Research on BP Neural Network-Based Technical Skills Training

Model Under the Background of College-Enterprise Integration

Xin Chen

, Wei Wang

2b*

and Yindong Zhu

Cyberspace Security Institute, Changchun University, Changchun 130022, China

Computer Science and Technology Institute, Changchun University, Changchun 130022, China

Keywords: College-Enterprise Integration, BP Neural Network, Talent Training Model, Technical Skills, Ecosystem.

Abstract: College-enterprise integration is a new model of technical and skilled personnel training produced with the

development of education in the new era, and it is also an important way to improve the quality of higher

vocational education personnel training. In this paper, we construct a talent cultivation model based on BP

neural network, and improve each link of the talent ecological cultivation circle by constantly correcting the

weights to arrive at the values of the vocational ability indexes possessed by the graduates satisfied by em-

ployers. The MATLAB simulation tests confirmed the scientific and reasonable nature of the talent cultivation

model. The MATLAB simulation tests confirmed the scientific and reasonable nature of the talent cultivation

model.

1 INTRODUCTION

It is generally believed that technically skilled person-

nel are composite talents who can use the basic theo-

retical knowledge of technology learned to create

wealth for society, but also work in front-line posi-

tions such as production or service, and can perform

practical operations skillfully (Cui 2021). Data show

that 70% of new front-line practitioners come from

vocational colleges every year, which shows that vo-

cational colleges are the main source of technical skill

talents. The "college-enterprise integration" means

that the school adopts the form of purchasing services

to seek cooperation with enterprises and make joint

efforts to carry out practical teaching for students, i.e.,

the school provides design plans and is responsible

for personnel management, the enterprise provides

equipment, technology and teachers, and the school

and the enterprise closely integrate to organize prac-

tical training, create a real environment of practical

training positions and professional atmosphere in the

school, so that Students can get out of the classroom,

experience the workflow, and master the operation

skills, and then cultivate high-quality technical skill

talents who have the knowledge, ability, and quality

https://orcid.org/0000-0003-3615-9833

https://orcid.org/0000-0001-6571-6972

required by the occupation and can be competent to

perform the job duties(Xinhua net 2021). Therefore,

this paper proposes to build a model of technical skills

training in the context of school-enterprise integra-

tion.

2 PROBLEMS

Insufficient substantial output of school-enter-

prise cooperative education model. Most higher vo-

cational colleges and local application-oriented un-

dergraduate colleges have carried out school-enter-

prise cooperative education model. However, in the

actual implementation process, the model is too sin-

gle, and the education effect and substantial output

are insufficient. The so-called school-enterprise cur-

riculum replacement refers to the fact that enterprises

do not fully understand the teaching management

mode and assessment requirements of institutions,

and participate in the on-campus practical course

guidance method for too short a time and too much

content, with little process for students to understand,

practice and improve. The institution has too little in-

tervention in the enterprise replacement course, the

https://orcid.org/ 0000-0003-4298-0003

542

Chen, X., Wang, W. and Zhu, Y.

Research on BP Neural Network-Based Technical Skills Training Model Under the Background of College-Enterprise Integration.

DOI: 10.5220/0011915700003613

In Proceedings of the 2nd International Conference on New Media Development and Modernized Education (NMDME 2022), pages 542-547

ISBN: 978-989-758-630-9

 2023 by SCITEPRESS – Science and Technology Publications, Lda. Under CC license (CC BY-NC-ND 4.0)

process supervision is not timely, and the final enter-

prise replacement course grade is higher and lacks

distinction. Literature (Yang 2021) suggests that the

modern apprenticeship system can solve the problems

of inadequate standards of vocational education sys-

tem, lack of "application-oriented curriculum" and

lack of corporate responsibility, but it does not clarify

the specific reform plan and lacks practical effects.

