Neural network-based optimization model for sprint data collection

Abstract

Sprint data has the characteristics of quality and continuity, but due to the limitations of optimization algorithm, the existing sprint data acquisition optimization model has the problem of low optimization performance parameters. Therefore, a data acquisition control optimization model based on neural network is proposed. This paper analyzes the advantages and disadvantages of neural network algorithm, combined with the sprint data collection optimization requirements, introduces BP neural network algorithm, based on this, uses multiple sensors, based on baud interval balance control to collect sprint data, applies BP neural network algorithm to compress, integrate and classify sprint data, realizes the sprint data collection and optimization. The experimental results show that the optimization performance parameters of the model are large, which fully shows that the model has good data acquisition optimization performance.

Keywords

Neural network sprint data dash data acquisition optimization

1. Introduction

Neural network is to use the cognitive basis of medical neurosurgery on human nervous system to explore the use of computer system to simulate the human nervous system, so that the computer system can also be like the human nervous system, which can remember, process and sublimate the external information. If the computer system can process all kinds of external information as intelligently as the human brain, then the computer system will become an information processing system with the style of human brain. Such an intelligent system is composed of a large number of simple artificial processing units. It is a nonlinear, self-organizing and adaptive system. With the development of the times, China’s competitive sports have also made rapid progress, especially in the Beijing Olympic Games is the first gold medal gratifying results, many people think that China has become a sports power from a sports power. But in sharp contrast, as the most important track and field event in the Olympic Games, it is much inferior [1]. In 2011 Daegu world track and Field Championships, China’s track and field events won a gold medal, two silver medals and a bronze medal, ranking seventh. In the 2012 London Olympic Games, China’s track and field events only won one gold medal and five bronze medals, ranking 13th. So far, track and field is one of the oldest events. It is the most wonderful event in the Olympic Games and Asian Games, and the number of medals allocated to it is also the largest. Therefore, we often say that “those who win track and field win the world”. Generally, the level of track and field has a direct impact on the overall level of competitive sports in any country. Track and field is put in the first place of gold medal strategy by many developed countries, such as the United States. China has also launched the Olympic glory program to promote the development of track and field [2].

Relevant scholars have done a lot of research on sprint. Reference [3] puts forward the impact of high-intensity interval training and sprint interval training on time trial results. The scores obtained by the international amateur track and field federation according to the individual best results of each group of sprinters are used as the parameters of sprint results. Reference [4] compared the physiological and clinical indexes of chronic sprint interval training under fasting or feeding. Sprint interval training and intermittent fasting are effective and independent methods to achieve clinical health results. The purpose of this study was to compare the effects of 6-week sedentary sitting under fasting and feeding on physiological and clinical health indexes of healthy adults. Reference [5] proposed a preliminary study on the potential relationship between leg bone length and sprint performance of sprinters. This study investigated the relationship between leg bone length and sprint performance. The leg bone length of 28100 m professional sprinters and 28400 m professional sprinters were measured by magnetic resonance imaging. The length of the femur and tibia was evaluated by calculating the length of the femur and tibia respectively. In order to minimize the difference in body size between participants, both bone lengths were normalized to height. The ratio of tibial length to femoral length was calculated to evaluate the interaction between upper and lower limb bone length.

