Content uploaded by Rui Liang
Author content
All content in this area was uploaded by Rui Liang on Aug 16, 2023
Content may be subject to copyright.
1
5
10
15
20
25
30
35
40
45
50
55
1
5
10
15
20
25
30
35
40
45
50
55
1
5
10
15
20
25
30
35
40
45
50
55
1
5
10
15
20
25
30
35
40
45
50
55
State-of-the-art applications of machine learning in the life
cycle of solid waste management
Rui Liang1, Chao Chen1, Akash Kumar1, Junyu Tao (✉)2, Yan Kang2, Dong Han2, Xianjia Jiang2, Pei Tang2,
Beibei Yan1,3, Guanyi Chen2,4
1SchoolofEnvironmentalScienceandEngineering,TianjinUniversity,Tianjin300350,China
2SchoolofMechanicalEngineering,TianjinUniversityofCommerce,Tianjin300134,China
3TianjinKeyLaboratoryofBiomassWastesUtilization/TianjinEngineeringResearchCenterofBioGas/OilTechnology,Tianjin300072,China
4SchoolofScience,TibetUniversity,Lhasa850012,China
HIGHLIGHTS GRAPHICABSTRACT
●State-of-the-art applications of machine
learning(ML)insolidwaste(SW)ispresented.
●Changesofresearchfieldovertime,space,and
hottopicswereanalyzed.
●Detailedapplication seniors of ML on the life
cycleofSWweresummarized.
●Perspectives towards future development of
MLinthefieldofSWwerediscussed.
ARTICLEINFO
Article history:
Received28May2022
Revised18August2022
Accepted6September2022
Availableonline20October2022
Keywords:
Machinelearning(ML)
Solidwaste(SW)
Bibliometrics
SWmanagement
Energyutilization
Lifecycle
ABSTRACT
Duetothesuperiorityofmachinelearning(ML)dataprocessing,itiswidelyusedinresearchofsolid
waste(SW). Thisstudyanalyzed theresearchand developmentalprogressof theapplications ofML
inthelifecycleofSW.Statistical analyseswereundertakenontheliteraturepublishedbetween1985
and2021intheScienceCitationIndex Expanded and Social Sciences Citation Index to provide an
overviewof the progress.Basedon the articlesconsidered,a rapid upwardtrendfrom 1985 to2021
wasfoundandinternational cooperatives were found to havestrengthened.Thethree topics of ML,
namely,SW categories, MLalgorithms,and specific applications,asapplied to thelifecycle of SW
werediscussed.MLhasbeenappliedduringtheentireSWprocess,therebyaffectingitslifecycle.
ML was used to predict the generation and characteristics of SW, optimize its collection and
transportation, and model the processing of its energy utilization. Finally, the current challenges of
applyingMLtoSWandfutureperspectiveswerediscussed.Thegoalistoachievehigheconomicand
environmentalbenefitsandcarbonreductionduringthe lifecycleofSW.MLplaysanimportantrole
inthe modernization andintellectualizationof SWmanagement.It is hopedthatthis workwouldbe
helpfultoprovideaconstructiveoverviewtowardsthestate-of-the-artdevelopmentofSWdisposal.
©HigherEducationPress2023
1Introduction
Solid waste (SW) disposal is an important international
issue in terms of resources and environmental aspects.
With respect to resources, the materials and energy
recovered from SW could contribute to the sustainable
✉Correspondingauthor
E-mail:taojunyu@tjcu.edu.cn
Special Issue—Artificial Intelligence/Machine Learning on Environ-
mentalScience&Engineering(ResponsibleEditors:YongshengChen,
XiaonanWang,JoeF.BozemanIII&ShouliangYi)
Front.Environ.Sci.Eng.2023,17(4):44
https://doi.org/10.1007/s11783-023-1644-x
RESEARCHARTICLE
FSE-22087-LR2022-09-28 17:16:24
1
5
10
15
20
25
30
35
40
45
50
55
1
5
10
15
20
25
30
35
40
45
50
55
andcarbon-neutraldevelopmentofhumansocieties.The
energy utilization of SW has a significant impact on
energyshortage(Zhang etal.,2021).Withrespect to the
environment, mass accumulation and improper disposal
of SW threaten the atmosphere, water, ecosystems, and
human health via the food chain (Liu et al., 2021; Shi
etal.,2022).Itsimportancehas led to growing concerns
worldwide. To enhance the overall performance of SW
recycling and disposal, it is necessary to take action at
each point during the life cycle of SW, which can be
divided into: 1) generation and characterization, 2)
collection and transportation, and 3) treatment and
utilization. Recently, processes related to these three
divisions have experienced an increasing demand for
data-driven techniques. Policymakers urgently need
instructiveguidancefromtherunningdataofthechainof
SWdisposaltoreducetheeconomiccostandobtainmore
environmentalbenefits.In contrast, citizens and workers
intheSWindustryrequiremoreconvenientandefficient
data-driven techniques, particularly machine learning
(ML).
Currently,MListhemostpopulardata-driventechno-
logy.Itsimulates orrecognizeshumanlearningbehavior
using computers. Due to of its time-saving, strong
learningabilityandhighaccuracy,MLhasbeenapplied
in many fields, such as population analysis (Wan et al.,
2009), soil conservation (Qu et al., 2018), and environ-
mentalscience(ErkinayOzdemir et al., 2021). Inrecent
years,MLhasalsobeenusedforatvariousaspectsofthe
SWlifecycle.For example, Taoetal.(Taoetal.,2020)
studied the elemental composition (C, H, O) and lower
heating value of SW and predicted these characteristics
accordingtoinfraredspectrawithanaccuracyofover85%.
Coskuneretal.(2021)proposedrobustpredictivemodels
for annual forest generation rates of domestic, comm-
ercial,construction, and demolition wastes;R2 values of
0.95, 0.99, and 0.91, respectively were obtained. Chen
et al. (2021a) established a model based on an artificial
neural network and decision tree to determine online
measurement and control of polychlorinated dibenzo-
dioxin and dibenzofuran emissions. Idwan et al. (2020)
developeda two-step heuristicalgorithm to discoverthe
idealtruckrouteforwastefleetmanagement.Zhengetal.
(Zheng and Gu, 2021) proposed an ensemble image-
learningmodelbasedonconvolutionalneuralnetworksto
classifydomesticSW.
These attempts to use ML in multiple ways for SW
disposalprovide a promising scope for the development
of this field. Meanwhile, a timely overview of relevant
achievementsis urgently neededto help researchersand
policymakers in this field find potential directions.
Severalreviewshavebeenconductedinrecentyearsthat
focusonMLapplicationsinthefieldofSWdisposal.For
example, Wang et al. (Wang et al., 2022) analyzed the
role of ML in the development of bioenergy and
conversionofbiofuel.Guoetal. (2021) summarized the
application of ML to predict organic SW treatment and
recyclingprocesses.Althoughthesestudiesareinsightful
andinformative,itisdifficultforthereadershiptoobtain
comprehensiveknowledgeoftheentirelifecycleofSW
becauseitcontainstoomanydivisions,makingitdifficult
to have an objective discussion based on the existing
literature.
Accordingly,aquantitativeassessmentofthepublished
literature is necessary to provide an overview of ML
appliedduringthelifecycleofSW,asthiswillbehelpful
for policymakers and researchers. In addition, quantita-
tiveassessmentofthisfieldisrare,andthisstudyusesthe
bibliometric method to facilitate a review on the use of
MLduringthelifecycleofSW.Thereafter,basedonthe
bibliometric analysis, deep perspectives on the research
progress and future development were provided in this
study.Bibliometrics is a literaturemanagement tool that
can use mathematical methods to quantitatively analyze
documentsin acertainfield.Ithasbeenapplied inmany
fields to explore technological progress, such as energy
development (Obileke et al., 2020), business economics
(Ding and Yang, 2020), and environmental science
(de Sousa, 2021). However, it has rarely been used to
analyzetheapplicationofMLinSWmanagement.
In this study, the basic characteristics, including the
total publication, total citations of articles, countries,
institutions,keywords,anddistributionofresearchareas,
wereinvestigatedusingthebibliometricmethodtoobtain
anoverviewofthedevelopmenttrendoverthepastyears.
Based on a quantitative and comprehensive analysis,
state-of-the-art advances and future perspectives of ML
ondifferentareasofSWdisposalwerealsoexplored.We
hope that this work will significantly contribute to the
environmentalandresourceconcernsassociatedwithSW
disposal.
