human body are proportional to the exercise intensity [52], and the exercise intensity can be reflected in the acceleration of the movement. In this study, the calculation of exercise energy consumption is performed through the weight and the output value of the triaxial acceleration sensor. According to physics, the relationship between the instantaneous speed in one direction and the acceleration in that direction during human movement is: $v (t) = v_{0} + a Δ t$ (21)

In the formula, Δt is a basic time unit, v (t) is the instantaneous speed after time Δt, v₀ is the initial speed, and a is the instantaneous acceleration when the human body moves.

In time Δt, the displacement Δs of the human body in this direction of movement is shown by the formula: $Δ S = v (t) Δ t$ (22)

According to the mechanical work formula W = FS and the constant acceleration of the human body during the movement, the work is differentiated to obtain the work Δw performed by the human body in the basic time unit, as shown in the formula: $W = \sum_{i = 1}^{T / Δ t} mav (t) Δ t = \sum_{i = 1}^{T / Δ t} ma (v_{i} + a Δ t) Δ t$ (23)

In the formula, T is the exercise time, W is the work performed by the human body during exercise time, m is the human body weight, v_i is the human body’s movement speed in the direction of movement after a certain basic time unit, and a is the instantaneous acceleration of the human body when it moves. During the movement of the human body, the direction of movement is constantly changing, so the acceleration value is positive or negative, which causes some mechanical work to be cancelled. Because the work done by the human body during exercise is linear with the energy consumption E and there is no negative energy consumption, the relationship between energy consumption and acceleration is as follows: $E = A \sum_{i = 1}^{T / Δ t} m Δ t | a \times (v_{i} + a Δ t) | + B$ (24)

In the formula, A and B are both constants. The acceleration sensor can measure the acceleration value in three directions: X-axis, Y-axis and Z-axis. In this subject, the X and Z axes of the sensor are parallel to the ground, and the measured acceleration value is the acceleration of the human body in that direction. Therefore, the energy consumption E_x and E_z of human exercise in the X and Z directions are as follows: $E_{x} = A \sum_{i = 1}^{T / Δ t} m Δ t | a_{x} \times (v_{i} + a_{x} Δ t) | + B$ (25) $E_{z} = A \sum_{i = 1}^{T / Δ t} m Δ t | a_{z} \times (v_{i} + a_{z} Δ t) | + B$ (26)

In the formula, a_x and a_z are the acceleration output of the X-axis and Z-axis of the sensor, respectively, and A and B are constant. The Y-axis measurement direction is the same as the direction of gravity. The measured values include the human body’s exercise acceleration and gravity acceleration g in this direction. The influence of gravity acceleration needs to be considered when calculating the energy consumption in this direction. The actual acceleration of the human body in the Y-axis should be a_y + g (g is a scalar). At this time, the energy consumption E_y of the human body movement in the Y-axis direction is as follows: $E_{y} = A \sum_{i = 1}^{T / Δ t} m Δ t | a_{y} + g | | v_{i} + (a_{y} + g) | + B$ (27)

In summary, according to the body weight m and the acceleration output value of the triaxial gravity sensor in each direction, the total energy consumption of the human body during exercise is shown by the formula: $\begin{matrix} EE = E_{x} + E_{y} + E_{z} = \\ A \sum_{i = 1}^{T / Δ t} m Δ t | a_{x} \times (v_{i} + a_{x} Δ t) | + B + \\ A \sum_{i = 1}^{T / Δ t} m Δ t | a_{y} + g | | v_{i} + (a_{y} + g) | + B + \\ A \sum_{i = 1}^{T / Δ t} m Δ t | a_{z} \times (v_{i} + a_{z} Δ t) | + B \end{matrix}$ (28)

6 Experimental verification

The subjects collected 600 sets of data at three different speeds of 6 km / h, 8 km / h, and 10 km / h. Figures 6 8 show the trend of physiological parameters of the subject during running. The subject’s heart rate change trend table is shown in Table 1.