Sprint is one of the most popular events in track and field, which integrates explosive power, speed, strength and other qualities, and plays a certain role in promoting the development of other events. Because of its strong ornamentals, it has always been regarded as an important part of the world track and field events. At the London Olympic Games, China’s Su Bingtian won the eighth place in the group and the title of “30-meter champion” in the men’s 100-meter sprint semi-final with a score of 10.28 seconds. He is the first person to enter the 100-meter semi-final in the history of China’s Olympic Games. Although failed to enter the final, in the London Olympic Games still gave us a big surprise, his start let us see the hope of Chinese spirit. In 2011, hurdlers such as Liu Xiang and Shi Dongpeng repeatedly performed well in the international sports arena, and Chinese athlete Zhang Peimeng won the 100-meter championship in the 18th Asian Track and Field Championships with an excellent result of 10.28 seconds, which gave us confidence in China’s sprint to a large extent, indicating that we have enough ability to improve the performance of China’s sprinters. However, generally speaking, the overall level of Chinese track and field has been in a relatively backward position in the world, and sprint is a major weakness of competitive sports. In the 18th Asian Track and Field Championships, the short distance race won only three gold medals, while the Japanese team won seven gold medals in the short distance race. In the London Olympic Games, China has won six medals in track and field events. One gold medal is men’s 20 km walking race, and five bronze medals are women’s 20 km walking race, men’s 50 km walking race, women’s discus and women’s shot put. It’s a pity that none of the medals belong to sprint events.

At present, there are only a few Excellent Sprinters in China. Su Bingtian broke the national record of men’s 100-meter with 10.16 seconds, which is still a certain gap compared with the 9.63 seconds of the world’s excellent sprinters. Xie Zhenye, a 200 m sprinter, tied the national record of men’s 200 m with 20.54 seconds, while the best result of London Olympic Games was 19.32 seconds, and the best result of women’s 200 m sprinter Wei Yongli was 24.36 seconds, which was 2.48 seconds different from the best result of world elite sprinters. At present, the 400 m event is still a vacancy in China, China did not arrange athletes to participate in this event in the Olympic Games. Yuan Guoqiang, the head coach of the national track and field men’s sprint, put the focus of the Olympic Games on the relay project. The improvement of the performance of the 4 $\times$ 100 meter relay requires the growth of each sprinter’s personal quality, and formulated corresponding training plans according to each athlete’s different conditions. So how to improve the performance of Chinese sprinters, how can we shorten the gap between China’s sprint and the world level, how to win more medals in sprint events in the World Series, and these are an urgent problem to be solved in China [6].

In order to improve the level and performance of sprinters, it is necessary to collect sprint data. Due to the quality and continuity of sprint data, it is necessary to optimize sprint data [7]. Due to the limitations of optimization algorithm, the existing sprint data acquisition optimization model has the defect of low optimization performance parameters. Therefore, an optimization model of sprint data acquisition based on neural network is proposed. Neural network is a network composed of a large number of artificial neurons. The innovation of the research content is artificial neuron is a mathematical model that approximately simulates biological neurons. Each neuron receives all neuron information connected to it, and after amplification and weighting, the weighted sum is compared with neurons one by one. If it is greater than the threshold, the artificial neuron is activated, the signal is transmitted to the higher-level neurons connected to it. Building the model refers to the sprint data acquisition optimization model based on neural network, which greatly improves the data acquisition and optimization performance parameters of the model, and provides more effective help and theoretical support for improving the data acquisition, application, level and performance in the sprint stage. Research shows that sprinting is an important skill of sprinters. In the fierce sprint competition, the technology at the end of sprint is also a key technical link that can not be ignored. In the process of training and competition, athletes and coaches should consciously strengthen the cultivation of this technical link, improve students’ ability to run the end through various auxiliary teaching means, improve athletes’ awareness of approaching and reaching the end, and achieve excellent results in the fierce competition.

2. Construction of sprint data collection optimization model

2.1 Introduction of neural network algorithm

Neurons are such a large number of artificial processing units [8, 9]. The network formed by extensive interconnection of a large number of neurons is called artificial neural network. It is an abstract network formed by abstraction, simplification and simulation of human brain, which reflects the basic characteristics and attributes of human brain [10]. It is similar to human brain in the following two aspects:

(1)
It has the ability of self-regulation, can carry out learning and training, and can acquire knowledge through training;
(2)
It can store the learned knowledge information by internal neurons (synaptic weight).

Neuron is the basic component of artificial neural network, and a large number of neurons are connected to form the adaptive nonlinear dynamic system of artificial neural network [11]. Neurons are similar to the synapses in the human nervous system and are the basic units of the nervous system [12]. Neuron is also a nonlinear dynamic system, which has the structural characteristics of multi input and single output. Its structural model is shown in Fig. 1.