2Data and methods
2.1Bibliometrics
Bibliometrics is a science that adopts mathematics,
statistics, and other measurement methods to study the
literature system and bibliometric characteristics (Bhatt
etal., 2020),suchasdistributionstructure,changeregul-
ation, and quantitative relationship. Key information,
such as the title, author, publication period, keywords,
and citation information, obtained from literature are
analyzedusingthismethod.Bibliometricscanbeusedto
explore specific aspects of science and technology and
havebecomea general method formeasuring regulation
(Keramatfar and Amirkhani, 2019). Bibliometrics pro-
vides a useful tool to reflect scenarios and trends in a
researchfield.Thisstudyconcludeswiththequantityand
visualprocessusedtoidentifypatternsandcharacteristics
ofchangesinscientificliterature.
2 Front.Environ.Sci.Eng.2023,17(4):44
FSE-22087-LR2022-09-28 17:16:24
1
5
10
15
20
25
30
35
40
45
50
55
1
5
10
15
20
25
30
35
40
45
50
55
2.2VOSviewer
VOSviewerisasoftwaretoolforcreatingmapsbasedon
networkdataandvisualizingandexploringthesemaps.It
iscommonlyusedforbibliometric analyses. VOSviewer
is based on sophisticated clustering and analysis
technology that analyze knowledge units in scientific
literature. Its biggest advantage is that it visualizes
knowledge units and has good image presentation
capabilities. A co-occurrence matrix and distance-based
groupingwere used to build maps in VOSviewer(Lima
et al., 2021). This software can perform co-authorship,
co-occurrence, co-citation, citation, and bibliographic
analyses by exploring the density between documents.
VOSviewer includes three browsing formats: network,
overlay, and density formats. Various browsing formats
reveal the flow and transfer of knowledge between
scientific documents based on color, size, width, and
other factors. This also reflects the similarity,
consistency,and citation relationshipsbetween scientific
documents.
2.3Literatureanalysismethods
Thestepsof this study were illustratedinFig. 1(a). The
firststep was to obtainliterature on MLapplied to SW,
forwhichtheWebofScience(WoS)corecollectionwas
used. WoS is a web-based product developed by
ThomsonScientificintheUnitedStates.Thisisalarge-
scale, comprehensive, and multidisciplinary journal
citation index database. This study selected two major
citation indices in the WoS core collection: Science
Citation Index Expanded (SCI-EXPANDED) and Social
Sciences Citation Index (SSCI). This study was
conductedaccordingtotopicandtitle.Thetopicsearchin
WoScoveredeacharticle’s title, abstract,andkeywords,
encompassingthemaximumamountofinformationinthe
database.Toobtainpreciseresults,atitlesearchwasalso
used. After repeated deliberation (Guo et al., 2021),
specificsearchformulaandkeywordsin this study were
selected and shown in Fig. 1(a). The search formula
stipulates the scope of the literature search and ensures
that the obtained results conform to the scope of
application of ML on SW. These three terms of search
formulawerelinkedbyBoolean“and”logic,considering
articles on this subject published between January 1985
and June 2021. The search was continued until June 6,
2021.
After the preliminary search, the search results were
refined to include journal articles, which were the
researchobjectsofthisarticle.Next,theserefinedresults
were exported in the format of “Full Record and Cited
References” from WoS. The final exported information
was analyzed using data processing, VOSviewer, and
summaries.
3Bibliometric analysis
3.1Generalintroduction
A total of 170 documents from 1985 to 2021 related to
ML applied to SW were retrieved from the SCI-
EXPANDED and SSCI databases. Of these, journal
Fig. 1Researchrouteofthearticle(a)andprogressofauthorkeywordanalysisinVOSviewer(b).
RuiLiangetal.Applicationsofmachinelearninginthelifecycleofsolidwaste 3
FSE-22087-LR2022-09-28 17:16:24
1
5
10
15
20
25
30
35
40
45
50
55
1
5
10
15
20
25
30
35
40
45
50
55
articles accounted for 88.82 % (151). Other types of
documents included reviews, early access, and editorial
materials. Only journal articles were considered in this
study.
Inviewofjournalarticlespublishedyearly(Fig.2),the
development of the application of ML to SW could be
clearlyidentified.As showninFig.2,journal articleson
the subject areas of this study were first published in
2001. The total number of publications (TP) showed a
rapidupwardtrendover time, although there were some
fluctuationsthroughoutthe studied period.From2001to
2010, research on ML applied to SW was still in its
infancyworldwide;theglobalTPdidnotexceed3.After
2010,theTPincreasedsignificantly,reaching29in2020,
an increase of nearly nine times compared to the TP in
2010.After 2010,MLandartificialintelligencebeganto
emerge.Thegrowthofthisfieldcanbeattributedtothe
required advanced data processing technologies and the
development of environmentally friendly concepts, and
many researchers began to apply ML on SW. For
example, many studies have highlighted the importance
ofartificialintelligence.Thisfieldisnowaresearchhot
spotandmayhavebetterprospectsinthefuture.
Changeinthe total number oftimesliterature cited in
the WoS (TC) was similar to that in the TP. The TC
showedanoverallincreasingtrend.However,therewasa
suddendeclinein 2007 and2014.In2007,theTC hada
minimum of 4. TC reached its maximum in 2018, with
354 citations, and showed a downward trend thereafter.
This may be because citations require long-term
accumulation.Thehighcitationnumberalsoprovedthat
MLhasgreatsignificanceinthedevelopment oftheSW
lifecycle.
3.2Ananalysisofcountries
From2001to2021,46countrieshavepublishedarticles
on ML applied to SW. Of the 151 journal articles
investigatedinthispaper,43(28.48%)wereinvolvedin
internationalcooperation,andtheremaining108onlyhad
participationfromindividual countries. Compared tothe
international cooperation (18.01 %) of waste-to-energy
incinerationin 2015(Wangetal., 2016),thecooperation
between international countries has greatly deepened.
Againstthebackgroundofglobalization,itisconceivable
that international cooperation will be stronger in the
futureinthecontextofnationalcooperation.
Fig.3(a)showedacomparisonofthecountrieswiththe
top eight TPs from 2001 to 2021. Of the eight most
productive countries, five were from Asia (Peoples R
China,Iran,Turkey,India,andMalaysia),twowerefrom
NorthAmerica(USA,Canada),andonewasfromEurope
(England). Peoples R China was the most productive
country,with 37 journal articles (18.41 %), followedby
Iran (26), and Turkey (26). Peoples R China and Iran
have remarkably more articles than other countries,
resulting in a significant research gap between other
countries.
In terms of the number of international collaborative
publications (CP), the USA and China had the best
performance compared to other countries. However, the
CPoftheUSAhadsignificantlyexceededthenumberof
single-countrypublications (SP).ExcepttheUSA,outof
theseeightcountries,onlyEngland hadahigherCPthan
SP. Furthermore, articles published in individual
countries were more common. This may be because
researchonMLappliedtoSWwasstillinitsinfancyand
domestic exploration was still in its initial stages.
National cooperation deepened after the initial develop-
ment. Peoples R China had the highest SP (20),
accountingfor18.52 %. Tran wasclose behind, ranking
second.FortheH-index,whichhas an important impact
onwastemanagementanddisposal,Chinarankedsecond
(256). Therefore, based on the comprehensive perfor-
mances of TP, SP, CP, and FP, China has an important
role in the field of ML applied to SW. However,
improvements in scientific technology and further
strengthininternationalcooperationarerequired.
Fig.3(b)showedtheTPchangesinthecountrieswith
thetopfiveTPsfrom2001to2021.Thefirststudyinthis
field appeared in USA in 2001. The first article from
Chinawaspublishedin2002.However,thisfielddidnot
show great development until 2014. In addition, the
developmentofMLappliedtoSWinthesefivecountries
fluctuatedbetween2001and2021andhasasimilartrend
to TP changes, as discussed under section 3.1. Prior to
2014, research in this field had appeared almost
exclusively in one country. After 2014, the number of
articles published in this field fluctuated. This may be
because of the higher requirements for sophisticated
equipment worldwide, which has led to technological
progress.
The co-authoring relationships of these 46 countries
were also investigated using VOSviewer. The network
was shown in Fig. S1. This showed that China and the
USA played a key role in international cooperation.
Moreover, these two countries had the closest connec-
tions.
Fig. 2TPandTCchangesfrom2001to2021.
4 Front.Environ.Sci.Eng.2023,17(4):44
FSE-22087-LR2022-09-28 17:16:24
1
5
10
15
20
25
30
35
40
45
50
55
1
5
10
15
20
25
30
35
40
45
50
55
3.3Ananalysisofinstitutions
Amongthe151 articlesonMLappliedtoSW, 212were
included. A total of 86 journal articles were completed
with the cooperation of multiple organizations, accoun-
ting for 56.9 5%. This was much higher than the
proportionofcooperation(28.48%)betweencountriesin
discussed under section 3.3, indicating that interchange
within one country was much more common than that
acrossmultiplecountries.Thus,internationalcooperation
must be strengthened further. More communication
betweencountriesmaybring vitality to the development
ofthisnewfield.
