sensors
Article
Development of Cotton Picker Fire Monitoring System Based
on GA-BP Algorithm
Weipeng Zhang 1 , Bo Zhao 1, *, Shengbo Gao 1 , Yuankun Zheng 2 , Liming Zhou 1 and Suchun Liu 1
Citation: Zhang, W.; Zhao, B.; Gao, S.;
Zheng, Y.; Zhou, L.; Liu, S.
Development of Cotton Picker Fire
Monitoring System Based on GA-BP
Algorithm. Sensors 2023, 23, 5553.
https://doi.org/10.3390/s23125553
Academic Editors: Nahina Islam and
Dinko Oletic
Received: 19 April 2023
Revised: 2 June 2023
Accepted: 9 June 2023
Published: 13 June 2023
Copyright: © 2023 by the authors.
Licensee MDPI, Basel, Switzerland.
This article is an open access article
distributed under the terms and
conditions of the Creative Commons
Attribution (CC BY) license (https://
creativecommons.org/licenses/by/
4.0/).
Sensors 2023, 23, 5553. https://doi.org/10.3390/s23125553
1 National Key Laboratory of Agricultural Equipment Technology, Chinese Academy of Agricultural
Mechanization Sciences, Beijing 100083, China; zhwellpeng@gmail.com (W.Z.); gaoshengbo95@163.com (S.G.);
caamsjds309@gmail.com (L.Z.); suchun1804@163.com (S.L.)
2 Department of Electrical and Mechanical Engineering, College of Engineering, China Agricultural University,
Beijing 100089, China; uiagri.caams@gmail.com
* Correspondence: zhaoboshi@126.com
Abstract: Due to the characteristics of the cotton picker working in the field and the physical
characteristics of cotton, it is easy to burn during the operation, and it is difficult to be detected,
monitored, and alarmed. In this study, a fire monitoring system of cotton pickers based on GA
optimized BP neural network model was designed. By integrating the monitoring data of SHT21
temperature and humidity sensors and CO concentration monitoring sensors, the fire situation
was predicted, and an industrial control host computer system was developed to monitor the CO
gas concentration in real time and display it on the vehicle terminal. The BP neural network was
optimized by using the GA genetic algorithm as the learning algorithm, and the data collected by the
gas sensor were processed by the optimized network, which effectively improved the data accuracy
of CO concentration during fires. In this system, the CO concentration in the cotton box of the cotton
picker was validated, and the measured value of sensor was compared with the actual value, which
verified the effectiveness of the optimized BP neural network model with GA. The experimental
verification showed that the system monitoring error rate was 3.44%, the accurate early warning rate
was over 96.5%, and the false alarm rate and the missed alarm rate were less than 3%. In this study,
the fire of cotton pickers can be monitored in real time and an early warning can be made in time, and
a new method was provided for accurate monitoring of fire in the field operation of cotton pickers.
Keywords: cotton picker; fire monitoring; GA optimized BP neural network; test verification
1. Introduction
China is the largest cotton producer and consumer country in the world. By 2019, the
cotton planting area exceeded 300 million hm2 and the output was about 5.889 million
tons [1]. As the major cash crop and strategic material of the national economy, its planting
area in Xinjiang is expanding continuously. Due to the labor shortage, cotton picker
machinery became the main machinery for cotton harvesting. It can improve cotton picking
efficiency, reduce labor costs, and increase economic benefits. However, the cotton picker
machinery can easily cause a fire during the process of picking cotton. The main causes of
fires are particularity of climate, terrain, as well as the physical characteristics of cotton,
such as flammability and smoldering [2]. Therefore, the fire monitoring of cotton picker
machinery is particularly important.
The main reason for fire in cotton pickers is the high concentration of inflammable
gases in the chassis during the process of operation [3]. Therefore, the gas sensor is used
to monitor the fluctuation of the concentration of inflammable gases in the cotton box of
the cotton picker to reduce the fire rate [4]. A gas sensor can be used to monitor the CO
concentration. The principle is that CO diffuses to the sensor and undergoes a reduction
reaction [5] or oxidation reaction [6] on the catalytic electrode, and a current is generated
on the working electrode [7]. The concentration of CO gas can be obtained by sampling
https://www.mdpi.com/journal/sensors
Sensors 2023, 23, 5553 2 of 13

and processing the current [8]. It is worth noting that high-precision gas monitoring sensor
probes and strong sensor devices should be selected during the process of sensor selec-
tion [9]. Due to the accumulation of a large amount of cotton in the cotton box, the harvested
cotton increases the difficulty of monitoring gas fluctuation with the ordinary sensor [10]
and causes friction damage to the sensor [11]. The neural network is an operational model
composed of a large number of connections between nodes [12]. It has an excellent ability
to learn data features, so the prediction of fire situation can be realized by using the neural
network to learn the data monitored by sensors [13]. One of the urgent problems to be
solved is how to introduce the neural network algorithm into the fire monitoring of cotton
picker and improve the monitoring performance [14]. Chen [15] integrated multi-modal
sensors to obtain data of carbon monoxide, smoke, temperature, and humidity, and devel-
oped an indoor fire alarm system. The accurate fire warning information can be quickly
obtained through the support vector machine algorithm. Sulistian [16] designed a fire early
warning system, developed an alarm device, compared four machine learning classification
methods, and determined that the classification algorithm can improve the performance of
the intelligent fire alarm system.
It can be seen from the literature that the existing cotton pickers do not consider
the problem of fire prevention, nor do they install relevant sensors on the cotton pickers.
There is little research on fire prediction systems, and the fire prevention system of the
cotton picker is very necessary. The optimized BP neural network model based on genetic
algorithm was widely used in the prediction of agricultural machinery [17–21]. It has high
accuracy and can be used for effective prediction. A fire monitoring system of a cotton
picker based on the BP neural network model optimized by the GA genetic algorithm
was designed.
In this paper, the CO concentration monitoring sensor and SHT21 temperature and
humidity sensor were adopted to monitor the cotton box in real time. At the same time,
the BP neural network was used to establish the prediction model of CO concentration,
and the GA genetic algorithm was then used to accelerate the convergence of weight and
threshold of BP neural network to optimize the prediction model. The prediction model of
CO concentration can predict the probability of fire when the data collected by gas sensor
were input into the prediction model. Finally, the automatic alarm and real-time monitoring
can be achieved by transmitting the monitoring data of gas and the predicting data of fire to
the on-board industrial control terminal system. Prior to this study, an intelligent detection
and monitoring system for cotton bale storage based on FFID was designed, laying the
foundation for this research.