Table 1
Trend table of heart rate of experimental subjects

a b c a b

20 66.91 71.00 65.96 620 158.59 164.23 177.66

40 88.57 82.00 83.31 640 155.38 169.53 179.70

60 90.73 94.16 92.24 660 154.97 167.30 178.42

80 107.19 97.16 92.62 680 158.23 162.36 177.24

100 108.46 97.19 94.50 700 159.32 169.03 177.25

120 133.74 99.73 96.13 720 157.55 162.98 176.77

140 153.59 108.00 96.80 740 159.91 161.35 175.11

160 154.03 120.36 103.56 760 151.77 162.82 179.29

180 151.25 135.90 108.10 780 151.49 166.40 179.01

200 151.52 150.39 134.57 800 152.34 168.08 178.27

220 155.59 151.00 143.73 820 159.70 162.60 176.13

240 152.55 151.60 154.98 840 159.00 161.42 177.26

260 151.04 162.21 176.33 860 155.77 160.06 177.68

280 151.71 165.38 175.42 880 155.08 164.33 178.28

300 150.59 160.57 177.50 900 156.18 162.39 178.66

320 154.77 163.65 176.85 920 153.01 163.28 175.75

340 150.62 163.53 178.76 940 157.96 162.04 176.07

360 157.35 168.73 179.00 960 151.14 161.13 179.21

380 150.25 165.42 178.23 980 154.35 166.92 176.16

400 158.85 164.14 175.77 1000 159.84 169.41 178.12

420 158.03 161.71 178.39 1020 150.51 160.46 179.07

440 156.17 169.37 178.20 1040 152.62 162.45 179.38

460 157.72 161.19 175.52 1060 155.73 168.18 179.37

480 158.21 169.19 177.57 1080 157.98 168.14 176.45

500 152.50 160.67 179.96 1100 150.22 166.62 179.83

520 153.61 165.89 179.68 1120 151.79 167.09 175.07

540 159.88 164.05 179.27 1140 157.81 165.67 176.03

560 156.93 164.49 179.14 1160 159.87 166.67 175.11

580 157.33 167.55 176.39 1180 156.22 163.17 177.40

600 156.39 164.23 177.83 1200 151.10 169.81 175.81

	a	b	c		a	b
20	66.91	71.00	65.96	620	158.59	164.23	177.66
40	88.57	82.00	83.31	640	155.38	169.53	179.70
60	90.73	94.16	92.24	660	154.97	167.30	178.42
80	107.19	97.16	92.62	680	158.23	162.36	177.24
100	108.46	97.19	94.50	700	159.32	169.03	177.25
120	133.74	99.73	96.13	720	157.55	162.98	176.77
140	153.59	108.00	96.80	740	159.91	161.35	175.11
160	154.03	120.36	103.56	760	151.77	162.82	179.29
180	151.25	135.90	108.10	780	151.49	166.40	179.01
200	151.52	150.39	134.57	800	152.34	168.08	178.27
220	155.59	151.00	143.73	820	159.70	162.60	176.13
240	152.55	151.60	154.98	840	159.00	161.42	177.26
260	151.04	162.21	176.33	860	155.77	160.06	177.68
280	151.71	165.38	175.42	880	155.08	164.33	178.28
300	150.59	160.57	177.50	900	156.18	162.39	178.66
320	154.77	163.65	176.85	920	153.01	163.28	175.75
340	150.62	163.53	178.76	940	157.96	162.04	176.07
360	157.35	168.73	179.00	960	151.14	161.13	179.21
380	150.25	165.42	178.23	980	154.35	166.92	176.16
400	158.85	164.14	175.77	1000	159.84	169.41	178.12
420	158.03	161.71	178.39	1020	150.51	160.46	179.07
440	156.17	169.37	178.20	1040	152.62	162.45	179.38
460	157.72	161.19	175.52	1060	155.73	168.18	179.37
480	158.21	169.19	177.57	1080	157.98	168.14	176.45
500	152.50	160.67	179.96	1100	150.22	166.62	179.83
520	153.61	165.89	179.68	1120	151.79	167.09	175.07
540	159.88	164.05	179.27	1140	157.81	165.67	176.03
560	156.93	164.49	179.14	1160	159.87	166.67	175.11
580	157.33	167.55	176.39	1180	156.22	163.17	177.40
600	156.39	164.23	177.83	1200	151.10	169.81	175.81

As can be seen from Fig. 7, the subject’s heart rate increased significantly within the first 2 minutes. The reason is that after the human body starts strenuous exercise, the oxygen consumption of the body’s tissues and organs increases sharply. At this time, the heart must pump more blood to increase the blood supply to ensure the oxygen demand of the tissues and organs of the body. Heart rate stabilizes after 2 minutes of body movement. However, as the body’s energy expenditure continues, the heart rate will slowly rise over time. The greater the intensity of exercise, the higher the average heart rate. The oxygen intake and oxygen consumption of the human body in a quiet state are basically in equilibrium, and the blood oxygen saturation of a normal human body is between 98% and 99%. In the state of exercise, the body’s oxygen consumption greatly increases, causing the oxygen uptake rate to keep up with the oxygen consumption rate. When the body’s stored oxygen is consumed in large quantities, the total oxygen content of the body will begin to decrease. As shown in Fig. 8, after about 5 minutes, the blood oxygen saturation of the subject began to decrease significantly and then gradually flatten again to some extent. When the body’s blood oxygen content decreases, the body’s lack of oxygen supply increases the fatigue response of various organs, in which the brain’s response is particularly obvious, and subjects suffers from decreased attention, confused thinking, and slow response. It can also be found from Fig. 8 that exercise intensity affects the rate of decline in blood oxygen saturation. Table 2 is the statistical table of the changes of the SpO₂ value of the experimental objects.