Mismatch between faculty technical skills up-

grading and rapid business development. For the

training of technical skill talents, the construction of

teachers is an important link and the key to ensure the

quality of talent training. At present, the construction

and continuous training of "dual-teacher" teachers

still face many problems. Teachers in colleges and

universities are less involved in actual projects in en-

terprises and less involved in front-line production

lines, and teachers who have obtained "dual-teacher"

qualifications lack follow-up continuous learning and

exercise process, which makes technical skills fail

over time. In the literature(Jiang 2021, Yi 2021, Zhao

2021)the "1+X" certificate system is used to train

technical skill talents in a complex way, and the re-

form ideas are mainly proposed in three aspects: fac-

ulty, curriculum system and quality evaluation, but

they only emphasize the reform measures without in-

dicating specific criteria, such as talent training qual-

ity evaluation criteria.

3 BP NEURAL NETWORK

3.1 Theory

BP neural network, a multi-layer feed-forward neural

network consists of information forward transmission

and error back propagation, and the topology is di-

vided into 3 layers, i.e., input layer, implicit layer and

output layer. In the process of forward propagation of

information, the signal will then be processed from

the input layer through the implicit layer by layer and

finally reach the output layer and output the infor-

mation processing results, completing a learning pro-

cess of forward propagation. When the output layer

does not get the expected target output value, it enters

the backward propagation phase of the error, adjusts

the network weights and thresholds according to the

error value, and trains repeatedly to make the BP and

repeated training to make the predicted output of the

BP neural network The network weights and thresh-

olds are adjusted according to the error value, and the

training is repeated so that the predicted output of the

BP neural network is continuously approximated to

finally achieve the expected output and complete the

overall modeling The modeling effect is completed.

3.2 BP Neural Network Mathematical

Model

The corresponding data model is constructed accord-

ing to the principle of BP neural network, and the spe-

cific construction process (Cong 2009, Zhang 2013)

is as follows.

Let the input layer of the BP network have neu-

rons(m), the output layer has neurons(n), the middle

layer(p) has one neuron, and the training samples are

N. Then the network is set up as follows: input sample

vector 𝐴



=(𝑥



,𝑥



,… ,𝑥



), expected output vec-

tor 𝑌



=(𝑦



,𝑦



,… ,𝑦



), actual output vector 𝑌





(𝑦



,𝑦



,…,𝑦



) , network weighted input for each

layer in the middle of the network 𝑆



𝑠



,𝑠



,…,𝑠



, network output of the intermediate

layers𝐵



=(𝑏



,𝑏



,…,𝑏



), the weighted inputs of the

output layer each layer 𝐿



=𝑙



,𝑙



,…,𝑙



, the con-

nection right of the Input layer to intermediate layer

ij,

intermediate layer to output layer connection

rights v

, threshold values for each cell in the middle

layer

, threshold values for each cell of the output

layer r

. Among them in 𝑖=1,2,…,𝑝 ;𝑡=

1,2,…,𝑛;𝑘 = 1,2,…𝑁.

The whole training process is as follows:

a) Initialize and assign random

values to connection rights w

, v

and thresh-

olds

, r

b) Input training samples (A

, Y

c) Using the inputs A

, connection

weights w

and thresholds

, compute the

output of each neuron in the interlayer.

𝑏



=𝑓

(



)

,𝑠



=𝑤



𝑥



+𝜃



(1)





d) Using b

the connection weights v

and

thresholds r

, the output of each neuron in the

output layer is calculated.

𝑦



=𝑓(𝑙



), 𝑙



∑

(𝑣



𝑏







𝑟



) (2)

e) Using Y

the actual output of the net-

work 𝑌





, calculate the correction error of

each output layer unit's correction error.

𝑑



=𝑦



∙

(

1−𝑦





)

∙

(

𝑦



− 𝑦

)

(3)

f) Using v

, b

, 𝑦



 to calculate the cor-

rection error for each cell in the hidden layer.

Research on BP Neural Network-Based Technical Skills Training Model Under the Background of College-Enterprise Integration

543

𝑒



=𝑏



∙1−𝑏



∙ 𝑦







∙

(

1−𝑦





)

∙

(

𝑦



− 𝑦





)

∙ 𝑣



(4)

g) Compute the new connection between

the hidden layer to the output layer using d

, v

, r

of the new connection rights.