2. Materials and Methods

2.1. Overall System Design
The overall structure of the cotton picker fire monitoring system consists of three
parts: a CO concentration sensor, a temperature and humidity sensor, and an on-board host
computer. The CO concentration sensor mainly monitors the changes in CO concentration
in the cotton box of the cotton picker. The temperature and humidity sensor mainly
monitors the change of temperature and humidity in the cotton box to prevent smoldering.
The data collected by the sensor were transmitted to the host computer system through
RS485 communication serial port and stored. Then, the stored data were trained by GA-
BP model to obtain an optimized model. As a display unit, the on-board terminal can
display the sensor data in real time, and can set thresholds to realize automatic alarm
and other functions. The overall structure of the cotton picker fire monitoring system is
shown in Figure 1. When a fire occurs, the CO sensor will detect the corresponding gas
concentration. When smoldering occurs, the temperature and humidity sensor will detect
the changes in temperature and humidity inside the box. Then, the industrial control screen
will display the curves of co concentration, temperature, and humidity. Then, the GA-BP
algorithm will predict the threshold of gas concentration, temperature, and humidity based
on the data collected by the sensor. Finally, according to the predicted values, the upper
Sensors
Sensors 2023,
2023, 23,
23, xx FOR
FOR PEER
PEER REVIEW
REVIEW 33 of
of 13
13

Sensors 2023, 23, 5553 3 of 13

GA-BP
GA-BP algorithm
algorithm will
will predict
predict the
the threshold
threshold ofof gas
gas concentration,
concentration, temperature,
temperature, and
and hu-
hu-
midity
midity based on the data collected by the sensor. Finally, according to the predicted val-
based on the data collected by the sensor. Finally, according to the predicted val-
computer
ues,
ues, the in thecomputer
the upper
upper cab sets off
computer inthe
in thesound
the cab and
cab sets
sets offlight
off the alarm and
the sound
sound to
andcomplete
light the fire
light alarm
alarm to monitoring
to complete
complete theand
the fire
fire
warning
monitoring function.
monitoring andand warning
warning function.
function.

Figure 1.
Figure1.
Figure The
1.The overall
Theoverall structure
overallstructure of
structureof the
ofthe fire
thefire monitoring
firemonitoring system
monitoringsystem of
systemof the
ofthe cotton
thecotton picker.
cottonpicker.
picker.

2.2.
2.2.Hardware
2.2. HardwareSystem
Hardware SystemDesign
System Design
Design
The
The hardware
The hardware
hardware ofof cotton
of cotton picker
cotton picker fire monitoring
picker fire
fire monitoringand
monitoring andearly
and earlywarning
early warningsystem
warning systemincludes
system includesaa
includes
aCOCO
CO gasgas monitoring
gas monitoring module,
monitoring module, a temperature
module, aa temperature
temperature and and humidity
and humidity monitoring
humidity monitoring module,
monitoring module,
module, and
andand a
aa in-
in-
industrial
dustrial controlhost
dustrial control
control hostcomputer.
host computer.The
computer. Thestructure
The structureof
structure ofofhardware
hardwaresystem
hardware systemis
system isisshown
shownin
shown ininFigure
Figure2.
Figure 2.2.

Figure 2.
Figure2. Hardware
2.Hardware system
Hardwaresystem structure
systemstructure diagram.
structurediagram.
diagram.
Figure

2.2.1.
2.2.1.
2.2.1.CO
COGas Monitoring
GasMonitoring
MonitoringModuleModule
Module
In
In order
In order to
order to effectively
to effectively
effectively improve
improve the monitoring accuracy
the monitoring accuracy of of the
the gas
gas sensor,
sensor, aa high-
high-
precision
precision gas
gas concentration
concentration monitoring
monitoring sensor
sensor was
was adopted
adopted inin CO
CO
precision gas concentration monitoring sensor was adopted in CO gas concentration mon- gas
gas concentration
concentration mon-
moni-
itoring module, with working temperature: −−30 ◦ C~+50 ◦°C,
toring
itoringmodule,
module,withwithworking
working temperature:
temperature: −30 °C~+50
30 °C~+50 °C, working
C, working humidity:
humidity:≤
workinghumidity: ≤90%RH,
90%RH,
≤90%RH,
DC24V ±
DC24V
DC24V ± 15%.
± 15%. The
15%.The sensor
Thesensor had
sensorhad had an LED
ananLEDLED digital tube
digital
digital tube display
tube display
display function, the
function,
function, CO
the the
CO CO gas measure-
gas gas mea-
measure-
ment
ment range
surement range was
range 0–1000
waswas ppm,
0–1000
0–1000 ppm, the
the resolution
ppm, the resolution
resolution was 11 ppm,
waswas and
1 ppm,
ppm, andandthe
the response
the response
response time
timetime 20≤
≤≤ 20 s. The
s.20 s.
The
sensor
The
sensor can
can monitor
sensor can CO
CO with
monitor
monitor high
CO with
with precision,
highhigh wide
precision,
precision, widewiderange,
range, short
range, response
shortshort time,
response
response and
and more
time,time, and
more
more
accurateaccurate
accurate monitoring.
monitoring.
monitoring. In In addition,
Inaddition,
addition, the the output
the output
output mode
modewas mode
was was a three-wire
aa three-wire
three-wire system,system,
system, with with a
withaacurrent
current
current
signal signal
range range
4~20 mA 4~20
and mA
the and
outputthe output
signal of signal
RS485 of RS485
isolation.
signal range 4~20 mA and the output signal of RS485 isolation. The incomplete combus-isolation.
The The
incomplete incomplete
combus-
combustion
tion
tion of
of flames
flamesofwill
flames
will will generate
generate
generate CO
CO with
withCO withamount
aa large
large a large amount
amount of
of smoke,
smoke, of so
smoke,
so the so the monitoring
the monitoring
monitoring module
module
module
needs
needs to needs
to measureto measure
measure the the
the fluctuation fluctuation
fluctuation of of CO of CO
CO concentration concentration
concentrationin in the
the air in
airin the
in real air in
real time. real
time. The time.
The gas The
gassensor
sensor
gas sensor included a digital tube display, real-time calibration, high-precision probe, and
Sensors 2023,
Sensors 23,23,
2023, x FOR PEER
x FOR REVIEW
PEER REVIEW 4 4of of1313
Sensors 2023, 23, 5553 4 of 13

included
includeda adigital
digitaltube
tubedisplay,
display,real-time
real-timecalibration,
calibration,high-precision
high-precisionprobe,
probe,andandmetal
metal
shell, as
metalas
shell, shown
shell, in Figure
as shown
shown 3. The
in Figure
in Figure sensor
3. The3.sensor was
The sensorpowered by
was powered
was powered 2 24 V
by V
by 2 24 power
2 24 supply,
V power
power and
andthe
supply,
supply, and
the
change
change ofofpotential
the change potentialofofoutput
of potential output terminal
of output
terminal was
wasdirectly
terminal was related
directly
directly totothe
related
related change
to
the ofofthe
the change
change CO
of
the gas
the
CO CO
gas
concentration.
gas concentration.
concentration.