The top eight productive institutions were shown in
Fig. 4. University of Tehran in Iran had the largest TP
(12,4.08 %), which wasmuch higher thanthat of other
organizations.The TP insubsequent institutions showed
littledifference,allinsingledigits.Theseeightorganiza-
tions were distributed across Iran, Malaysia, Canada,
Turkey, and Singapore. In particular, Iran, Malaysia,
Canada,Turkey,andSingaporeranked2,7,5,3,and13,
respectively,according to theirTP from 2001–2021. No
Fig. 3Comparisons of the top 8 productive countries (a) and TP changes by year in the top 5 productive countries (b) during
2001–2021.
RuiLiangetal.Applicationsofmachinelearninginthelifecycleofsolidwaste 5
FSE-22087-LR2022-09-28 17:16:24
1
5
10
15
20
25
30
35
40
45
50
55
1
5
10
15
20
25
30
35
40
45
50
55
Chineseinstitutionsappearedin the top eight productive
institutions,althoughChinahadthelargestTP. Fromthe
calculatedstatisticalresults,itcanbeseen thattheTPof
Chinese institutions was 3 or below, but the institutions
had TP that fell within a wide range. This may be the
reasonwhyChinahadthehighestyieldinthefieldofML
appliedtoSW.
Theco-authoring relationship ofthese 212 institutions
wasalsoinvestigatedusingVOSviewer.Thenetworkwas
showninFig. S2. Ton Duc ThangUniversity(Vietnam)
and the University of Tehran play key roles in interna-
tionalcooperation.
3.4Ananalysisofkeywords
Thekeywordswerethegistofanarticleandcouldreflect
themaintopicoftherelevantresearcharea.Thiscanhelp
researchers better understand the emerging trends in a
field. In WoS, keywords contain two parts: author
keywordsand keyword plus.The author keywords were
those we commonly referred. Keyword plus was a
keyword added by WoS to increase the hit rate of the
articleunderarelevanttopic. Therefore, author keyword
analysis was adopted using VOSviewer to acquire the
hotspotsaswellasfuturetrendofMLappliedtoSW.
Fig. 1(b) present author keyword analysis steps. First,
Openrefine was used to sort out the duplicate author
keywords in the search record. If synonymous author
keywordswithdifferentspellingsweredirectlyenteredin
VOSviewerforanalysis, theaccuracyofauthorkeyword
occurrence would be affected. A synonym dictionary,
thesaurus_terms.txt file, was created in this step. It was
thenimportedtoVOSviewertoperformauthorkeyword
analysis. In VOSviewer analysis, minimum number of
occurrencesofakeywordwassetasfour.
Atotalof443authorkeywordswereidentifiedin151
journal articles. Fig. 5 showed the results of author
keyword co-occurrence in VOSviewer. The nodes
representdifferentauthorkeywords.The largerthenode,
the more occurrences of author keywords in the 151
journal articles. The lines between the nodes reflect the
relationships between the author keywords. If the line
betweentwonodeswasthickerorshorter,therelationship
betweenthetwoauthor keywords wasstrengthened.The
different node colors represent the different types of
clusters. Each cluster was determined based on its
weight/significance (Lima et al., 2021). This cluster
algorithmofVOSviewerissimilartoaprobability-based
measurement method. The more similar the two author
keywords,themorerelatedtheyare.
The central author keyword of this network was
“municipalsolidwaste”.Thenodesaroundthecentercan
be divided into two types: 1) keywords related to
algorithms, such as artificial neural networks, genetic
algorithms,andsupportvectormachineand2)keywords
related to applications, such as waste management and
waste-to-energy.Inaddition,“municipalsolidwaste”and
“artificialneuralnetwork”appearedmostfrequently.The
link strengths of these two keywords were also the
strongest.
Fig. 4TopeightproductiveinstitutionsforMLappliedtoSW,during2001–2021.
6 Front.Environ.Sci.Eng.2023,17(4):44
FSE-22087-LR2022-09-28 17:16:24
1
5
10
15
20
25
30
35
40
45
50
55
1
5
10
15
20
25
30
35
40
45
50
55
Therewerefourclustersofdifferentcolors,asshownin
Fig.5, which alsoshowed the main topics of these four
clusters. The top three keywords with the highest
occurrences in these four clusters were summarized in
TableS2. Twelve keywords withhigh occurrences were
obtained and can be divided into three categories. 1)
Municipalsolidwaste(withthehighestoccurrenceoutof
allkeywords)andsolidwaste,whichwererelatedtoSW
itself; 2) artificial neural network, support vector
machine,geneticalgorithm,machinelearning,andneural
network,whichwererelatedtotheMLalgorithm;and3)
optimization, forecasting, recycling, waste management,
and life cycle assessment, which focus on different
aspectsofthelifecycleofSW.Accordingly,researchon
ML applied to the life cycle of SW can be divided into
three parts: SW categories, ML algorithms, and specific
applications. These three aspects were discussed in
furthersections.
WeanalyzedtheliteratureonMLappliedtoSWusing
thebibliometricmethodsinprovidedundersections3.1-
3.4.Thecharacteristicsoftherelatedliteraturefrom1985
to 2021, based on the SCI-EXPANDED and SSCI
databases,were examined. This provideda development
venation for the application of ML in SW management.
Thefirst literatureonthisresearchareawaspublishedin
2001.TheTPandTCofthearticlesshowedrapidupward
trends,withtheexceptionofafewfluctuations.
Publications on ML applied in the lifecycle of SW in
countriesandinstitutionsworldwidewerealsoexamined.
This provided information on the international
cooperativerelations and their respective research levels
on ML applied to SW. People R China was the most
productivecountryandplayedakeyrole ininternational
cooperation.Morethanaquarterofthearticles involved
international cooperation. The TP between institutions
showedafewdifferences, all in single digits,exceptfor
University of Tehran, which had a TP of 12. The
cooperationofinstitutionswithinone country was much
more common than in multiple countries. The top 5
productive journals in the publication of ML applied in
SW during 2001–2021 were provide in Table S1.
Keywordsreflected themainfocusofthisresearchstudy
and can be divided into SW categories, ML algorithms,
and specific applications. Based on this bibliometric
analysis,thedevelopmentvenationofMLappliedinSW
in terms of time, international state, and research topics
wereanalyzed.This provided a clearand overall insight
to researchers and policymakers in this area. It was
necessary to summarize the different aspects of ML
application to the lifecycle of SW. Moreover, it can
significantlycontributetothefutureprospectsofMLand
SW.
4Literature review
According to the bibliometric analysis of the keywords,
as mentioned under section 3.4, ML applied to the life
cycle of SW can be divided into three parts: SW
categories, ML algorithms, and specific applications.
Fig. 5Authorkeywordsnetworkof151journalarticlesinVOSviewer.
RuiLiangetal.Applicationsofmachinelearninginthelifecycleofsolidwaste 7
FSE-22087-LR2022-09-28 17:16:24
1
5
10
15
20
25
30
35
40
45
50
55
1
5
10
15
20
25
30
35
40
45
50
55
Based on the literature and bibliometric analysis, a
detailed introduction, and an in-depth analysis of these
threepartswerecarriedout.
SWwastherawmaterialinvestigatedinthisstudy.SW
refers to the production, consumption, living, and other
activitiesofsolidandsemisolidwastematerialsproduced
by human beings. It consists of a wide range of
substances and can be divided into industrial SW,
municipal SW (MSW), construction SW, agricultural
SW, and hazardous SW. MSW was highly prevalent in
the151journalarticles(Fig.5).ItistheSWgeneratedin
the daily life of a city or the activities that provide
servicestothecity.Withtheincreaseinhumanprogress,
urbanization has advanced. Thus, human activities have
becomemoreabundant.Consequently,theproductionof
MSW has become aggravated (Noori et al., 2009), and
thenumberofscientificstudiesonMSWhasincreased.
MLwasa method used to studythelifecycle of SW.
ML is a multidisciplinary subject that specializes in
simulating human learning behaviors in computers. It
containsvarioustypesofalgorithms.Intheliterature,we
found that artificial neural network (ANN), support
vectormachine(SVM), random forest (RF),and genetic
algorithm (GA) were the most used ML algorithms
applied to the lifecycle of SW. In particular, GA is an
optimization algorithm that can optimize the parameters
ofotheralgorithmssuchasANN,SVM,andRF.
ML has been applied to the entire SW process,
affecting its life cycle. According to the 151 journal
articlesobtainedinthisstudy,thespecificapplicationsof
ML on SW can be divided into three areas. ML can be
usedto1)predictthegenerationcharacteristicsofSW,2)
collectandtransportSW,and3)treatandutilizeSW.The
status of ML was discussed in detail in the subsequent
threesections.