Figure 1.
Neuron model.

As shown in Fig. 1, the input signal of the neuron is $x_{j}\left({j=1,2,\ldots,n}\right)$ , and the connection weight of the neuron is $\omega_{ij}$ . $\mu_{i}$ is the strength of the end of a neuron, which is the output of the linear combination of the connection weights of the previous input signals, and also becomes the net input of neuron $i$ . After the neuron gets the input signal, it is excited and ignited by a function. This function is usually a nonlinear function with values between 0 and 1. The commonly used excitation functions include ladder function, piecewise linear function and Sigmoid function [13]. $\theta_{i}$ is the additional bias term of neurons. In the absence of any input signal, the initial state of the neuron is determined by $\theta_{i}$ . $\theta_{i}$ also becomes the natural excitation of neurons, which is represented by $b_{i}$ . $V_{i}$ is the value adjusted by $\theta_{i}$ , and $y_{i}$ is the final output of neurons [14].

There is a mathematical relationship between the value $V_{i}$ adjusted by $\theta_{i}$ , the intensity $\mu_{i}$ of the end point of the neuron, the deviation $b_{i}$ of the neuron, the input signal $x_{i}$ of the neuron, the output signal $y_{i}$ of the neuron and the connection weight $\omega_{ij}$ of the neuron:

$\displaystyle\left\{{{\begin{array}[]{l}{V_{i}=\mu_{i}+b_{i}}\\ {\mu_{i}=\sum\limits_{j}{\omega_{ij}x_{j}}}\\ {y_{i}=f\left({\sum\limits_{j}{\omega_{ij}x_{j}}+b_{i}}\right)}\\ \end{array}}}\right.$ (1)

In Eq. (1), $f\left(\cdot\right)$ represents the excitation function, the mathematical relationship between the output signal of the end point of the neuron and the value adjusted by the connection weight of the neuron is used as the premise parameter.

Neural network is formed by a large number of neurons connected with each other. Although the structure of neurons is more complex, its function is very simple. The reason why the artificial neural network has powerful information processing ability is that it interconnects numerous neurons with complex structure and simple function according to a certain arrangement. Through nearly 20 years of efforts and research, researchers have invented hundreds of neural network model structures. These neural network structural models are constructed by scientists aiming at different focuses and different angles [15]. However, according to the traditional classical classification pattern of neural networks, neural networks are divided into two categories: feedback neural networks and feedforward neural networks without feedback. The schematic diagram is shown in Fig. 2.

Figure 2.
Schematic diagram of neural network.

It is found that BP neural network is more suitable for sprint data collection. Feedforward neural network can approach any continuous function, that is to say, using multilayer feedforward network to establish the prediction model of athletes’ special performance can fit any functional relationship between special performance and quality training indicators, reflecting their internal essence. Therefore, the BP neural network algorithm is introduced into the model construction to prepare for the subsequent data acquisition optimization.
2.2 Sprint data collection

In the multi-sensor fusion model, the sprint data is collected based on baud interval balance control. The baud interval equalization control technology is applied to accurately deploy sensor nodes, and the transmission power of sensor output nodes is calculated:

$\displaystyle p_{i}=\frac{\sigma^{2}}{1-\sum\limits_{j=k+1}^{n}{\frac{\gamma_{% i}}{\gamma_{j}+G}}}\times\frac{\gamma_{i}}{h_{i}\left({\gamma_{i}+G}\right)}$ (2)

In Eq. (2), $\gamma_{i}=\gamma_{th}$ . Under the constraint of minimum output power, the optimal localization of sensor nodes is carried out, and the control model of node localization is equivalent to the following form:

$\displaystyle\mathop{\max}\limits_{p_{i}\in P_{i}}u_{i}\left({p_{i},p_{-i}}% \right)=\ln\left({1+\gamma_{i}}\right)-\beta_{i}\gamma_{i}$ (3)