Figure 3. 3.
Figure
Figure Structure
3. block
Structure
Structure diagram
block
block ofof
diagram
diagram CO
of gas
CO
CO monitoring
gas
gas sensor.
monitoring
monitoring sensor.
sensor.

2.2.2.
2.2.2.
2.2.2. Temperature
Temperature
Temperature andand
and Humidity
Humidity
Humidity Module
Module
Module
The
The
The SHT21
SHT21
SHT21 temperature
temperature
temperature and
and
and humidity
humidity
humidity sensor
sensor
sensor was
was
was selected
selected
selected forfor
for thethe
the temperature
temperature
temperature and
and
and
humidity detection module. The temperature range of the sensor was: − 30~70 ◦ C (±0.2 ◦ C)
humidity
humidity detection
detection module.
module. The temperature
The temperature range
range ofof
the
thesensor
sensor was:
was: −30~70
−30~70 °C°C (±0.2
(±0.2°C)
°C)
and
and thethe humidity
humidity range
range of of
the the sensor
sensor was was 0~100%RH
0~100%RH ( ±
(±3%RH).3%RH).
and the humidity range of the sensor was 0~100%RH (±3%RH). The ◦response◦ time was◦ <1 The The response
response time time
was was
<1
<1
S,S, s, temperature
temperature
temperature ofof of working
working
working environment
environment
environment temperature
temperature
temperature was
was−55 was
−55 −55 °CC~120
°C~120
°C~120 °C(±0.1
(±0.1 C
°C),
°C),±and
(and0.1theC),
the
and the of
humidity humidity
working ofenvironment
working environmentwas ≤95%RH ≤95%RH (±
was(±0.1%RH). 0.1%RH).
The RS485 The RS485
with standard with
humidity of working environment was ≤95%RH (±0.1%RH). The RS485 with standard
standard Modbus-RTU
Modbus-RTU protocol was adopted for communication serial port, with the
Modbus-RTUprotocol protocolwas wasadopted
adoptedfor forcommunication
communicationserial serialport,
port,withwiththe thedevice
devicede- de-
device
fault default
address 01.01.address
The 01.
temperature The temperature
and humidity and humidity
module had module
a sensor had
probe, a sensor
temperature probe,
fault address The temperature and humidity module had a sensor probe, temperature
temperature
and and humidity digital display, sensor housing, and communication interface,
and humidity digital display, sensor housing, and communication interface, asshown
humidity digital display, sensor housing, and communication interface, as showninin
as
Figureshown in Figure 4. Among them, the sensor probe had the function of high-precision
Figure4.4.Among
Amongthem, them,the thesensor
sensorprobe
probehad hadthethefunction
functionofofhigh-precision
high-precisioninformation
information
information
acquisition. acquisition.
It It
can display It can
temperaturedisplay temperature
and humidity and humidity
changes inin changes
real time, andin can
real convert
time, and
acquisition. can display temperature and humidity changes real time, and can convert
can
the converttemperature
received the receivedvariabletemperature
signals variable
into signals into output
standardized standardized
signal output
and inputsignal
themand
the received temperature variable signals into standardized output signal and input them
input
into the them into
industrial the industrial
control computer control
after computer
processing. after
Theprocessing.
advantages The
ofof advantages
the sensor of
probe the
into the industrial control computer after processing. The advantages the sensor probe
sensor probe are fast response speed and high accuracy.
are
arefast
fastresponse
response speed
speed and
and high
highaccuracy.
accuracy.

Figure
Figure 4.
4. 4. Structure
Structure block
block diagram
diagram of
ofof temperature
temperature and
and humidity
humidity monitoring
monitoring sensor.
sensor.
Figure Structure block diagram temperature and humidity monitoring sensor.
2.2.3. Host Computer Display Module
2.2.3. Host
2.2.3. HostComputer
Computer Display Module
The MCGS touchDisplay Module Technology, Shenzhen, China) used as the human–
screen (Kunlun
The
The
computer MCGS
MCGS touch
touch
interaction screen
screen
system(Kunlun
(Kunlun
module Technology,
Technology,
of the upperShenzhen,
Shenzhen,
computer China)
China)used
used
monitoring asterminal
the
as thehuman–
human–
had a
computer
computer interaction
interaction system
system module
module of the
of theupper
upper computer
computer monitoring
monitoringterminal
terminal
4-core (main frequency 1 GHz) CPU as the core. Additionally, the communication interface had
hada 4-
a 4-
core (main
core frequency
(main1*RS232;
included frequency 1 GHz) CPU
1 GHz) 1*LAN
2*RS485; as the core.
CPU as network Additionally,
the core. Additionally, the
port, built-in thecommunication
MCGScommunication interface
professionalinterface
version,
included
included
four-wire1*RS232;
1*RS232;
resistive2*RS485;
2*RS485; 1*LAN
1*LAN
touch screen, network
networkport,
TPC1031Ki, port,built-in
built-in
resolution MCGS
MCGS
1024 professional
professional
× 600, supported version,
version,
a serial
four-wire
four-wire resistive
resistivetouch
port multi-protocol touch screen,
screen,TPC1031Ki,
communication, and canresolution
TPC1031Ki, resolution1024*600,
communicate 1024*600,supported
directly with theaCO
supported aserial port
serial port
sensor, as
 Sensors 2023, 23, x FOR PEER REVIEW 5 of 13
Sensors 2023, 23, 5553 5 of 13

multi-protocol communication, and can communicate directly with the CO sensor, as well
well as temperature
as the the temperature
and and humidity
humidity sensor.
sensor. As shown
As shown in Figure
in Figure 5, the
5, the upper
upper computer
computer dis-
display included
play included communication
communication status
status information,
information, thethe temperature
temperature andand humidity
humidity of
of the
the cotton
cotton boxbox of the
of the cotton-picking
cotton-picking machine,
machine, andCO
and the thegas
COconcentration
gas concentration
value.value. The
The upper
upper computer
computer had anhad an indicator
alarm alarm indicator light,
light, etc., etc.,can
which which can the
display display
changetheofchange
the fireof the
warn-
fire warning situation in the cotton box in real time, and the observation
ing situation in the cotton box in real time, and the observation was simple and conven-was simple
and
ient.convenient.