𝑣



(

𝐿+1

)

=𝑣



(

𝐿

)

+𝛼𝑑



𝑏



𝑟



(

𝐿+1

)

=𝑟



(

𝐿

)

+ 𝛼𝑑



(L indicates

the number of training) (5)

h) Use e

, x

to compute the input

layer to the hidden layer of the new connec-

tion rights between.

𝑤



(

𝐿+1

)

=𝑤



(

𝐿

)

+𝛽𝑒



𝑥



𝜃



(

𝐿+1

)

=𝜃



(

𝐿

)

+𝛽𝑒



(6)

i) Select the second set of training sam-

ples and repeat steps c to h. until all N sam-

ples have been trained.

j) Start the second training, starting from

the first sample, and Repeat steps c to h until

the global error function e is less than the pre-

defined value (network error) or the number

of training sessions reaches The whole train-

ing process is finished.

4 TALENT CULTIVATION

ECOSYSTEM

In the process of talent cultivation in institutions, the

concept of "talent cultivation ecosystem" has been

proposed in the literature, which means that multiple

subjects within the educational ecosystem are inter-

connected and closely integrated, while energy, ma-

terial and information are exchanged outside the sys-

tem to form a larger synthesis with the educational

ecosystem (Ling 2005). The details are shown in the

figure1.

Figure 1: Talent Cultivation Ecosystem

5 BUILDING A BP NEURAL

NETWORK MODEL

TECHNICAL SKILLS

TRAINING

BP neural network has the characteristics of forward

transmission of input signal, backward propagation of

insufficient information and nonlinearity, which are

associated with this talent cultivation model. This pa-

per takes the employment situation of the graduates

of Changchun University in 2020 and 2021 as an ex-

ample, and evaluates the technical skill cultivation

level in terms of students' vocational ability, and di-

vides it into two levels of vocational technical skill

level evaluation and vocational literacy level for con-

struction.

5.1 Build Model

The structure of the BP neural network is determined

according to the input and output data characteristics

of the system. The evaluation index system is estab-

lished by influencing certain key factors of talent

training, and the weight of each index is determined

by the neural network after training through adaptive

learning. As shown in Table 1-2.

Table 1: Students' professional competence evaluation in-

dex system—Tier 1

Professional

competence

evaluation

index sys-

tem

Tier 1 Indicator--Vocational technical

skill level

Usual grades; Hands-on assignments;

Scientific Research Competition; Stage

Assessment; Professional qualification

level certificate; Academic Certificates;

Internship Certification; Industry Certi-

fication

Table 2: Students' professional competence evaluation in-

dex system—sec

Professional

competence

evaluation

index sys-

tem

Secondary indicator-- Professional qual-

ity level

Thoughtful Literacy; Psychological Lit-

eracy; Behavioral Literacy; Social Liter-

In this paper, the set of occupational competence

levels is divided into four levels: = {excellent, good,

pass, fail}. According to Table 1, the input layer is

assumed to be 12 nodes, i.e., the 12 indicators mainly

included in the influence of occupational competence

as the input layer. The evaluation result is the output

of the network, and its indicators are taken as the em-

ployer satisfaction to establish the original model; the

NMDME 2022 - The International Conference on New Media Development and Modernized Education

544

evaluation result is the output of the network, so the

number of output layers is assumed to be n=1. In the

neural network of BP algorithm, the selection of the

number of nodes in each layer has a great impact on

the performance of the network, and if there are too

many nodes in the implicit layer, then it will have a

negative impact on the generalized reasoning ability

of the network, i.e., it will affect the network for new

input adaptability. A small number of nodes in the

hidden layer also affects the accuracy of network

learning and increases the number of local minima, so

the number of nodes in the hidden layer needs to be

chosen appropriately. Currently, the number of nodes

in the implicit layer is mainly based on experience,

and according to the empirical formula, the number

of implicit layers is 5; the S-type function is generally

used as the neuron conversion function of BP neural

network, and the specific function form is:

f(x)=1/(1+e^(-x) ) (7)

The insufficient information is passed backwards

through training, and then analyzed and adjusted the

structure and data to further optimize the talent train-

ing model. The algorithm model of talent training

based on BP neural network includes BP neural net-

work construction, BP neural network training and

BP neural network fitting walk. The specific algo-

rithm flow is shown in Fig2.