Based on the design of wireless sensor network node aggregation link gain control model in sprinting process of athletes, the data acquisition channel model of wireless sensor network under clustering routing detection protocol is established. Through baud interval balanced configuration method, the power consumption output percentage of wireless sensor network transmitting node receiving data is obtained:

$\displaystyle\eta_{\textit{comm}}=\frac{k_{1}\cdot l}{E_{\textit{comm}}}\cdot% \left({1-p_{\textit{drop}}}\right)$ (4)

In Eq. (4), $p_{\textit{drop}}$ represents the start-up energy consumption of data collection, and establishes the adaptive deployment of WSN nodes under the clustering routing detection protocol. For the transmission power $\textit{CH}_{i}$ of the cluster head node, the optimal deployment spacing of WSN nodes for data collection is as follows:

$\displaystyle\frac{\partial u_{i}}{\partial p_{i}}=\frac{\textit{Gh}_{i}}{\sum% \limits_{j\neq i}{h_{j}p_{j}+\sigma^{2}}}\left({\frac{1}{1+\gamma_{i}}-\beta_{% c_{1}}}\right)$ (5)

The calculation formula of the optimal deployment spacing of WSN nodes is used as the result parameter. Through the above process, the sprint data collection of athletes is completed, which makes preparation for the subsequent sprint data optimization.

2.3 Sprint data optimization

2.3.1 Sprint data compression

Sprint data transmission adopts GPRS (General Packet Radio Service) mode, the data stream is ASCII (Agile supply chain II) character stream, its data compression belongs to the lossless compression of text files, which requires that the original state data value can be accurately restored when decompressing at the receiving end.

At present, there are three common lossless compression techniques: Huffman coding, arithmetic coding and LZW. Among them, Huffman coding is based on the probability of character appearance to construct the shortest average length of different prefix, sometimes called the best coding. Dynamic Huffman coding adopts the method of scanning data while coding (that is, building a tree, it is a multivariable decision tree), and the coding is completed after scanning. In the process of scanning, data can be transmitted at the same time. This solves the problem of real-time transmission. At the same time, the receiver receives data while building a tree, which is exactly the same as the sender’s way of building a tree. During this period, the decoding is also in progress, and the decoding is completed after receiving [16]. In decoding, Huffman tree is not needed, which saves storage space and reduces channel occupation and transmission time. It is shown that mastering Huffman coding is the most important and basic coding method, which has far-reaching significance for future study in this field.

Arithmetic coding maps the input stream to a certain amplitude value according to the probability distribution of the source, which makes it possible to represent an input with a codeword length less than 1 bit. Theoretically, it can achieve the compression efficiency of the first-order entropy of the source [17, 18]. Arithmetic coding is not only a lossless data compression method, but also an entropy coding method. Different from other entropy coding methods, other entropy coding methods usually divide the input message into symbols, and then encode each symbol, while arithmetic coding directly encodes the whole input message into a number, a decimal $n$ . For example, if the output of an arithmetic code for a piece of information is 1010001111, then it means decimal 0.1010001111, that is, decimal 0.64. The decimal system is temporarily used to represent the decimal in the algorithm, which will not affect the feasibility of the algorithm at all. Considering that there are only abc characters in a message, the information to be compressed is bccd. The adaptive model is adopted. At the beginning, it is temporarily considered that the occurrence probability of the three is equal that is, 1/3. The 0–1 interval is assigned to three characters according to the proportion of probability, that is, $a$ from 0.0000 to 0.3333, $b$ from 0.3333 to 0.6667, $c$ from 0.6667 to 1.0000.