Figure5.5.Host
Figure Hostcomputer
computersystem.
system.

2.3.
2.3.Software
SoftwareSystem
SystemDesign
Design
2.3.1.
2.3.1. Host ComputerSystem
Host Computer SystemDesign
Design
In
Inthe
thedesign,
design,the
thecollection
collectionofoftemperature
temperatureand andhumidity,
humidity,COCOconcentration
concentrationof ofthe
the
sensing module, logical judgment and alarm, data transmission, and display
sensing module, logical judgment and alarm, data transmission, and display of monitor-of monitoring
data werewere
ing data realized. Firstly,
realized. thethe
Firstly, system
systemwas initialized,
was including
initialized, includingRS485
RS485communication
communication
rate
rateand
andmode
modesetting,
setting,temperature
temperatureand andhumidity
humiditysensor,
sensor,COCOconcentration
concentrationsensor
sensorandand
alarm threshold setting, etc., and the prescription chart was loaded. The
alarm threshold setting, etc., and the prescription chart was loaded. The initial alarminitial alarm
threshold
thresholdwaswas50
50ppm,
ppm,butbutthe
thethreshold
thresholdvalue
valuewould
wouldbe beaffected
affectedbybythe
theenvironment,
environment,andand
the
the algorithm would be used to predict it later. Then, the data receiving timerwas
algorithm would be used to predict it later. Then, the data receiving timer was started,
started,
and the information of CO sensor and temperature and humidity sensor was received
and the information of CO sensor and temperature and humidity sensor was received and
and processed through the RS485 communication line. At the same time, the dynamic
processed through the RS485 communication line. At the same time, the dynamic concen-
concentration change curve was obtained to complete the fire warning. The software
tration change curve was obtained to complete the fire warning. The software program
program flow is shown in Figure 6.
flow is shown in Figure 6.
2.3.2. System Algorithm Design
(1) GA-BP neural network definition
The BP back propagation algorithm is a well-known method for training multilayer
feedforward artificial neural networks. By adding gradually trained data and hidden units,
the learning convergence speed can be accelerated to avoid falling into local minimum. The
GA algorithm is a method to search the optimal solution by simulating the natural evolu-
tion process. The GA-optimized BP algorithm model can improve the global optimization
ability of traditional GA algorithm and BP algorithm, respectively, and improve the pre-
diction efficiency and accuracy of the algorithm. Therefore, through the CO concentration
and temperature and humidity information uploaded by the gas sensor, the GA genetic
algorithm was used to train the weights and thresholds of the fire monitoring and early
warning neural network, which can avoid the problem of falling into the local optimal
solution due to the use of the traditional BP neural network, and can improve the accuracy
of fire monitoring and early warning.
 Sensors 2023, 23, 5553 6 of 13
Sensors 2023, 23, x FOR PEER REVIEW

(2) GA-BP neural network model construction

Figure
Figure 6. Fire
6. Fire warning
warning systemflow
system software software
chart. flow chart.
As shown in Figure 7, the specific algorithm process of optimizing the neural network
2.3.2. System
connection weightAlgorithm Design
based on genetic algorithm was as follows:
(1) GA-BP neural network definition
The BP back propagation algorithm is a well-known method for train
feedforward artificial neural networks. By adding gradually trained data an
the learning convergence speed can be accelerated to avoid falling into lo
The GA algorithm is a method to search the optimal solution by simulati
Sensors 2023, 23, x FOR PEER REVIEW 7 of 13

(2) GA-BP neural network model construction

Sensors 2023, 23, 5553 7 of 13
As shown in Figure 7, the specific algorithm process of optimizing the neural net-
work connection weight based on genetic algorithm was as follows:

Figure 7. 7.
Figure Algorithm
Algorithmblock
blockdiagram ofGA
diagram of GAtotooptimize
optimizethethe weights
weights of neural
of BP BP neural network.
network.

(1)(1)A Asetset
ofofdistributions
distributions are randomlygenerated
are randomly generatedthrough
through thethe information
information of COofcon-
CO con-
centration, temperature, and humidity uploaded by the gas sensor, and each
centration, temperature, and humidity uploaded by the gas sensor, and each weight weight
in in
thethe groupisisencoded
group encoded by
by aa certain
certaincoding
codingscheme,
scheme,so so
as to
as construct a code
to construct chain.chain.
a code
Under the premise that the network structure and learning algorithm have been
Under the premise that the network structure and learning algorithm have been de-
determined, the code chain corresponds to the neural network with specific weights
termined, the code chain corresponds to the neural network with specific weights
and thresholds.
(2)andThethresholds.
error function of the generated neural network is calculated to determine its
(2) The errorfunction
fitness function of the
value, andgenerated
the error isneural network
inversely is calculated
proportional to determine its fit-
to the fitness.
(3)ness
Thefunction
individuals are sorted by the method of fitness ratio, and a number
value, and the error is inversely proportional to the fitness. of individuals
with larger fitness values are selected and inherited directly to the next generation.
(3)(4)The individuals are sorted by the method of fitness ratio, and a number of individuals
Crossover operation is performed, and two individuals are randomly selected from
with
thelarger fitness
population to values arewith
exchange selected and inherited
each other directly
according to the set to the next generation.
probability.
(4)(5)Crossover
The mutation operation
operation is carried out,
is performed, and random mutation are
two individuals points are defined,
randomly and from
selected
improved genetic operators are used to form a new generation of groups
the population to exchange with each other according to the set probability.through
adaptive adjustment of individuals such as crossover and mutation.
(5)(6)The mutation operation is carried out, random mutation points are defined, and im-
Steps (2) to (5) are repeated to make a set of initially determined weight distribution
proved
evolvegenetic operators
continuously untilare
theused to form
training a new
target generation
is satisfied or the of groups
number of through
iterationsadap-
tive adjustment of individuals
reaches the preset target. such as crossover and mutation.
(6) Steps (2)paper,
In this to (5) the
are traditional
repeated toBPmake a set of neural
feed forward initially determined
network model,weight distribution
the IPSO opti-
evolve
mized continuously
BP neural network until the
model, training
and target
the GA-BP is satisfied
neural network or the were
model number of iterations
trained, the
training results
reaches were obtained,
the preset target. and the best neural network model was selected to carry
out the fire prediction analysis of the cotton picker. The model capabilities were compared
In this paper, the traditional BP feed forward neural network model, the IPSO opti-
by NSE, MAE, RMSE, and the formula was as follows:
mized BP neural network model, and the GA-BP neural network model were trained, the
training results were obtained, and the best n neural network model was selected to carry
2
∑ ( Xiρ − Xiξ )
out the fire prediction analysis of the cotton picker. The model capabilities were compared
NSE = 1 − i=n1 (1)
by NSE, MAE, RMSE, and the formula was as ( Xfollows:
2
∑ −X ) iρ 0
i =1 n