Figure 2: Schematic diagram of BP neural network algo-

rithm

5.2 Implementation

In the implementation process of the model, accord-

ing to the criteria and numbers of each level of voca-

tional ability level division, as well as according to

the employment quality report of Changchun Univer-

sity, the satisfaction of employers to graduates (very

satisfied, relatively satisfied, average, unsatisfied)

against the level of students' vocational ability, as

shown in Table 3.

Table 3: Occupational competence level details

occupa-

tional

compe-

tence

levels

Score

Criteria

Number （

2020）

Employer sat-

isfaction（

2020）

Excellent 90-100 2751 68.14%

Goo

75-89 1009 25%

Pass 60-74 139 3.43%

Fail 59- 139 3.43%

5.3 Results and Analysis

A three-layer BP neural network model is constructed

and then implemented in simulation using MATLAB.

This software is widely used in the field of data anal-

ysis and can analyze the most valuable information in

the shortest possible time. Since the concept of talent

training ecosystem was proposed, it has been of great

significance in the process of vocational education re-

form and practice, and has successfully cultivated a

batch of high-quality technical skill talents. This cul-

tivation model can be described as a spiral structure

model. In order to prove the correct guiding role of

the spiral theoretical structure model in practical edu-

cation, BP neural network is used to test and prove it.

Therefore, more than 8,000 sample data from the last

2 years were taken as training test data. Normalizing

the sample data, the formula is

𝑥



=(𝑥



−min

(

𝑥

)

)/(max

(

𝑥

)

−

min

(

𝑥

)

)(𝑥



𝑖𝑠 𝑎 𝑐𝑒𝑟𝑡𝑎𝑖𝑛 𝑠𝑎𝑚𝑝𝑙𝑒 𝑜𝑓 𝑑𝑎𝑡𝑎 ,𝑖 =

1,2,…,𝑛) (8)

The error accuracy is set to”1e-8”(Error sum of

squares), Training function is “trainlm”, Learning

Rate is “1r=0.4”,”err-goal=1e-5”,

“max_epoch=100”.

net=newff(minmax(p),[s1,s2],{‘tansig’,

’purelin’},’trainlm’);

net.trainParam.lr=0.4;

net.trainParam.show=5;

net.trainParam.epochs=100;

net.trainParam.goal=1e-5;

net.trainParam.min_grad=1e-8;

The sample data are trained iteratively at third

time (Figure 3), and the actual output fits very well

with the target output. It is easy to see that the errors

Research on BP Neural Network-Based Technical Skills Training Model Under the Background of College-Enterprise Integration

545

obtained from training the sample data for the last two

years are very small, basically hovering around, so

this network model is considered to be very success-

ful, and it also confirms the scientific validity and rea-

sonableness of this model in the talent training eco-

system. the training generations of the BP model are

shown in Figure 4.

Figure 3: The effect of training and fitting

Figure 4: Network training algebra

The test data of this model are the specific data of

vocational ability and level of graduates in 2021,

which are conducted according to Tables 1 to 3. The

absolute and relative errors obtained from the training

tests are shown in Table 4.

Table 4: Absolute and Relative Errors

occupational

competence

levels

Actual

number

Predicted

number

AE RE/%

Excellen

3209 3200 9 0.2

Goo

629 625 4 0.6

Pass 130 129 1 0.7

Fail 130 128 2 1.5

6 PRACTICE EFFECTIVENESS

According to the results of the tracking survey and

external evaluation of the training quality of gradu-

ates, all network engineering students of Changchun

University participate in innovation and entrepre-

neurship practice, and are awarded more than 10 in-

novation and entrepreneurship projects at provincial

level and above every year; in the past three years,

they have won in the skills competition, network se-

curity competition, robotics competition, program-

ming competition and other events In the past three

years, they have won more than 80 awards at the pro-

vincial level or above in skills competitions, network

security competitions, robotics competitions and pro-

gramming competitions.