4.1Generationandcharacterization
During the life cycle of SW, it is important to first
consider its properties. The generation of SW and its
changes reflect the development of a country. The
generation of SW is related to economic, social,
population, and other factors. In addition, SW
characteristics are greatly important for its utilization,
such as elemental composition, heating value, and ash
content. This influenced the treatment method or
conditionoftheSWdisposal.MLcanbeusedtopredict
SWgenerationandcharacteristicsthroughdatalearning.
QuantitativepredictionofSWproductioniscriticalfor
thedesign andplanningofSW managementsystems.As
mentioned above, SW is produced by human beings
through their social activities and is difficult to
understand. Thus, the quantity of SW is influenced by
various uncertainty parameters related to human beings.
Additionally, the amount of SW is highly volatile.
Therefore, it is difficult to precisely predict SW
generationusing conventional statistical methods. Based
on the robust properties of data processing, ML can be
usedto learnpreviousdataon wasteproduction,monitor
itsgenerationandpredicttheevolutionregulation.SVM,
ANN, and RF were typically used in research.
Furthermore, prediction models were improved in
subsequentstudies.They canreflectmoreinformationin
additiontoSWgenerationinSWmanagement.Daietal.
(2011) developed a support vector regression model to
predict the amount of waste generated in a city. It also
reflects the dynamic, interactive, and uncertain
characteristics of SW management systems. These SW
generation prediction models were helpful for SW
managementinMSWplants.
As mentioned above, waste generation is a complex
process closely correlated with human social life. It is
influencedby some uncertainfactors and thepopulation
andeconomic aggregates of a city, including population
size, city size, GDP, income level, capita electricity
demand,employmentstatus,urbanizationlevel,education
level, consumption level, etc. The relationship between
SW generation and these factors may be complex and
nonlinear. In some cases, ML has a better processing
ability than do traditional methods. Some studies have
investigated the effects of these factors on waste
generation.JiangandLiu (2016) analyzedtheinfluences
ofsocialandeconomicchangesonSW generationbased
onastatisticallearningapproach.Magazzinoetal.(2020)
investigated the relationship between SW production,
greenhouse gas emissions, and GDP in Switzerland.
Nguyenet al. (2020) explored the relationships between
SW, individual, and socioeconomic factors. The major
socioeconomic factors that influenced SW composition
weresafetyconcerns,economicactivities,andlifestyles.
MLmodelshavealsobeenusedtopredictthephysical
and chemical properties of SW. Physical properties
includethe wear rate (Nayak and Satapathy,2020), fiber
content, moisture content, dry unit weight, axial strain
(Falamaki and Shahin, 2019), and compression ratio
(Heshmati et al., 2014). Chemical properties include
chemicalcomposition (C,H,O,N)andheatingvalue.In
popular prediction models, the input variables are
spectrum data (Tao et al., 2020; Yan et al., 2021) or
physical composition (Lin et al., 2015). However, in
some studies, the input parameters of the characteristic
prediction model could be plant operating (You et al.,
2017) or seasonal variation (Adeleke et al., 2021).
Perhapstheaccuracyofpredictionsislowerthanthatof
actualmeasurements,butit is acceptable in manyactual
downstream treatments (Tao et al., 2020). In addition,
these predictions are obtained faster and are more
convenientthanthosebycomplexmeasurements.
In general, ML contributed to the quantitative
prediction of SW production, its relationship with
economicandsocialfactors,anditscharacterization.The
current aim of urban management is to promote SW
8 Front.Environ.Sci.Eng.2023,17(4):44
FSE-22087-LR2022-09-28 17:16:24
1
5
10
15
20
25
30
35
40
45
50
55
1
5
10
15
20
25
30
35
40
45
50
55
reductionandenergyutilizationofsolidwaste.Therefore,
itiscrucialtounderstandtherootcauseofSWtocontrol
its generation. The SW is a product of human social
activities. Establishment of the quantitative relationship
between SW generation and socioeconomic factors was
helpful to fundamentally clarify the root cause of SW
production and, consequently, reduce SW generation
fromeconomicandsocialactivities.Moreover,MLhelps
to construct the forecast model of weekly, monthly, or
yearly SW using a variety of linear and non-linear
methods.ThisprovidedanintuitiveunderstandingofSW
structure and quantity changes. Additionally, the SW
characterizationmodel constructed using multiple inputs
overcomes the drawbacks of traditional testing methods
andismoresuitableforautomationtechnology.MLplays
animportantroleinthegenerationandcharacterizationof
SW.
4.2Collectionandtransportation
LargeamountsofSWaregeneratedannually.Thus,they
requireappropriatemanagement.Inappropriateregulation
and disposal of SW have a negative influence on the
environmentandecosystems.Theopen-airstorageofSW
causessoil pollution.Toxicandharmfulcomponentscan
harmthesoilandenterthefoodchain,evenreachingthe
human body (Zhang et al., 2010; Liu et al., 2021). In
addition, SW seriously destroys aquatic environments.
SWdischargeintoriverscausesdamagetowaterquality,
river siltation, and blockage (Chandra et al., 2006). In
addition,dustandSWparticlespresentintheatmosphere
can cause atmospheric pollution (Rabl et al., 1998).
Therefore,itisnecessarytostudydifferentcategoriesof
SW, investigate their properties, and identify efficient
treatmentmethods.
Asmentionedabove,thegenerationofSWiscorrelated
with the economy, population dynamics, and policies.
These factors also influence SW management. In
addition, SW management is of great significance for
waste disposal and environmental improvement. It
containsseveraltopics.The collection andtransportation
ofSWwereinvestigatedinthissection.SelectionofSW
collection sites, proper transportation routes, treatment
plantsites,andproperproportionsofdifferent plants are
complexoptimizationproblems that need toconsider all
factors,includingprivateand external costs(Korucuand
Karademir, 2014). In recent years, ML has been
commonlyusedtosolvecomplexoptimizationproblems.
First, SW drop-off or collection points must be
selected. If the collection location is suitable, it can
contribute to reducing the cost and potential
environmental risks. ML has been used to study the
allocationoftheSWcollectionbins.In1998,Huangetal.
(1998)adjusted trash-flow allocationusing a graylinear
programming model. This was helpful in formulating
local policies. However, studies on collection bin
allocation were also considered along with the
optimization of the transportation path. The locations of
thecollectionpointswerealsotheinputparametersofthe
transport-optimization model. This topic is discussed
later.
Thus, it is important to consider waste transportation.
SW at the waste drop points should be transported by
trucksandothervehicles to thecentraltreatmentstation.
Because of differences in collection point locations,
disposal, city construction, SW components, collection
frequency, collection type (Vu et al., 2020), etc., the
transport route was different for different scenarios.
Furthermore, the rational truck route significantly
decreased the transport time, distance, and fuel
consumption. Therefore, to save energy and costs, it is
necessary to choose an appropriate transport path and
vehicle(including vehicle amount andvehiclecapacity).
Using a literature review, heuristic procedures (Beliën
et al., 2014) and genetic algorithms (Viotti et al., 2003)
were used to optimize SW transport routes. Geographic
information systems are also necessary for SW route
optimization.Theyareusuallystudiedusingacasestudy.
For example, Vu et al. (2020) developed and solved 48
vehicle routing problem models to optimize travel
distance and time in Austin, Texas, United States. The
input of the models contained geographical locations of
collection points, location of landfills, road network,
number of collection points, volume of waste at each
collectionpoint,andpick-uptime,andothers.Thereafter,
the model can output the optimized truck routes,
collection frequency, truck compartment configuration,
andtypeandcapacityoftrucks.Vuetal.(2019)explored
the effect of SW compositional features on optimized
truckroutetime,distance,andairemissions.Thishelped
savebetween 10.3% and 16.0 % in travel distance and
slightlyreducedemissions.Thesecasestudieswillbe of
great significance to waste system designers and
policymakers.
Akey factoraffectingthecollection andtransportation
of SW is the location of SW treatment plants. ML has
also been used to assist in selecting the location of SW
treatmentplantsandexploringtheoptimalproportionsof
differenttypes of plants.It is importantto consider cost
andenvironmentaleffectswhenselectingthelocationof
plants.Thepositionsoflandfills,waste-to-energyplants,
and MSW disposal sites have been studied previously.
Simseketal.(2006)introducedanewwastesiteselection
toolbasedonageographicinformationsystem.Basedon
fivedifferenthydrogeologicalparametersofgroundwater,
the tool developed different disposal area schemes and
was used in the Torbali Basin to select the optimal
project.However, ifthereareseveralSWdisposalplants
in a city, a considerable arrangement is required.