( X ρ − X ξ )
i i
2

NSE = 1 − i =1
n (1)
 ( Xiρ − X 0)2
i =1
 Sensors2023,
Sensors 2023,23,
23,5553
x FOR PEER REVIEW 88 of
of13
13

n
n  Xρ − Xξ
i i
(2)
MAE∑= iX=1
iρ − Xiξ
MAE = i =1 n (2)
n
n
v  ( Xi ρ − Xiξ )2
u n i =1
u= (3)
RMSE u ∑ ( Xiρ n− Xiξ )2
RMSE = i=1
t
(3)
where Xiρ is Actual value uploaded by gas sensor. Xiξ is predicted value by GA-BP
n
where Xiρ is Actual value uploaded by gas sensor. Xiξ is predicted value by GA-BP
algorithm. XX is
algorithm.
0 is arithmetic mean of the actual values uploaded by the gas sensor. n is
arithmetic mean of the actual values uploaded by the gas sensor. n is
0
number of data points.
number of data points.

2.4.Test
2.4. TestVerification
Verification
Themodel
The modelcotton
cottonpicking
pickingbox boxwas
wastaken
taken asas
thethe research
research object.
object. AsAs shown
shown in Figure
in Figure 8,
8, the
the length,
length, width,
width, andand height
height of box
of the the box
werewere
3 × 23*2*2
× 2 m.m. The COCO sensor
sensorandandtemperature
temperature
andhumidity
and humiditysensor
sensorwere
werefixed
fixedand
andinstalled
installedininthe
thecotton
cottonbox,
box,and
andthethedisplay
displayscreen
screenofof
the industrial computer
the industrial computer was connected to the cab. During the test, the cotton
connected to the cab. During the test, the cotton was was lit lit
in the
in
cotton
the boxbox
cotton of of
thethe
cotton
cottonpicker,
picker,the
thestate
stateofofthe
the cotton
cotton fire
fire was tested during
during the
thewhole
whole
process,the
process, thechanges
changesofof the
the characteristic
characteristic parameters
parameters were
were observed,
observed, andandthe the warning
warning effectef-
fectrecorded.
was was recorded. The system
The system was to
was able able to operate
operate continuously
continuously for a for
longa long
time time and meet
and meet the
operation requirements
the operation of cotton
requirements pickers.
of cotton pickers.

CO
CO concentration
concentration sensor
sensor

On board terminal

Temperature
Voltage module
and humidity
sensor

Figure8.8.Test
Figure Testsite
sitephotos.
photos.

3.3.Results
Results
Algorithm
AlgorithmPerformance
PerformanceEvaluation
Evaluation
Since
Sincethe
thenumerical
numericaldimensions
dimensionsand andranges
rangesofofCO
COconcentration,
concentration,temperature,
temperature,and
and
humidity were all different, in order to make reasonable use of these three characteristic
humidity were all different, in order to make reasonable use of these three characteristic
variables,
variables,normalization
normalizationwaswasperformed.
performed.The
Thenormalization
normalizationformula
formulaisis
Y − −min
Yi Y Y min
Yi ' = i
Yi0 = (4)
(4)
Y max −min
Ymax − Y Y min
where YiYisi Current input data. YminYismin
the smallest value in the data. Ymax Yis the
maxlargest
where is Current
0 input data. is the smallest value in the data. is the
value in the data. Yi is Normalized value.
Yi '
largest value in the data. is Normalized value.
 Sensors2023,
Sensors
Sensors 2023,23,
2023, 23,x5553
23, xFOR
FORPEER
PEERREVIEW
REVIEW 999ofof 13
of13
13

COconcentration
CO
CO concentrationwas
concentration wastaken
was takenas
taken asan
as anexample
an examplein
example inthis
in thispaper
this paperto
paper toevaluate
to evaluatethe
evaluate theperformance
the performance
performance
of the
of
of the algorithm.
the algorithm. The
algorithm. The parameters
parametersof of CO
CO concentration
concentration werewere input
input into three BP
into three
three BP neural
BP neural
neural
networks.Additionally,
networks.
networks. Additionally, one
one ofofthem
of themwas
them wasoptimized
was optimizedby
optimized byIPSO
by IPSO(Improved
IPSO (ImprovedParticle
(Improved ParticleSwarm
Particle Swarm
Swarm
Optimization),and
Optimization),
Optimization), andthe
the other
theother
otherbyby GA
byGAGA (Genetic
(Genetic
(Genetic Algorithm).
Algorithm).
Algorithm). Inorder
In order
In totruly
order
to truly reflect
to truly thedif-
reflect
reflect the dif-
the
difference
ference of of prediction
prediction results
results among
among the the three
three models,
models, the the three
three models
ference of prediction results among the three models, the three models were trained 20models werewere trained
trained 20
20 times, respectively,
times,respectively,
times, respectively,in in
inthis this
thispaper,paper,
paper,and and statistical
andstatistical analysis
statisticalanalysis
analysiswas was
wasperformed performed
performedon onthe on
theNSE the
NSE(NashNSE
(Nash
(Nash
efficiency
efficiency efficiency coefficient)
coefficient)
coefficient) andMAE
and MAEand(mean
MAE absolute
(mean (mean
absoluteabsolute error),
error),and
error), andthetheandRMSE
RMSE the(root
RMSE
(rootmean(rootsquare
mean mean
square
square
error) error)
was was calculated
calculated for eachfor each prediction,
prediction, as shownas
error) was calculated for each prediction, as shown in Figure 9.shown
in in
Figure Figure
9. 9.