The exploration and practice of "excellent engi-

neer" training in computer science under the deep in-

tegration mode of industry-university was awarded

the second prize of teaching achievement in Jilin

Province; 2 golden courses were built in the institute;

the first 32km-long quantum-secure communication

demonstration network based on commercial optical

fiber was built in the northeast provinces for textbook

research practice. The first 32km long quantum con-

fidentiality communication demonstration network

based on commercial optical fiber has been built; the

faculty team has been reasonably constructed, and

nearly all of them are equipped with "double teach-

ers". The faculty team is reasonably constructed, with

nearly all of them possessing "double teacher" quali-

fication.

7 CONCLUSIONS

The cultivation of technical skills is a long-term pro-

cess, and students need to accumulate and improve

their technical skills under the framework of lifelong

learning if they are to be equipped with more tech-

nical skills that can be applied in practice. By using

BP algorithm to build talent training data model, com-

bined with the annual school dynamic data analysis

and subject research results, quantitative analysis and

optimization, tracking to find the lack of indicators

and the number of deficiencies affecting talent train-

ing, and constantly revised. However, the program

has not been promoted, and we will cooperate with

other institutions and enterprises in the future to fur-

ther optimize the talent training model, adjust the net-

work weights and thresholds, so that the predicted

output of the BP neural network can be approximated

NMDME 2022 - The International Conference on New Media Development and Modernized Education

546

to reach the desired output, and provide a more scien-

tific basis for vocational education talent training.

ACKNOWLEGEMENTS

Jilin Province Vocational Education Project

(2021XHZ040)

REFERENCES

CONG S. Neural Network Theory and Applications for

MATLAB ToolboX[M]. Hefei: University of Science

and Technology of China Press,2009.

CUI F J. What is Technical Skilled Personnel?[J]. Industrial

Technology and Vocational Education,

2021,19(03):3.DOI:10.16825/j.cnki.cn13-

1400/tb.2021.03.001.

JIANG P, HAO J, MENG S Y. Reform practice of BIM

technical and technical personnel training mode for

higher vocational architectural design majors under the

background of "1+X" certificate system[J]. Guangxi

Journal of Light Industry, 2021(31):137-140.

LING L, HE Z B. Regional Educational Planning from the

Perspective of Educational Ecology[J]. Research in Ed-

ucational Development, 2005(9):66-68.

XINHUA net. Opinions on promoting the high-quality de-

velopment of modern vocational education.[OL]

.http://www.moe.gov.cn/jyb_xxgk/moe_1777/moe_17

78/202110/t20211012_571737.html.

YANG H J. An Analysis of the Training Mode of Technical

Skilled Personnel——Based on the Integrated Con-

struction of Schools and Enterprises under the Modern

Apprenticeship System[J]. Journal of Yuzhang Normal

University, 2021,36(05):114-118.

YI Y, RONG X, DING M J. Exploration on the cultivation

of compound technical and technical talents in intelli-

gent manufacturing professional groups in higher voca-

tional colleges from the perspective of 1+X certificate

system[J]. Education and Vocation, 2021(16):65-

68.DOI:10.13615/j.cnki.1004-3985.2021.16.010.

ZHAO D, WANG J, LIU M X. Based on "1+X", research

on the training path of compound technical and tech-

nical talents in engineering construction majors[J]. Vo-

cational Technology, 2021,20(09):38-

42.DOI:10.19552/j.cnki.issn1672-0601.2021.09.007

ZHAO Z Y. ZAHNG P Z. The construction and implemen-

tation of higher vocational student learning evaluation

model based on BP neural network[J]. Journal of Hubei

University of Education,2013,30(08):75-76.

Research on BP Neural Network-Based Technical Skills Training Model Under the Background of College-Enterprise Integration

547