LZW algorithm, also known as dictionary coding or general coding, is a lossless and efficient sequence coding method independent of probability and statistics. Through the analysis of the input stream, a string table is generated adaptively, which contains the non repeating substrings in the input stream. Each string is mapped to an independent codeword output, which makes full use of the correlation between adjacent inputs, and achieves the coding efficiency higher than the first order of the source. The compression effect is limited by the capacity of dynamic dictionary, which needs a large data buffer. On the other hand, when the dictionary capacity is large, the computational complexity is also very high. If the data packets of continuous monitoring are connected in series, they can be regarded as continuous code stream, and LZW method can be used to achieve certain compression effect, but the requirements for on-board hardware are very high and the software programming is very complex [19].

Through the analysis of the existing data, it can be seen that the arithmetic coding method is more suitable for the compression of sprint data collection of athletes, so this study uses the arithmetic coding method to compress sprint data [20].

2.3.2 Sprint data classification

In order to integrate sprint data, BP neural network algorithm is used to classify sprint data. The specific steps of BP neural network algorithm are as follows:

Step 1:
BP network initialization: take $n$ random numbers in $\left({-1,1}\right)$ and copy these random numbers to the connection weights between the layers, set the error function $e$ , give the calculation accuracy $\varepsilon$ and the maximum learning times $M$ ;
Step 2:
Randomly select an input sample and its corresponding expected output from the sample data:

$\displaystyle\left\{{{\begin{array}[]{l}{x\left(k\right)=\left({x_{1}\left(k% \right),x_{2}\left(k\right),\cdots,x_{n}\left(k\right)}\right)}\\ {d_{o}\left(k\right)=\left({d_{1}\left(k\right),d_{2}\left(k\right),\cdots,d_{% n}\left(k\right)}\right)}\\ \end{array}}}\right.$ (6)
Step 3:
Calculate the input and output of the hidden layer according to the mathematical formula;
Step 4:
Calculate the partial derivative $\delta_{o}\left(k\right)$ , first, the error between the actual output and the expected output is calculated, and then the partial derivative $\delta_{o}\left(k\right)$ of the error function to each node of the output layer is calculated;
Step 5:
Calculate the partial derivative $\delta_{h}\left(k\right)$ , using the connection weight between hidden layer and output layer, $\delta_{o}\left(k\right)$ of output layer and hidden layer calculate the partial derivative of error function to each node of hidden layer $\delta_{h}\left(k\right)$ ;
Step 6:
Modify the connection weight $w_{ho}\left(k\right)$ , the $\delta_{o}\left(k\right)$ of each node in output layer and the output of each node in hidden layer are used to modify the connection weight $w_{ho}\left(k\right)$ ;
Step 7:
Modify the connection weight $w_{ih}\left(k\right)$ , the $\delta_{h}\left(k\right)$ of each node in the hidden layer and the input of each node in the input layer are used to modify the connection weight $w_{ih}\left(k\right)$ ;
Step 8:
Calculate the global error and judge whether to stop learning on this basis. When the error of BP network reaches the set error or the maximum learning times, the algorithm ends. Otherwise, go to step 3 to continue learning. The specific process is shown in Fig. 3.

Figure 3.
BP neural network flow chart.

Through the above process, the sprint data collection and optimization are realized, which provides better help for the improvement of Sprinters’ level and performance, and also provides support for the research on data collection optimization [21].
3. Simulation experiment

In order to verify the performance difference between the constructed model and the existing model, the MATLAB software is used to design the simulation experiment. There is a clear functional mapping relationship between athletes’ special performance and physical fitness training level. By establishing the optimization model of athletes’ data acquisition and control, we can realize the accurate prediction of athletes’ special performance and provide a scientific theoretical basis for the preparation training of physical fitness. The experimental process is as follows:

3.1 Experimental preparation

In order to ensure the smooth progress of the experiment, the sensor position is arranged reasonably, as shown in Fig. 4.

Figure 4.

Sensor location layout.

3.2 Optimization of performance parameters

In order to objectively display the optimal performance of sprint data collection, the calculation formula of optimal performance parameters is determined as follows:

$\displaystyle\psi=\frac{Q_{*}+Y_{*}}{Q_{0}}\times 100\%$ (7)

In Eq. (7), $\psi$ represents the optimization performance parameters; $Q_{*}$ represents the difference of data compression quantity; $Y_{*}$ represents the accuracy of data classification; $Q_{0}$ represents the total amount of sprint data collected.