(a)
(a) (b)
(b) (c)
(c)
Figure9.
Figure
Figure 9.Performance
9. Performancecomparison
Performance comparisonof
comparison ofdifferent
of differentalgorithms
different algorithmsfor
algorithms for(a)
for (a)NSE
(a) NSE(Nash
NSE (Nashefficiency
(Nash efficiencycoefficient),
efficiency coefficient),
coefficient),
(b)
(b) MAE (mean absolute error), and (c) RMSE (root mean square error).
(b) MAE
MAE (mean
(mean absolute
absolute error),
error), and
and (c)
(c) RMSE (root mean
RMSE (root mean square
square error).
error).

Thegreater
The
The greaterthe
greater thedifference
the differencebetween
difference betweenthe
between theboxes
the boxesin
boxes inthe
in thebox
the boxplot,
box plot,the
plot, thegreater
the greaterthe
greater thedifference
the difference
difference
in
in the
in the evaluation
the evaluation indicators
evaluation indicators
indicators of of the
of the model
the model prediction
model prediction results.
prediction results. It can
results. ItIt can
can bebe seen
be seen from
seen from Figure9a
from Figure
Figure 9a
9a
thatthere
that
that therewas
there wasaaasignificant
was significantoverall
significant overalldifference
overall differencein
difference inthe
in theevaluation
the evaluationindicators
evaluation indicatorsof
indicators ofthe
of theprediction
the prediction
prediction
results
results between
between the
the BP
BP neural
neural network
network model
model and
and the
the
results between the BP neural network model and the GA-BP neural network model. GA-BP
GA-BP neural
neural network
network model.
model. Ad-
Ad-
ditionally,
ditionally,
Additionally, there
there was
waswas
there a significant
a significant
a significant overall
overall difference
difference
overall difference in the
in the evaluation
inevaluation
the evaluation indicators
indicators of the pre-
of theofpre-
indicators the
dictionresults
prediction
diction results between
results
betweenbetweentheIPSO-optimized
the IPSO-optimized
the IPSO-optimized BPneural
BP neural
BP neural network
network networkmodel
model andthe
model
and the GA-BP
andGA-BP
the GA-BPneu-
neu-
ralnetwork
ral network
neural network model.
model. TheGA-BP
model.
The GA-BP
The GA-BP boxin
box in
boxFigure 9b,cwas
in Figure
Figure 9b,c was the
9b,cthe was smallest, andthe
the smallest,
smallest, and the
andGA-BP neural
the GA-BP
GA-BP neural
network
neural
network modelhad
network
model hadthe
model thehad
smallest variability.
the smallest
smallest variability. Thesmaller
variability.
The smaller
The smallertheMAE
the MAE andRMSE,
theand
MAE RMSE,
and RMSE,themore
the more
the
stablestable
more
stable thealgorithm.
the algorithm.
the algorithm.Thevalues
The values ofMAE
The values
of MAE and
ofand
MAE RMSE
and of
RMSE ofthe
RMSE theGA-BP
GA-BP
of the GA-BPneuralnetwork
neural network
neural model
network
model
wereslightly
model
were slightly smallerthan
were slightly
smaller thanthose
smaller thosethan ofthe
of theBP
those BPof neural
the BP
neural network
neuralmodel
network model
network and
and theIPSO-optimized
model
the IPSO-optimized
and the IPSO-
BP neural
optimized network
BP neural model,
network and the
model, NSEand was
the slightly
NSE
BP neural network model, and the NSE was slightly higher than the BP neural was higher
slightly than the
higher BP
than neural
the BP network
neural
network
modeland
network
model and
modelthe IPSO-optimized
the IPSO-optimized
and the IPSO-optimized BP neural
BP neural BP network model. The
neural network
network model. The
model. network modelshowed
The network
network model showed
model
thatGA-BP
showed
that GA-BP neural
thatneural
GA-BP network
neural can
network caneffectively
networkeffectively improvethe
can effectively
improve the stability
improve
stability and
theand accuracy
stability
accuracy andofof predic-
accuracy
predic-
tion.
of
tion.prediction.
The total
Thetotal
The number
totalnumber
numberof of iterations
ofiterations
iterationsof of the
ofthe
theBPBP neural
BPneural network
neuralnetwork
networkmodel model optimized
modeloptimized
optimizedby by the
bythe IPSO
theIPSO
IPSO
particle
particle swarm algorithm,
algorithm, the
the BP
BP neural
neural network
network model
model
particle swarm algorithm, the BP neural network model optimized by the GA genetic al- optimized
optimized by bythetheGA GA genetic
genetic al-
algorithm,
gorithm,and
gorithm, and
andthe the traditional
thetraditional
traditionalBP BP model
BPmodel
modelwere were statistically
werestatistically analyzed,
statisticallyanalyzed,
analyzed,as as shown
asshown
shownin in Figure
inFigure
Figure10.10.
10.

Figure 10.
Figure10.
Figure Comparison
10.Comparison of
Comparisonof iteration
ofiteration times
iterationtimes of
timesof different
ofdifferent algorithms.
differentalgorithms.
algorithms.
Sensors 2023, 23,
Sensors 2023, 23, 5553
x FOR PEER REVIEW 10 of
10 of 13

The number of iterations of the GA-BP neural network model in Figure 10 was less
than that of the BP neural network model and the BP neural network model optimized by
IPSO, indicating that the GA-BP neural network can reduce the number of iterations. The
fewer
fewer iterations,
iterations, the
the greater
greater the
the memory
memory consumption
consumption caused
caused byby the
the experiment
experiment andand the
the
lower
lower calculation cost. Therefore, the GA-BP model was selected in this paper to predict
calculation cost. Therefore, the GA-BP model was selected in this paper to predict
the
the fire
fire situation
situation of
of the
the cotton
cotton picker.
picker. The
The CO
CO concentration
concentration parameters
parameters uploaded
uploaded byby the
the
gas
gas sensor
sensor were
were uploaded
uploaded toto the
the trained
trained GA-BP
GA-BP neural
neural network
network model,
model, and
and the
the output
output
value
value of
of the
the fire
fire prediction
prediction time
time was
was obtained.
obtained. The
The prediction
prediction accuracy
accuracy curve
curve is
is shown
shown in
in
Figure 11.
Figure 11.

Figure 11. GA-BP neural network model prediction.

It can be seen from Figure 11 that

that the
the predicted
predicted curve
curve had a high
high coincidence
coincidence rate
with the actual curve, with an error rate less than 3%, and an accuracy rate of 96.84%. The
GA-BP neural network model can effectively
effectively predict and warn the fire situation, which
greatly increased the safety.
greatly increased the safety.