Fig. 7

Trend diagram of heart rate change of experimental subjects.

Fig. 8

Trend diagram of the change of the SpO₂ value of the experimental object.

Table 2

Statistical table of the changes of the SpO₂ value of the experimental objects

	a	b	c		a	b	c
20	98.04	98.97	98.33	620	96.79	95.89	95.49
40	98.05	98.97	98.09	640	96.64	95.87	94.84
60	98.07	98.61	98.19	660	96.52	95.78	94.69
80	98.10	98.68	98.23	680	96.36	95.77	94.66
100	98.13	98.51	98.11	700	96.28	95.75	94.33
120	98.18	98.99	98.05	720	96.14	95.52	94.31
140	98.20	98.80	98.22	740	96.08	95.38	94.29
160	98.22	98.62	98.14	760	96.08	95.18	94.23
180	98.24	98.94	98.38	780	96.07	95.17	94.19
200	98.26	98.71	98.14	800	96.00	95.12	94.11
220	98.26	98.97	98.23	820	95.62	94.31	93.80
240	98.29	98.53	98.40	840	95.79	94.65	93.89
260	98.31	98.69	98.03	860	95.83	94.76	93.23
280	98.33	98.58	98.22	880	95.57	94.41	93.58
300	98.33	98.70	98.09	900	95.69	94.56	93.80
320	98.40	98.77	98.33	920	95.86	94.09	93.89
340	98.44	98.53	98.38	940	95.57	94.53	93.62
360	98.49	98.83	98.13	960	95.92	94.91	93.72
380	97.83	97.83	97.76	980	95.83	94.57	93.41
400	97.81	97.83	97.10	1000	95.57	94.16	93.52
420	97.72	97.60	96.82	1020	95.81	94.40	93.25
440	97.67	97.58	96.81	1040	95.72	94.92	93.66
460	97.57	97.16	96.70	1060	95.53	94.71	93.44
480	97.52	97.04	96.50	1080	95.54	94.05	93.61
500	97.49	96.83	96.48	1100	95.95	94.53	93.79
520	97.41	96.71	96.38	1120	95.62	95.00	93.46
540	97.15	96.67	96.20	1140	95.83	94.69	93.02
560	97.04	96.55	96.07	1160	95.54	94.87	93.72
580	96.99	96.25	95.64	1180	95.78	94.41	93.62
600	96.92	96.08	95.51	1200	95.82	94.35	93.90

Body temperature is a commonly used physiological measurement parameter, which can reflect the physiological metabolism of the human body. Table 3 is a statistical table of the subject’s body temperature changes, and Fig. 9 is a trend chart of the subject’s body temperature changes. It can be seen from Fig. 9 that the subject’s body temperature maintained an upward trend within 4 to 5 minutes of running, which is the reason for the accelerated physiological metabolism of the human body after exercise and the heat generation of muscle tissue. After 4 to 5 minutes, the rising trend of body temperature slows down, and some even show a decreasing trend. The reason is that the body’s sweat removes some of the heat. All in all, body temperature is on the rise, and the greater the intensity of exercise, the faster the rise.