Otherwise,resources such as land, energy,or costs may
be wasted. Farzaneh et al. (2021) developed three
scenariostoconsiderincineration,composting,recycling,
RuiLiangetal.Applicationsofmachinelearninginthelifecycleofsolidwaste 9
FSE-22087-LR2022-09-28 17:16:24
1
5
10
15
20
25
30
35
40
45
50
55
1
5
10
15
20
25
30
35
40
45
50
55
and landfilling, thereby minimizing environmental
burdensandcosts.Inmy opinion, cloud-based resource-
allocation methods can be applied to optimize different
ratiosofplantsinthefuture.
Ingeneral,MLcontributed to optimizing the selection
of SW collection sites, transportation routes, treatment
plant sites, and the proportions of different plants.
Economic development, social form, topography of an
area,andrural-urbandifferenceshave important impacts
onSWdistributionandSWcollectionandtransportation.
Optimization of SW collection and transportation is not
only an environmental necessity but also an economic
necessity. ML is helpful for finding a fast and optimal
way to solve this problem and reduce energy and
economic consumption. ML plays an important role in
SWcollectionandtransportation.
4.3Treatmentandutilization
After the collection and transportation of SW, the
disposal of SW is the next important issue in SW
management. Energy utilization of SW can not only
reduce environmental hazards but also produce energy.
However, the heterogeneity of SW has a negative
influence on treatment. Thus, it is very important to
classify SW before energy utilization. Incineration,
landfills, anaerobic digestion, pyrolysis, and gasification
are commonly used to treat SW. With the rapid
development and good performance of ML, it has been
usedtostudySWclassificationandtreatment.
Waste classification is of great significance for
downstreamapplications.Ontheonehand,itensuresthat
recyclables can be recycled and reduce resource waste.
However, as mentioned before, there are different types
of SW and they have heterogeneous properties.
Therefore, different SW require different disposal
conditions to ensure an optimal energy conversion
efficiency. Moreover, it is necessary to classify SW
before utilization. Sophisticated SW classification
methodsbasedonimages,spectra,andsoundshavebeen
proposed in recent years. ML was used to establish the
classification models and performed well. Hannan et al.
(2014) achieved SW classification according to their
images based on a feedforward neural network with the
highest accuracy being 0.9875. Junjuri and Gundawar
(2020) identified ten post-consumer plastics combined
with laser-induced breakdown spectroscopy and an
artificialneuralnetwork,withanaccuracyofover97%.
Korucu et al. (2016) developed sound classification
systemsbasedonasupportvectormachine andahidden
Markov model to separate packaging waste with an
accuracy of over 88 %. Lai et al. (2017) explored food
waste recovery using an electrostatic separation process
basedonanartificialneuralnetwork.
Finally, almost all unrecoverable SW is sent for
disposal or energy utilization. Some points should be
considered in the treatment processes: reaction monito-
ring, product yield, pollutant control, and assessment of
the whole process. ML plays an important role in this
process. Input parameters of equipment, such as the
characteristics of SW, structure of equipment, and
reaction conditions, can influence the reaction process
and even the yield of the products and pollutants. By
learning previously processed data, ML can be used to
constructcomplex functionstosimulatetheircorrelation.
IncinerationisacommonmethodofSWtreatment,anda
detailedanalysisofMLunderdifferentenergyutilization
methodsisdiscussedbelow.
ML is typically used to simulate SW incineration and
predictpollutant emissions ofSW incineration. In 2000,
ChangandChen(2000b) designed a fuzzy controllerfor
municipal incinerators using genetic algorithms. Fuzzy
control technology is helpful for reducing operational
risks. It was also studied subsequently to provide more
reliable control of the combustion process (Chen et al.,
2002).ML was alsoused to testSW combustion (Dong
etal., 2002;Kabugoetal., 2020)andoptimizethe waste
incinerationplantusingmulti-objectives(Andersonetal.,
2005).They werehelpfulinprovidingclearobservations
and decision support to the human operator. Pollutant
emission control is of great significance for environme-
ntalprotection.In2000,ChangandChen (2000a) found
that artificial intelligence was useful for identifying
nonlinear structures in relation to polychlorinated
dibenzodioxin and dibenzofuran emissions from SW
incinerators. The emission characteristics of other
pollutants (e.g., SO2 (Wen et al., 2006) and HCl (Chi
et al., 2005; Zhang et al., 2019)) were studied using a
neural network. The pollutants were caused by
incompletecombustionofSW.Toreduceenvironmental
problems, it is necessary to find a more optimal
managementstrategy. Diagnosticanalysis of the Garson
index is useful solution (Chen and Chen, 2008).
Giantomassietal.(2011) predicted thesteamproduction
of an SW incinerator using fully tuned minimal RBF
neuralnetworks.
MLwasalsousedtosimulatetheSWlandfills,predict
the physical properties of SW in landfills, and evaluate
the environmental impacts (subsurface temperatures,
leachate, and methane emissions) and energy
consumptionofthelandfills.In2006,Ozcanetal.(2006)
first used artificial neural networks to simulate SW
landfills and predict CH4 production. Moreover, the
correlationswere 0.983 and 0.806 forprogramming and
testing, respectively. In addition, this year, a neural
networkwasusedtomodeltheleachateflowrateataSW
site (Karaca and Özkaya, 2006). Subsequently, ML was
applied to predict the physical properties of SW in
landfills. The long-term settlement of SW landfills
occurred in almost every landfill. This was attributed to
the long-term mechanical or decomposition-based
compression. A genetic algorithm was used to estimate
10 Front.Environ.Sci.Eng.2023,17(4):44
FSE-22087-LR2022-09-28 17:16:24
1
5
10
15
20
25
30
35
40
45
50
55
1
5
10
15
20
25
30
35
40
45
50
55
long-term and short-term settlements (Park and Park,
2009). The compression ratio of SW is an essential
parameter for evaluating settlement and landfill design.
The decision tree method (Heshmati et al., 2014),
adaptive fuzzy neural network, and ANN models
(Mokhtarietal.,2015)wereusedtopredictthecompre-
ssion ratio. In addition, landfill temperature is an
importantfactorforsafety. Satellite images andartificial
neuralnetworkswereusedtosimulateandpredictlandfill
surfacetemperature(Abu Qdais and Shatnawi,2019). A
risk assessment of elevated subsurface temperatures in
landfills is necessary. Sabrin et al. (2021) proposed a
comprehensive risk assessment framework specific to
subsurface elevated-temperature mitigation. In addition,
leachateandfugitivelandfillmethane emissions must be
considered in actual applications. They are harmful to
groundwater and air. ML has been used to predict the
generation (Abunama et al., 2018), COD load (Azadi
et al., 2016) and treatment process (Azadi et al., 2021;
Vazetal.,2021).MLhasalsobeenusedtosimulateand
predictmethaneemissions(Kormietal.,2018;Mehrdad
etal.,2021)andtheireffects(Singhetal.,2021).Finally,
landfillarea estimation is helpfulfor the better planning
and management of landfill sites (Hoque and Rahman,
2020).
Theenergyconsumptionandenvironmentalimpactsof
incinerationandlandfillswerestudiedusingML.Nabavi-
Pelesaraei et al. developed an ANN for forecasting and
modeling the energy and life cycles of incineration and
landfilling of SW. This is helpful for SW management
(Nabavi-Pelesaraeietal.,2017).
MLwasusedfortheanaerobicdigestionofSW.Itcan
predict methane production, simulate the process
performance,andevaluate the massandenergy balance.
Methane yield prediction and evaluation are important
topics in anaerobic research. In 2002, Holubar et al.
(2002) first used feed-forward backpropagation neural
networks to simulate methane production in anaerobic
digesters. Subsequently, biogas production on different
substratesusingdifferent operations was predicted using
different ML algorithms (Turkdogan-Aydınol and
Yetilmezsoy, 2010; Li et al., 2022b). In addition, SW
withhighbiodegradabilityandhighorganicandnutrient
contents is suitable for producing hydrogen through
anaerobic digestion. Elsamadony et al. (2015) predicted
biohydrogen production using an artificial neural
network.Inaddition,thefatesofCandNwerepredicted
basedonanartificialneuralnetwork(Lietal.,2016).ML
canalsobe usedtosimulatetheprocess (Saghourietal.,
2020)and predict control parameters (Flores-Asis et al.,
2018).Thisishelpful for monitoring and optimizing the
reactionprocess.Moreover,themassandenergybalances
andeconomicsofanaerobicdigestionmustbeconsidered
inactualapplications.Dahunsietal.(2017)optimizedthe
anaerobic co-digestion process based on the response
surfacemethodologyandanartificialneuralnetwork.