4. Discussion
4. Discussion
4.1. System Monitoring Effect Comparison
4.1. System Monitoring Effect Comparison
In order to verify the effect of prediction algorithm, the GA-BP algorithm was com-
In order to verify the effect of prediction algorithm, the GA-BP algorithm was com-
pared with the GA algorithm and the IPSO-BP algorithm. The temperature, humidity,
pared with the GA algorithm and the IPSO-BP algorithm. The temperature, humidity, and
and co concentration were monitored 100 times by the sensor, and the average value of
co concentration
the were
actual value was monitored
compared 100the
with times by the
average sensor,
value and
of the the average
predicted value
value. The of the
value
actual value
measured bywas compared
KT-600 with the
gas detector average
(Henan value of
Baoshian theLtd.,
Co., predicted value.China)
Zhengzhou, The value
andmeas-
Xima
ured by KT-600 gas detector (Henan Baoshian Co., Ltd., Zhengzhou, China)
AS847 industrial grade temperature and humidity meter (Dongguan Wanchuang Electronic and Xima
AS847 industrial
Products Co., Ltd.,grade temperature
Dongguan, China) and humidity
was taken meter
as the (Dongguan
real value, Wanchuang
and the value judgedElec-
by
tronic Products Co., Ltd. Dongguan, China) was taken as the real value, and
adding the sensor to the algorithm was selected as the monitoring value. The results arethe value
judged in
shown byTable
adding
1. the sensor to the algorithm was selected as the monitoring value. The
results are shown in Table 1.
Table 1. Comparison of monitoring effects of different algorithms.
Table 1. Comparison of monitoring effects of different algorithms.
Temperature Humidity CO Mean
Algorithm Times Temperature (Average
(Average Humidity Concentration
CO Concen- Relative
Mean Rela-
Algorithm Times Value)
(Average Value)
(Average (Average
trationValue)
(Aver- Error
tive Error
Actual value 100 Value)
32.7 Value)
50.6 age Value)
35.1 -
BP value
Actual 100100 33 32.7 48.350.6 32.9
35.1 3.91%
-
IPSO-BP 100 33.2 44.6 32 7.40%
BP 100 33 48.3 32.9 3.91%
GA-BP 100 32.9 47.7 33.7 3.44%
IPSO-BP 100 33.2 44.6 32 7.40%
Table 1 shows that in the comparison process of the three algorithms, by calculating the temperature.
GA-BP 100 32.9 47.7 33.7 3.44%
TableThe average
1 shows that inerror was obtained
the comparison byofcomparing
process the humidity
the three algorithms, and COthe
by calculating concentration
temperature.
error comparison, and the error used by the system was 3.44%. The fire monitoring and
early warning algorithm of the cotton picker based on GA-BP designed in this study
Sensors 2023, 23, 5553 11 of 13

was more accurate than the other two algorithms. The reason for this was that in the
algorithm, the GA genetic algorithm was used to replace the traditional method of randomly
generating weights and thresholds, and the adjustment of the BP neural network error
back propagation was used to optimize the weights and thresholds gradients to reduce
errors, avoid falling into the local minimum value, obtain the optimal network structure,
and make the model more robust. The error curve of the predicted value of the algorithm
in this paper was in good agreement with the experimental measurement value, and the
error was small, which also reflected the systematic and scientific nature of cotton picker
operation safety. The GA-BP algorithm was adopted in the early warning and monitoring
system of the cotton picker to realize the fire early warning model of the cotton picker, so
as to obtain better fire warning effects.

4.2. System Alarm Performance Comparison

The performance of early warning system is an important indicator to measure the fire
early warning effect of the cotton picker. The fire alarm performance will affect the safety
of the cotton picker to a large extent. Therefore, in order to enhance the interpretability of
the experimental results, the effectiveness of the algorithm can be effectively highlighted
by taking the fire early warning results as the index in this paper. BP and IPSO-BP algo-
rithms were compared and the performance verification was completed together with the
algorithm in this paper, as shown in Table 2.

Table 2. Comparison of fire warning algorithms.

Algorithm Test Number Accurate Recognition Times/n Accuracy

BP 700 641 91.57%
IPSO-BP 700 652 93.14%
GA-BP 700 678 96.86%

It can be seen from the table that the fire warning algorithm based on the BP neural
network had a recognition accuracy of 91.57% in 700 fire recognitions. The BP algorithm fire
warning based on IPSO optimization had an accurate recognition rate of 93.14% in 700 fire
identifications. The BP fire warning algorithm based on GA optimization had an accurate
recognition rate of 96.86% in 700 fire identifications. Through comparison, it can be seen
that the accuracy of GA-BP algorithm was higher than that of BP neural network algorithm
and IPSO-BP algorithm. The GA-BP algorithm was superior to the other two methods.

5. Conclusions
Taking the CO concentration value, temperature, and humidity of the cotton box of
the cotton picker as the fire monitoring object, the gas concentration and temperature and
humidity changes in the cotton box of the cotton picker were monitored in real time by using
the CO sensor and temperature and humidity sensor. Combined with the GA-optimized BP
neural network intelligent algorithm, the accurate monitoring system of CO concentration
in cotton box realized the functions of gas concentration data collection and storage, data
real-time transmission of RS485, real-time display of vehicle-mounted industrial computer
and alarming when data exceeded the threshold. The system was convenient, simple, and
clearly displayed, and had a good human–computer interaction interface.
In view of the problems of high monitoring sensitivity during cotton pickers fires,
sensor damage caused by long-term extrusion, and low accuracy of data in cotton box
due to interference of gas sensor by field environment, high-precision gas detection sensor
probes, and metal shell sealed sensor were adopted, and the data measured by the gas
sensor were optimized by the BP network model optimized by the GA genetic algorithm,
which effectively improved the data accuracy. The CO gas concentration was used as model
input for the experiment. The experimental results showed that the monitoring system
worked well and realized the function of system design. At the same time, after BP neural
Sensors 2023, 23, 5553 12 of 13

network model was optimized by the GA genetic algorithm, the measurement accuracy of
CO was not less than 91.23%, which can effectively improve the monitoring accuracy of
early warning of cotton box fire. In order to ensure reliable data transmission, the RS485
transmission mode was adopted in the system.