Table 3

Statistical table of body temperature changes of experimental subjects

	a	b	c		a	b	c
20	34.71	34.70	34.67	620	34.82	34.88	35.03
40	34.71	34.72	34.68	640	34.83	34.89	35.05
60	34.72	34.72	34.70	660	34.83	34.90	35.06
80	34.72	34.72	34.70	680	34.83	34.90	35.07
100	34.73	34.73	34.72	700	34.84	34.90	35.07
120	34.74	34.73	34.72	720	34.84	34.90	35.08
140	34.75	34.74	34.72	740	34.84	34.91	35.10
160	34.76	34.75	34.74	760	34.84	34.91	35.11
180	34.76	34.75	34.74	780	34.85	34.91	35.12
200	34.77	34.75	34.75	800	34.85	34.92	35.12
220	34.77	34.76	34.75	820	34.85	34.93	35.13
240	34.77	34.78	34.75	840	34.85	34.93	35.15
260	34.78	34.78	34.76	860	34.86	34.94	35.15
280	34.78	34.79	34.76	880	34.86	34.94	35.17
300	34.79	34.80	34.78	900	34.86	34.94	35.19
320	34.80	34.80	34.81	920	34.86	34.94	35.19
340	34.80	34.81	34.81	940	34.87	34.95	35.22
360	34.80	34.82	34.87	960	34.87	34.96	35.23
380	34.80	34.82	34.89	980	34.87	34.96	35.23
400	34.80	34.82	34.90	1000	34.87	34.96	35.23
420	34.81	34.83	34.92	1020	34.87	34.97	35.24
440	34.81	34.83	34.94	1040	34.87	34.97	35.24
460	34.81	34.85	34.95	1060	34.88	34.98	35.25
480	34.81	34.85	34.96	1080	34.88	34.98	35.26
500	34.81	34.85	34.96	1100	34.88	34.98	35.26
520	34.81	34.87	34.98	1120	34.89	34.98	35.27
540	34.81	34.87	34.99	1140	34.89	34.99	35.27
560	34.82	34.87	35.00	1160	34.90	34.99	35.28
580	34.82	34.87	35.02	1180	34.90	34.99	35.29
600	34.82	34.88	35.03	1200	34.90	34.99	35.30

Fig. 9

Trend diagram of body temperature changes of experimental subjects.

A set of physiological parameters of the subject during exercise is taken as the input of the BP neural network, and then the fatigue value is predicted in real time. As shown in Fig. 10, the red marks in the figure are the fatigue values predicted by the BP neural network based on a number of physiological parameters when the subject is in exercise, and the blue mark is the fatigue value obtained through the comprehensive analysis of the main observation evaluation method, that is, the expected output value of the neural network. It can be seen that although the predicted output of the fatigue value fluctuates up and down, it is generally distributed around the expected output, and the trends of the two are generally consistent.

Fig. 10

Comparison diagram of predicted output and expected output of fatigue value.

7 Conclusion

With the increasing awareness of fatigue hazards, more and more researches have been done on fatigue, but the research on the dynamic detection of human exercise fatigue is still rare. Based on the research on the existing fatigue detection technology, this study carries out a dynamic fatigue detection study based on BP neural network and multiple physiological parameters. To meet the needs of sample data collection, a dynamic monitoring system for physiological parameters is designed. One part of the system is the detection device, and the other part is the host computer interface monitoring system. This study uses samples to construct a BP neural network and achieves dynamic detection of fatigue through multiple physiological parameters. In addition, this study uses athlete fatigue simulation as an example to perform the study.

In this project, five physiological parameters of the test object are used as the input of the BP neural network to predict fatigue. The results show that the predicted value of the fatigue value is in good agreement with the expected output, and the error ratio between the two is less than±20%. Moreover, the experimental results prove that the BP neural network can predict the fatigue value better through dynamic physiological parameters, so as to reflect the fatigue degree of the human body during exercise.

References

Schütte Kurt

, Maas

E.A.

, Vasileios

, et al., Wireless tri-axial trunk accelerometry detects deviations in dynamic center of mass motion due to running-induced fatigue, PLoS ONE 10(10) (2015), e0141957.

Ding

, Zhao

, Han

, et al., Adaptive hybrid classifier for myoelectric pattern recognition against the interferences of outlier motion, muscle fatigue, and electrode doffing, IEEE Transactions on Neural Systems and Rehabilitation Engineering 27(5) (2019), 1071–1080.

Tian

, Chang

and Li

, Development of standard fatigue vehicle force models based on actual traffic data by weigh-ln-motion system, Journal of Testing and Evaluation 45(3) (2017), 799–807.

Huether

, Abbott

, Cullen

, et al., Energy through motion?: An evidence-based exercise program to reduce cancer-related fatigue and improve quality of life, Clinical Journal of Oncology Nursing 20(3) (2016), 60–70.

Appelbaum

L.G.

and Erickson

, Sports vision training: A review of the state-of-the-art in digital training techniques, International Review of Sport and Exercise Psychology 11(1) (2018), 160–189.

Rein

and Memmert

, Big data and tactical analysis in elite soccer: future challenges and opportunities for sports science, SpringerPlus 5(1) (2016), 1–13.

Ridderinkhof

K.R.

and Brass

, How kinesthetic motor imagery works: a predictive-processing theory of visualization in sports and motor expertise, Journal of Physiology-Paris 109(1-3) (2015), 53–63.

Mulligan

, Lohse

K.R.

and Hodges

N.J.

, An action-incongruent secondary task modulates prediction accuracy in experienced performers: evidence for motor simulation, Psychological Research 80(4) (2016), 496–509.

Khudolii

O.M.