ML was also applied during pyrolysis, where it was
used to model the pyrolysis process and predict the
product yield. In 2016, a least-squares support vector
machinewasusedtopredictthebiocharyieldfromcattle
manure pyrolysis (Cao et al., 2016). ML has also been
used to predict the characteristics of products. Li et al.
(2020a) used an artificial neural network to predict a
higher heating value for syngas pyrolyzed from sewage
sludge, and ML was used to simulate the progress. A
geneticalgorithmandneuralfuzzymodelwereappliedto
determinetheoptimaloperatingconditionsoverdifferent
temperatureranges(Pan et al., 2021). MLhasalso been
used to investigate kinetic parameters (Pan et al., 2022)
andwastepyrolysisthermodynamics(Chenetal.,2021b)
andevaluate the potential ofpyrolysis (Ye et al.,2018).
Moreover, the distribution of special elements during
pyrolysiscanbepredictedusingML(Sunetal.,2020).
MLwasappliedto model the gasification processand
predictthecharacteristicsandyieldsofproducts.In2016,
anartificialneuralnetworkwasusedtopredictthelower
heating values of gas, tar, and entrained char (Pandey
etal.,2016).Subsequently,severalalgorithmshavebeen
used to model gasification progress and predict reaction
properties(Kardanietal.,2021).ML has also been used
to explore the influence of different variables on its
utilizationand to identifymore suitable conditions(Yan
etal.,2018;Kardanietal.,2021).
Furthermore,MLhasbeenappliedinthehydrothermal
reaction (hydrothermal carbonization, hydrothermal
liquefaction, and hydrothermal gasification) of SW to
model the hydrothermal reaction process, predict the
yieldandcharacteristicsoftheproducts,andoptimizethe
reactionconditions.In2020–2022,MLwillplay a more
important role in the hydrothermal reaction of SW.
Different ML algorithms, such as the gradient boost
regressor(Liet al.,2021b)andneuralnetwork(Lietal.,
2021a), were applied to model the process of
hydrothermal gasification of SW. ML has been used to
predict the characteristics of products, such as yield (Li
et al., 2021d), heating value of hydrochar, energy
recovery efficiency, energy densification (Li et al.,
2020b),yield,andOandNcontentinbio-oil(Zhangetal.,
2021).Inaddition,ML has been applied tooptimizethe
yieldofsyngasfromhydrothermalgasification(Li etal.,
2021b), optimize the production of bio-oil using high-
energy recovery and low nitrogen content from the
hydrothermal liquefaction of biomass (Li et al., 2021c),
andaidinthescreeningofcatalysts(Lietal.,2021a).
Ingeneral,MLhascontributed to the establishment of
SWclassificationmodels,modeleddifferentprocessesof
SW energy utilization (incineration, landfill, anaerobic
digestion, pyrolysis, gasification, and hydrothermal
reaction), predicted the characteristics of products,
optimized the reaction process, and evaluated the
environmentalandenergyperformanceofthetechnology.
SWclassificationbasedonMLisbeneficialforachieving
RuiLiangetal.Applicationsofmachinelearninginthelifecycleofsolidwaste 11
FSE-22087-LR2022-09-28 17:16:24
1
5
10
15
20
25
30
35
40
45
50
55
1
5
10
15
20
25
30
35
40
45
50
55
automatedand intelligentclassification,whichissuitable
for development direction of modernization and
intellectualization. In addition, for SW industrial energy
utilization,itisimportanttorecognizeefficientautomatic
monitoring, as well as to save manpower and material
resources.Moreover,MLishelpfulforexploring amore
efficient energy conversion and breaking the current
upperlimit.MLplaysanimportantroleinSWtreatment
andutilization.
5Future perspectives
Based on the quantitative bibliometric analysis and a
comprehensive summary of previous studies, a holistic
landscapeof ML applied to SWwas provided. Perspec-
tivesonfuturedevelopmentopportunitieswerediscussed
inthissection.
5.1Timeevolutionofauthorkeywordsfrom1985to2021
Thetimeevolutionofauthorkeywordscouldbeused to
reflectthedevelopmentofhottopicsinacertainresearch
field,evenprobablefutureperspectives.Thevisualization
ofauthorkeywordevolution(occurrences≥4)hasbeen
reducedinVOSviewer,asshowninFig.6.Themeansof
thecode sizeandlinesbetween thenodeswerethe same
as those mentioned under section 3.4. The colors of the
codes have slightly changed. They reflect the average
publication year of the author keyword, that is, the
weightedaverageofitsappearanceyearandtimes.Ifthe
colorof the node is closer to red, this indicatesthat the
authorkeywordappearedmorefrequentlyaround2020.If
it was closer to blue, it indicated that the keyword
appearedmorefrequentlybefore2014.
Theaveragepublicationyearof“Waste-to-energy”was
2020, followed by machine learning (ML) (2019.64),
recycling (2019.20), prediction (2019), and pyrolysis
(2018.75),therebyreflectingtherecentresearchhotspots.
The ongoing severe environmental problems and
energy crises may account for the occurrence of these
author keywords. On the one hand, mass accumulation
andimproperdisposal of SW havedamaged ecosystems
and the living environment. On the other hand, severe
energy shortage problems have attracted increasing
attentionworldwide.Waste-to-energyisnotonlyofgreat
significance for environmental governance but also for
solvingthecurrentenergycrisis.Theenergyutilizationof
SW is an effective method for solving these problems.
Pyrolysis, gasification, and hydrothermal methods have
several potential uses. Through these technologies, SW
can achieve high-value transformation and produce oil,
syngas, charcoal, etc. In addition, they produce less
pollutionthandoeswasteincineration.Recyclinghasalso
been emphasized because of its energy-saving and
environmental friendliness. It aims to recycle valuable
objectsfromalargeamountofwasteandachievereuse.
Wasteclassification could be helpful in realizing source
recycling, and has been carried out in many countries,
especially China, in recent years. This is beneficial for
Fig. 6OverlyVisualizationofauthorkeywordco-occurrence(>3).
12 Front.Environ.Sci.Eng.2023,17(4):44
FSE-22087-LR2022-09-28 17:16:24
1
5
10
15
20
25
30
35
40
45
50
55
1
5
10
15
20
25
30
35
40
45
50
55
carbonneutrality.
In addition, author keywords with lower occurrence
(<4)mayprovidenewideasforthefuturedevelopment
of ML applied to SW, such as waste collection (2021),
economic analysis (2021), environmental sustainability
(2020.50), and waste heat recovery (2019.67). The
economic and environmental assessment of waste
accumulation and disposal is of increasing concern,
which is consistent with the global call for carbon
reductionandhigherbenefits.
5.2Curerntchallenges
Data collection is the first step in an ML application.
However, there were some problems during this step.
First,somedata, including temperature measurements in
multipleareasoftheincinerator,weredifficulttoacquire
and could not be measured by equipment. Second, data
collection may consume lots of time. These data may
come from numerous repeated experiments or related
publishedarticles.Ittakesaconsiderableamountoftime
to manually repeat experiments and find data from
published articles. However, there was another problem
with the data from the published articles. In articles by
different researchers, the experimental operation would
differ, which could affect the uniformity of data.
Nonetheless,thisisdifficulttoavoid.
Most algorithms, including four algorithms (ANN,
SVM,RF, and GA) commonlyusedin SW, sufferfrom
the “black box” effect. The internal procedures of these
algorithms are unclear. We do not know the detailed
processing and calculation routes of the results after
inputting the variables. Further studies exploring these
routesare,therefore,required.
The results of the ML models should be critically
analyzed. A degree of bias may occur in model
processing but has not been discovered. Such biases
includedatabiasandalgorithmbias(Zhongetal.,2021).
Theyshouldbejudgedbasedonadeepunderstandingof
ML, professional knowledge of SW, or a team of
environmentalresearchers.
5.3Opportunitiesoffuturedevelopment
Increasingtheavailabilityofdataisimportant.Advanced
text-processingmethodsshouldbeexploredtoobtaindata
fromnetworksorpublishedarticles.Theresearchercould
voluntarily share the data on an open platform, such as
GitHub.An openandcomprehensivedatabase ofthelife
cycleofSWisrequiredtopromotetheapplicationofML
inthis field.Morefreedata, suchasthoseobtained from
PhyllisandBIOBIB,areencouraged.
The internal learning of integrated ML packages is
necessary. The transparency of ML is extremely
important when it is used, as this can increase its
credibility. Using something unknown in commercial
projectsposescertainrisksandlimitsitapplication.
Underglobalintelligentdevelopment,anefficientdata-
processingmethod,ML,willplayanimportantroleinthe
lifecycle of SW. The distribution of SW structure and
characteristics caused by waste classification will
continuetochangeinChina.Itrequirestimelymonitoring
and subsequent responses, to which ML can greatly
contribute. In addition, because of the change in SW
structure and characteristics, the progress of waste to
energy would be different, and it needs to be further
explored using ML to find more accurate scenarios.