6. Limitations and Future Research

Because the existing cotton pickers do not consider fire protection, there is no sensor
installed in the cotton picker. There is little research on fire prediction systems, and so, the
fire protection system of cotton picker is very necessary. The sensors used in the study
were not fully functional and in the future, data transmission can be carried out through
mature wireless equipment. At the same time, fire-fighting and rain-spraying devices can
be installed for fire monitoring. It can be seen from the test and detection that the system
has the potential to be applied to a cotton picker fire warning, and the system will be further
improved in the future.

Author Contributions: Conceptualization, W.Z. and B.Z.; methodology, W.Z. and Y.Z.; writing—
original draft, W.Z. and L.Z.; funding acquisition, B.Z.; project administration, W.Z. and B.Z.; formal
analysis, W.Z. and S.G.; investigation, B.Z.; data curation and software. visualization, W.Z. and S.L.
All authors have read and agreed to the published version of the manuscript.
Funding: The work was sponsored by the National Key R&D Program Project of China (2020YFB1709603).
Institutional Review Board Statement: Not applicable.
Informed Consent Statement: Not applicable.
Data Availability Statement: Not applicable.
Acknowledgments: The authors thank the editor and anonymous reviewers for providing helpful
suggestions for improving the quality of this manuscript.
Conflicts of Interest: The authors declare no conflict of interest.

References
1. Caicong, W.; Ying, C.; Weizhong, Y.; Lili, Y.; Peng, Q.; Qin, M.; Weixin, Z.; Dong, L.; Xiaoqiang, Z.; Chuanfeng, W.; et al.
Construction of big data system of agricultural machinery based on BeiDou. Trans. Chin. Soc. Agric. Eng. 2022, 38, 1–8.
2. Wang, Z.-S.; Liu, W.-F.; Ni, Z.-P.; Wang, L.; Gao, B. An Experimental Investigation on the Effect of Wind Speed on Cotton
Combustion. Procedia Eng. 2018, 211, 788–793. [CrossRef]
3. Wang, S.; Xiao, X.; Deng, T.; Chen, A.; Zhu, M. A Sauter mean diameter sensor for fire smoke detection. Sens. Actuators B Chem.
2019, 281, 920–932. [CrossRef]
4. Cheng, R.; Chai, L.; Hu, S. Research on application of artificial intelligence technology in fire alarm system of substations. E3S
Web Conf. 2021, 257, 1–4. [CrossRef]
5. Sakthivel, M.; Weppner, W. A portable limiting current solid-state electrochemical diffusion hole type hydrogen sensor device for
biomass fuel reactors: Engineering aspect. Int. J. Hydrogen Energy 2007, 33, 905–911. [CrossRef]
6. Park, C.-O.; Akbar, S.; Hwang, J. Selective gas detection with catalytic filter. Mater. Chem. Phys. 2002, 75, 56–60. [CrossRef]
7. Rong, C.L.; Hua, N.T.; Rong, C.L. Research on Fire Alarm Monitoring System of Fire Engineering Based on Network. J. Phys.
Conf. Ser. 2021, 1792, 1–6.
8. Hong, K.H.; Oh, K.W.; Kang, T.J. Polyaniline–nylon 6 composite fabric for ammonia gas sensor. J. Appl. Polym. Sci. 2004, 92, 37–42.
[CrossRef]
9. Martoiu, S.; Muller, H.; Tarazona, A.; Toledo, J. Development of the scalable readout system for micro-pattern gas detectors and
other applications. J. Instrum. 2013, 8, 1–10. [CrossRef]
10. Massie, C.; Stewart, G.; McGregor, G.; Gilchrist, J.R. Design of a portable optical sensor for methane gas detection. Sens. Actuators
B Chem. 2006, 113, 830–836. [CrossRef]
11. Stetter, J.R.; Li, J. Amperometric gas sensors a review. Chem. Rev. 2008, 108, 352–366. [CrossRef] [PubMed]
12. Anonymous. Neural Network Numerosity. Am. Sci. 2019, 107, 207.
13. Li, G.; Deng, L.; Chua, Y.; Li, P.; Neftci, E.O.; Li, H. Editorial: Spiking Neural Network Learning, Benchmarking, Programming
and Executing. Front. Neurosci. 2020, 14, 276. [CrossRef] [PubMed]
14. Jain, A.K.; Goel, D.; Agarwal, S.; Singh, Y.; Bajaj, G. Predicting Spam Messages Using Back Propagation Neural Network. Wirel.
Pers. Commun. 2020, 110, 403–422. [CrossRef]
Sensors 2023, 23, 5553 13 of 13

15. Chen, S.; Ren, J.; Yan, Y.; Sun, M.; Hu, F.; Zhao, H. Multi-sourced sensing and support vector machine classification for effective
detection of fire hazard in early stage. Comput. Electr. Eng. 2022, 101, 108046. [CrossRef]
16. Sulistian, G.; Abdurohman, M.; Putrada, A.G. Ieee, Comparison of Classification Algorithms to Improve Smart Fire Alarm System
Performance. In Proceedings of the 2019 4th International Workshop on Big Data and Information Security (IWBIS 2019), Bali,
Indonesia, 11 October 2019; pp. 119–124.
17. Yuan, F.; Yu, L.; Yan, Y. Multi-Sensor GA-BP Algorithm Based Gearbox Fault Diagnosis. Appl. Sci. 2022, 12, 3106. [CrossRef]
18. Zheng, F.; Zeng, L.; Lu, Y.; Kai, G.; Fu, S. Fault Diagnosis Research for Servo Valve Based on GA-BP Neural Network. J. Comput.
Theor. Nanosci. 2015, 12, 2846–2850. [CrossRef]
19. Wen, Y.; Jia, M.; Luo, C. Study of Fault Diagnosis for Rolling Bearing Based on GA-BP Algorithm. In Proceedings of the 2017 2nd
International Conference on Automation, Mechanical Control and Computational Engineering (AMCCE 2017), Beijing, China,
25–26 March 2017.
20. Tang, J.; Li, W.X.; Zhao, B. The Application of GA-BP Algorithm in Prediction of Tool Wear State. IOP Conf. Ser. Mater. Sci. Eng.
2018, 398, 93–97. [CrossRef]
21. Du, J.; Dong, Y.-C.; Xia, J.; Li, H. Research on the Equipment Fault Diagnosis Based on GN-BP Neural Network. In Proceedings
of the 2017 2nd International Conference on Automation, Mechanical Control and Computational Engineering (AMCCE 2017),
Beijing, China, 25–26 March 2017.

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual
author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to
people or property resulting from any ideas, methods, instructions or products referred to in the content.