, Ivashchenko

O.V.

, Iermakov

S.S.

, et al., Computer simulation of junior gymnasts’ training process, Science of Gymnastics Journal 8(3) (2016), 215–228.

10.

Seo

S.W.

, Yun

, Kim

M.G.

, et al., Screen-based sports simulation using acoustic source localization, Applied Sciences 9(15) (2019), 29–38.

11.

Tivener

K.A.

and Gloe

D.S.

, The effect of high-fidelity cardiopulmonary resuscitation (CPR) simulation on athletic training student knowledge, confidence, emotions, and experiences, Athletic Training Education Journal 10(2) (2015), 103–112.

12.

Ivashchenko

O.V.

and Kapkan

O.O.

, Simulation of process of 14–15 years old girls’ training of light athletic and gymnastic exercises, Pedagogics, Psychology, Medical-Biological Problems of Physical Training and Sports 22(8) (2015), 32–40.

13.

Owen

P.D.

and King

, Competitive balance measures in sports leagues: The effects of variation in season length, Economic Inquiry 53(1) (2015), 731–744.

14.

Yang

, The simulation of table tennis during the course of sports, Caribbean Journal of Science 52(4) (2019), 1561–1564.

15.

Bulat

, Can

N.K.

, Arslan

Y.Z.

, et al., Musculoskeletal simulation tools for understanding mechanisms of lower-limb sports injuries, Current Sports Medicine Reports 18(6) (2019), 210–216.

16.

Payne

, Mitchell

, Bibb

, et al., The evaluation of new multi-material human soft tissue simulants for sports impact surrogates, Journal of the Mechanical Behavior of Biomedical Materials 41 (2015), 336–356.

17.

, Hu

, Xue

, et al., Preliminary teaching reform of experiment of sports human science based on virtual simulation technology, Journal of Huaibei Normal University (Natural Science Edition) (2) (2016), 17-29.

18.

Lopatiev

, Ivashchenko

, Khudolii

, et al., Systemic approach and mathematical modeling in physical education and sports, Journal of Physical Education and Sport (JPES) 17(1) (2017), 146–155.

19.

Bennour

, Teaching practices and student action in physical education classes: perspectives for teacher education, Creative Education 6(10) (2015), 934–944.

20.

Sghaier

, Elandoulsi

, Mami

, et al., Physical education teacher’s training in swimming under the joint didactic action, Creative Education 6(22) (2015), 2433–2437.

21.

Souza

ítalo Matheus Maligere de

and Silva

C L D.

, Body practices and Brazilian culture: pedagogical contributions to physical education professionals,434-440, Procedia Social & Behavioral Sciences 174(23) (2015).

22.

Cutforth

and Belansky

E.S.

, A community-engaged approach to translating research into practice: A physical education story, Progress in Community Health Partnerships Research Education & Action 9(4) (2015), 571–582.

23.

Hwang,

EuiRyong

, Kim and TaeYoung , Intensification of the education of public health, hygiene, and martial arts during the Japanese colonial period (1937–1945), Journal of Exercise Rehabilitation 14(2) (2018), 160–167.

24.

Paul

, Jiang

Y.C.

, Wang

J.F.

and Yang

J.F.

, Parallel reconfigurable computing-based mapping algorithm for motion estimation in advanced video coding, ACM Transactions on Embedded Computing Systems (TECS) 11(S2) (2012), 1–18.

25.

Rathore

M.M.

, Paul

, Hong

W.H.

, Seo

H.C.

, Awan

and Saeed

, Exploiting IoT and big data analytics: Defining smart digital city using real-time urban data, Sustainable Cities and Society, 40 (2018), 600–610.

26.

Paul

, “Internet of Things: A primer’, R Jeyaraj Human Behavior and Emerging Technologies 1(1) (2019), 37–47.

27.

Paul

, Daniel

, Ahmad

and Rho

, Cooperative cognitive intelligence for internet of vehicles, IEEE Systems Journal 11(3) (2017), 1249–1258.

28.

Markov Anton Iliev

, Rahnev

Asen

and Kyurkchiev

, A note on the three–stage growth model, Dynamic Systems and Applications 28 (2019), 63–72.

29.

Zhang

Yuanfeng

and Han

Hongxia

, Study on the maximum value of flight distance based on the fractional differential equation for calculating the best path of shot put, Dynamic Systems and Applications 29(3) (2020), 707–718.

30.

Paul

Anand

, Ahmad

Awais

, Mazhar Rathore

and Jabbar

Sohail

, Smartbuddy: defining human behaviors using big data analytics in social internet of things, IEEE Wireless Communications 23(5) (2016), 68–74.