Furthermore, the SW reactor may require more precise
adjustment. ML can aid in the optimization of reactor
system design and monitoring in internal reactors (Li
et al., 2022a). In addition, multiple types of ML input
(e.g., text, number, graph, image) may be helpful in
creating new knowledge, such as new catalysts (Zhong
etal.,2021).
6Conclusions
ThisstudyprovidesanoverviewofMLappliedinthelife
cycleof SW. Characteristics,including the totalnumber
of publications and citations, countries, institutions,
keywords, and distribution of research areas, of the
literatureonMLappliedtoSWfrom1985to2021based
on the SCI-EXPANDED and SSCI databases were
examined. They could be helpful for researchers and
policymakers to macroscopically study changes in this
researchfieldtemporallyandspatially.
ThethreepopulartopicsonMLappliedtothelifecycle
SW are SW categories, ML algorithms, and specific
applications. ML is mainly applied throughout the life
cycle of SW, including generation, characteristics,
collection,transportation,andutilization.MLcanbeused
topredictthegenerationandcharacteristicsofSWandto
exploretherelationship betweenSWgeneration,society,
andeconomicdevelopment.MLhasalsobeenappliedto
optimize the allocation of SW collection sites, SW
transportation, and SW treatment plant arrangement.
Moreover,ML can be usedto classify SW andsimulate
theenergyutilizationprocess.
Finally,perspectiveson the futuredevelopmentofML
applied to SW were discussed. The goal has been to
achieve high economic and environmental benefits and
carbon reduction during the lifecycle of SW. Increasing
the availability of data, internal learning of ML, and
enhancing the appropriate application of in SW may
create opportunities. We hope that this work will
contribute to addressing concerns associated with SW
disposalintheenvironmentandresources.
Notations
SW Solidwaste
RuiLiangetal.Applicationsofmachinelearninginthelifecycleofsolidwaste 13
FSE-22087-LR2022-09-28 17:16:24
1
5
10
15
20
25
30
35
40
45
50
55
1
5
10
15
20
25
30
35
40
45
50
55
ML Machinelearning
WoS WebofScience
TP Thetotalnumberofpublications
TC ThetotalnumberoftimesliteraturecitedintheWoS
SP Thenumberofsinglecountriespublications
CP Thenumberofinternationalcollaborativepublications
FP Thenumberoffirstcountrypublications
ANN Artificialneuralnetwork
SVM Supportvectormachine
RF Randomforest
GA Geneticalgorithm
AcknowledgementThisresearch was supportedbythe National Natural
ScienceFoundationofChina(No.52100157).
Electronic Supplementary MaterialSupplementarymaterialisavailable
in the online version of this article at https://doi.org/10.1007/s11783-023-
1644-xandisaccessibleforauthorizedusers.
References
Abu Qdais H, Shatnawi N (2019). Assessing and predicting landfill
surface temperature using remote sensing and an artificial neural
network. International Journal of Remote Sensing, 40(24):
9556–9571
Abunama T, Othman F, Younes M K (2018). Predicting sanitary
landfill leachate generation in humid regions using ANFIS
modeling. Environmental Monitoring and Assessment, 190(10):
597–611
AdelekeO, AkinlabiSA, JenTC,DunmadeI (2021).Applicationof
artificialneuralnetworksforpredictingthephysicalcompositionof
municipal solid waste: an assessment of the impact of seasonal
variation.WasteManagement&Research,39(8):1058–1068
Anderson S R, Kadirkamanathan V, Chipperfield A, Sharifi V,
SwithenbankJ (2005). Multi-objective optimization of operational
variables in a waste incineration plant. Computers & Chemical
Engineering,29(5):1121–1130
AzadiS,AmiriH,RakhshandehrooGR(2016).Evaluatingtheability
of artificial neural network and PCA-M5P models in predicting
leachate COD load in landfills. Waste Management (New York,
N.Y.),55:220–230
Azadi S, Karimi-Jashni A, Javadpour S, Mahmoudian-Boroujerd L
(2021).PhotocatalyticlandfillleachatetreatmentusingP-typeTiO2
nanoparticles under visible light irradiation. Environment,
DevelopmentandSustainability,23(4):6047–6065
Beliën J, De Boeck L, Van Ackere J (2014). Municipal solid waste
collection and management problems: a literature review.
TransportationScience,48(1):78–102
Bhatt Y, Ghuman K, Dhir A (2020). Sustainable manufacturing.
Bibliometricsandcontentanalysis.JournalofCleanerProduction,
260:120988
CaoH, XinY,YuanQ(2016). Predictionofbiochar yieldfromcattle
manure pyrolysis via least squares support vector machine
intelligentapproach.BioresourceTechnology,202:158–164
ChandraS,ChauhanLKS, Murthy R C, Gupta S K (2006). In vivo
genotoxic effects of industrial waste leachates in mice following
oral exposure. Environmental and Molecular Mutagenesis, 47(5):
325–333
Chang N B, Chen W C (2000a). Prediction of PCDDs/PCDFs
emissionsfrommunicipalincineratorsbygeneticprogrammingand
neural network modeling. Waste Management & Research, 18(4):
341–351
ChangNB,ChenWC(2000b).Fuzzycontrollerdesignformunicipal
incinerators with the aid of genetic algorithms and genetic
programming techniques. Waste Management & Research, 18(5):
429–443
Chen J C, Chen W H (2008). Diagnostic analysis of a small-scale
incinerator by the Garson index. Information Sciences, 178(23):
4560–4570
ChenK,PengY, LuS, LinB,LiX(2021a).Baggingbased ensemble
learning approaches for modeling the emission of PCDD/Fs from
municipalsolidwasteincinerators.Chemosphere,274:129802
Chen R, Zhang D, Xu X, Yuan Y (2021b). Pyrolysis characteristics,
kinetics, thermodynamics and volatile products of waste medical
surgical mask rope by thermogravimetry and online
thermogravimetry-Fourier transform infrared-mass spectrometry
analysis.Fuel,295:120632
Chen W C, Chang N B, Chen J C (2002). GA-based fuzzy neural
controller design for municipal incinerators. Fuzzy Sets and
Systems,129(3):343–369
ChiY, WenJM, ZhangDP,YanJ H,NiM J,CenKF(2005). HCl
emission characteristics and BP neural networks prediction in
MSW/coal co-fired fluidized beds. Journal of Environmental
Sciences(China),17(4):699–704
CoskunerG,JassimMS,ZontulM,KaratekeS(2021).Applicationof
artificial intelligence neural network modeling to predict the
generationofdomestic,commercialandconstructionwastes.Waste
Management&Research,39(3):499–507
Dahunsi S O, Oranusi S, Efeovbokhan V E (2017). Pretreatment
optimization, process control, mass and energy balances and
economicsof anaerobicco-digestionof Arachishypogaea(Peanut)
hullandpoultrymanure.BioresourceTechnology,241:454–464
Dai C, Li Y P, Huang G H (2011). A two-stage support-vector-
regression optimization model for municipal solid waste
management: a case study of Beijing, China. Journal of
EnvironmentalManagement,92(12):3023–3037
deSousaFDB(2021).Managementofplasticwaste:abibliometric
mapping and analysis. Waste Management & Research, 39(5):
664–678
DingX,YangZ(2020).Knowledgemappingofplatformresearch:a
visual analysis using VOSviewer and CiteSpace. Electronic
CommerceResearch,4:1–23
DongC,JinB,ZhongZ,LanJ(2002).Testsonco-firingofmunicipal
solid waste and coal in a circulating fluidized bed. Energy
ConversionandManagement,43(16):2189–2199
Elsamadony M, Tawfik A, Suzuki M (2015). Surfactant-enhanced
biohydrogen production from organic fraction of municipal solid
waste(OFMSW)viadryanaerobicdigestion.AppliedEnergy,149:
272–282
14 Front.Environ.Sci.Eng.2023,17(4):44
FSE-22087-LR2022-09-28 17:16:24
1
5
10
15
20
25
30
35
40
45
50
55
1
5
10
15
20
25
30
35
40
45
50
55
ErkinayOzdemirM,AliZ,SubeshanB,AsmatuluE(2021).Applying
machinelearningapproachinrecycling.Journalof MaterialCycles
andWasteManagement,23(3):855–871
Falamaki A, Shahin S (2019). Determination of shear strength
parameters of municipal solid waste from its physical properties.