It can be seen from Eq. (7) that the larger the optimization performance parameter $\psi$ is, the better the data acquisition optimization performance of the model is; Otherwise, the worse the data acquisition optimization performance of the model is.

3.3 Experimental results

According to the above experimental preparation, the sprint data acquisition optimization experiment is carried out, and the optimized performance parameters are shown in Table 1. The existing model refers to an optimization model based on BP neural network [22].

Table 1
Data table of optimized performance parameters

Number of experiments	Optimize performance parameters
	Existing model	Building model
1	2.45	5.64
2	3.01	6.25
3	3.12	6.00
4	2.59	6.28
5	3.10	6.95
6	3.11	7.14
7	2.99	7.52
8	2.87	6.94

As shown in Table 1, the optimal performance parameters of the existing model range from 2.45 to 3.12, and the optimal performance parameters of the constructed model range from 5.64 to 7.52. Through the comparative study, it is found that compared with the existing models, the optimization performance parameters of the built model are larger, which fully indicates that the built model has better data acquisition optimization performance. The reason is that this method uses BP neural network algorithm to classify the sprint data in the integration of sprint data, which is conducive to the performance optimization of the model to a certain extent.

Suggestions are put forward for sprinters to finish the sprint: speed endurance training should be strengthened to maintain the high speed of the finish run, interval method should be used for distance running training of main event and close to main event, and at the same time, attention should be paid to experience the technical movements of the finish run. The coach should pay close attention to the reality and training content of the sprinters, actively mobilize all favorable factors, let the sprinters know how to cherish the sprint, grasp the sprint, show the willpower to overcome difficulties, and improve the sprinters’ ability of finishing sprint. Finally, whether it is teaching or training, the coach must abide by the physiological characteristics of the human body and the principle of gradual progress. Only by strictly abiding by the physiological principles of the human body and not violating the objective laws, can sprint teaching and training go to a higher level and better tap the potential of athletes.

4. Conclusion

In this study, neural network is introduced to establish a new data acquisition optimization model in the sprint stage, which greatly improves the data acquisition optimization performance parameters of the model, and provides more effective help and theoretical support for the improvement of data acquisition, application, level and performance in the sprint stage. Research shows that sprinting is an important skill of sprinters. In the fierce sprint competition, the technology at the end of sprint is also a key technical link that can not be ignored. In the process of training and competition, athletes and coaches should consciously strengthen the cultivation of this technical link, improve students’ ability to run the end point through a variety of auxiliary teaching means, enhance athletes’ awareness of approaching the end point and hitting the end point, and achieve excellent results in the fierce competition. The best performance parameters of the constructed model range from 5.64 to 7.52.

It is meaningful to study the problem of human movement ability by establishing neural network model, which is worthy of further discussion and research in the future. In the future research, we can also have an in-depth discussion on how to reduce energy consumption during exercise.

Footnotes

Acknowledgments

This work was supported by the Special Scientific Research Plan Project of Shaanxi Provincial Department of Education (No. 21JK0014), the Regular Project of Shaanxi Provincial Sports Bureau (No. 2020245), and the Scientific Research Project of Baoji University of Arts and Sciences (No. ZK2018020).

References

Ferreiro-Cabello

Fraile-Garcia

Martinez

DPAE

Martinez-de-Pisonm

. Metamodel-based design optimization of structural one-way slabs based on deep learning neural networks to reduce environmental impact. Engineering Structures. 2018; 155(15): 91-101.

O’Conor

Wang

Gibney

Young

Jones

Alexandrov

Johnson

Cushman

Tsao

. Lowering systolic blood pressure does not increase stroke risk: An analysis of the SPRINT and ACCORD trial data. Annals of Clinical and Translational Neurology. 2019; 6(1): 1-10.

Conceio

Neto

Carvalho

. Comment on: “Effect of high-intensity interval training versus sprint interval training on time-trial performance: A systematic review and meta-analysis”. Sports Medicine. 2020; 50(12): 2263-2264.

Tan

Lim

Tan

Aziz

. Comparison of physiological and clinical markers for chronic sprint-interval training exercise performed either in the fasted or fed states among healthy adults. Current Research in Physiology. 2021; 4: 192-201.

Daichi

Tadashi

Masafumi

Takahiro

Yuto

Hiromasa

Mitsuo

Akinori

Tadao

. A pilot study on a potential relationship between leg bone length and sprint performance in sprinters; are there any event-related differences in 100-m and 400-m sprints? BMC Research Notes. 2021; 13(1): 297-303.

Fei

Tian

. Research on data mining algorithm based on neural network and particle swarm optimization. Journal of Intelligent and Fuzzy Systems. 2018; 35(4): 1-6.

Giorgio

Melibeo

. Data acquisition and stream processing with Matlab using system objects. International Journal of Computer Applications in Technology. 2020; 62(3): 246-256.

Smeal

Ermentrout

White

. Phase-response curves and synchronized neural networks. Philosophical Transactions of the Royal Society of Series B. 2018; 365(1551): 2407-2422.

Andre

Brett

Novoa

Justin

Swetter

Blau

Sebastian

. Dermatologist-level classification of skin cancer with deep neural networks. Nature. 2017; 542(7639): 115-118.

10.

Hoppe

Baumgart

Polglaze

Freiwald

. Validity and reliability of GPS and LPS for measuring distances covered and sprint mechanical properties in team sports. PLoS ONE. 2018; 13(2): 19-27.

11.

Hervé

Jean-Claude

Béatrice

Frédéric

Francis

. Semantic and episodic memory of music are subserved by distinct neural networks. NeuroImage. 2019; 20(1): 244-256.

12.

Desai

Conway

Crook

Overstreet

. Credit-scoring models in the credit-union environment using neural networks and genetic algorithms. IMA Journal of Management Mathematics. 1997; 8(4): 323-346.

13.

Sexton

Canney

Kenny

. Limitations of observational data in interpreting SPRINT result-reply. JAMA Internal Medicine. 2018; 178(1): 1-9.

14.

Müller

Hiss

Schneider

. Recurrent neural network model for constructive peptide design. Journal of Chemical Information and Modeling. 2018; 58(2): 472-479.

15.

Prasad

Yadav

. Use of response surface methodology and artificial neural network approach for methylene blue removal by adsorption onto water hyacinth. Water Conservation and Management. 2020; 4(2): 83-89.

16.

Boesl

Engel

Eco

Galang

JAB

Bruckner

. Digital mapping of coastal boulders-high-resolution data acquisition to infer past and recent transport dynamics. Sedimentology. 2020; 67(11): 1393-1410.

17.

Yamamoto

Asanuma

Takahashi

Hirata

. In situ isotopic analysis of uranium using a new data acquisition protocol for 10 13 ohm Faraday amplifiers. Journal of Analytical Atomic Spectrometry. 2021; 36(3): 668-675.

18.

Mehto

Tapaswi

Pattanaik

. Virtual grid-based rendezvous point and sojourn location selection for energy and delay efficient data acquisition in wireless sensor networks with mobile sink. Wireless Networks. 2020; 26(5): 3763-3779.

19.

Kang

Zhang

Gao

. Research on neural network model for new energy industry economy based on particle swarm optimization. Journal of Intelligent and Fuzzy Systems. 2020; 38(5): 1-11.

20.

Zhou

Zhang

. Lung mass density analysis using deep neural network and lung ultrasound surface wave elastography. Ultrasonics. 2018; 89: 173-177.

21.

Barrow

Kourentzes

. The impact of special days in call arrivals forecasting: a neural network approach to modelling special days. European Journal of Operational Research. 2018; 264(3): 967-977.

22.