CivilEngineering(Shiraz),43(S1):193–201
FarzanehG,KhorasaniN,GhodousiJ,PanahiM(2021).Application
of MCAT to provide multi-objective optimization model for
municipal waste management system. Journal of Environmental
HealthScience&Engineering,19(2):1781–1794
Flores-Asis R, Méndez-Contreras J M, Juárez-Martínez U, Alvarado-
LassmanA,Villanueva-VásquezD,Aguilar-LasserreAA(2018).
Use of artificial neuronal networks for prediction of the control
parameters in the process of anaerobic digestion with thermal
pretreatment.JournalofEnvironmentalScienceandHealth.PartA,
Toxic/Hazardous Substances & Environmental Engineering,
53(10):883–890
GiantomassiA,IppolitiG,LonghiS,BertiniI,PizzutiS(2011).On-
line steam production prediction for a municipal solid waste
incineratorbyfullytunedminimalRBFneuralnetworks.Journalof
ProcessControl,21(1):164–172
GuoHN,WuSB,Tian YJ, ZhangJ,LiuHT(2021).Applicationof
machinelearningmethodsfor thepredictionoforganicsolidwaste
treatment and recycling processes: A review. Bioresource
Technology,319:124114
Hannan M A, Zaila W A, Arebey M, Begum R A, Basri H (2014).
Featureextraction usingHoughtransformfor solidwastebin level
detection and classification. Environmental Monitoring and
Assessment,186(9):5381–5391
HeshmatiRAA,MokhtariM,ShakibaRadS(2014).Predictionofthe
compression ratio for municipal solid waste using decision tree.
WasteManagement&Research,32(1):64–69
Holubar P, Zani L, Hager M, Fröschl W, Radak Z, Braun R (2002).
Advanced controlling of anaerobic digestion by means of
hierarchicalneuralnetworks.WaterResearch,36(10):2582–2588
HoqueMM,RahmanMTU(2020).Landfillareaestimationbasedon
solidwastecollectionpredictionusingANNmodelandfinalwaste
disposaloptions.JournalofCleanerProduction,256:120387
Huang G H, Baetz B W, Patry G G (1998). Trash-flow allocation:
planningunderuncertainty.Interfaces,28(6):36–55
Idwan S, Mahmood I, Zubairi J A, Matar I (2020). Optimal
managementof solid wasteinsmart cities usinginternetof things.
WirelessPersonalCommunications,110(1):485–501
Jiang P, Liu X (2016). Hidden Markov model for municipal waste
generation forecasting under uncertainties. European Journal of
OperationalResearch,250(2):639–651
JunjuriR,GundawarMK(2020). A low-cost LIBS detection system
combined with chemometrics for rapid identification of plastic
waste.WasteManagement(NewYork,N.Y.),117:48–57
Kabugo J C, Jamsa-Jounela S L, Schiemann R, Binder C (2020).
Industry 4.0 based process data analytics platform: a waste-to-
energyplantcasestudy.InternationalJournalofElectricalPower&
EnergySystems,115:105508
KaracaF,ÖzkayaB(2006).NN-LEAP:aneuralnetwork-basedmodel
forcontrollingleachateflow-rateinamunicipalsolidwastelandfill
site.EnvironmentalModelling&Software,21(8):1190–1197
KardaniN, ZhouA, NazemM,LinX(2021). Modellingofmunicipal
solid waste gasification using an optimised ensemble soft
computingmodel.Fuel,289:119903
Keramatfar A, Amirkhani H (2019). Bibliometrics of sentiment
analysisliterature.JournalofInformationScience,45(1):3–15
KormiT, Mhadhebi S, Bel Hadj Ali N, Abichou T,Green R (2018).
Estimation of fugitive landfill methane emissions using surface
emission monitoring and Genetic Algorithms optimization. Waste
Management(NewYork,N.Y.),72:313–328
KorucuMK,KaplanÖ,BüyükO,GüllüMK(2016).Aninvestigation
of the usability of sound recognition for source separation of
packagingwastesinreversevendingmachines.WasteManagement
(NewYork,N.Y.),56:46–52
Korucu M K, Karademir A (2014). Siting a municipal solid waste
disposalfacility, part II:theeffects of externalcriteriaon the final
decision. Journal of the Air & Waste Management Association,
64(2):131–140
Lai K C, Lim S K, Teh P C, Yeap K H (2017). An artificial neural
networkapproachtopredictingelectrostaticseparationperformance
forfoodwasterecovery.PolishJournalofEnvironmentalStudies,
26(4):1921–1926
LiH, KeL,ChenZ,Feng G,Xia D,WangY, ZhengY, LiQ(2016).
EstimatingthefatesofCandNinvariousanaerobiccodigestionsof
manure and lignocellulosic biomass based on artificial neural
networks.Energy&Fuels,30(11):9490–9501
LiH,XuQ,XiaoK,YangJ,LiangS,HuJ,HouH,Liu B (2020a).
Predicting the higher heating value of syngas pyrolyzed from
sewage sludge using an artificial neural network. Environmental
ScienceandPollutionResearchInternational,27(1):785–797
LiJ,LiL,SuvarnaM,PanL,TabatabaeiM,OkYS,WangX(2022a).
Wetwastesto bioenergy andbiochar:acritical review withfuture
perspectives.ScienceoftheTotalEnvironment,817:152921
LiJ, PanL, SuvarnaM,TongYW, WangX(2020b).Fuelproperties
of hydrochar and pyrochar: prediction and exploration with
machinelearning.AppliedEnergy,269:115166
Li J, Pan L, Suvarna M, Wang X (2021a). Machine learning aided
supercriticalwatergasificationforH2-rich syngas production with
processoptimization and catalystscreening.ChemicalEngineering
Journal,426:131285
Li J, Suvarna M, Pan L, Zhao Y, Wang X (2021b). A hybrid data-
driven and mechanistic modelling approach for hydrothermal
gasification.AppliedEnergy,304:117674
Li J, Zhang L, Li C, Tian H, Ning J, Zhang J, Tong Y W, Wang X
(2022b). Data-driven based in-depth interpretation and inverse
designofanaerobicdigestionforCH4-richbiogasproduction.ACS
ES&TEngineering,2(4):642–652
LiJ,ZhangW,LiuT,YangL,LiH,PengH,JiangS,WangX,LengL
(2021c). Machine learning aided bio-oil production with high
energy recovery and low nitrogen content from hydrothermal
liquefaction of biomass with experiment verification. Chemical
EngineeringJournal,425:130649
LiJ,ZhuX, Li Y, Tong Y W,OkYS,WangX (2021d). Multi-task
prediction and optimization of hydrochar properties from high-
moisturemunicipalsolidwaste:applicationofmachinelearningon
waste-to-resource.JournalofCleanerProduction,278:123928
Lima L, Trindade E, Alencar L, Alencar M, Silva L (2021).
RuiLiangetal.Applicationsofmachinelearninginthelifecycleofsolidwaste 15
FSE-22087-LR2022-09-28 17:16:24
1
5
10
15
20
25
30
35
40
45
50
55
1
5
10
15
20
25
30
35
40
45
50
55
Sustainability in the construction industry: a systematic review of
theliterature.JournalofCleanerProduction,289:125730
LinX, WangF,ChiY, HuangQ,Yan J(2015).A simplemethodfor
predicting the lower heating value of municipal solid waste in
China based on wet physical composition. Waste Management
(NewYork,N.Y.),36:24–32
Liu C, Dong H, Cao Y, Geng Y, Li H, Zhang C, Xiao S (2021).
Environmentaldamagecostassessmentfrom municipalsolidwaste
treatmentbasedonLIME3model.WasteManagement(NewYork,
N.Y.),125:249–256
MagazzinoC, MeleM,SchneiderN (2020).Therelationship between
municipalsolidwasteandgreenhousegasemissions:evidencefrom
Switzerland.WasteManagement(NewYork,N.Y.),113:508–520
MehrdadSM,AbbasiM,YeganehB,KamalanH(2021).Predictionof
methane emission from landfills using machine learning models.
EnvironmentalProgress&SustainableEnergy,40(4):13629
MokhtariM,HeshmatiR A A, Shariatmadari N (2015).Compression
ratio of municipal solid waste simulation using artificial neural
networkandadaptive neurofuzzy system.EarthSciencesResearch
Journal,18(2):165–171
Nabavi-PelesaraeiA,BayatR,Hosseinzadeh-Bandbafha H, Afrasyabi
H, Chau K W (2017). Modeling of energy consumption and
environmental life cycle assessment for incineration and landfill
systems of municipal solid waste management: a case study in
Tehran Metropolis of Iran. Journal of Cleaner Production, 148:
427–440
NayakSK, Satapathy A(2020).Wearanalysis of wastemarbledust-
filled polymer composites with an integrated approach based on
designofexperimentsandneuralcomputation.Proceedingsofthe
Institutionof MechanicalEngineers.Part J,Journalof Engineering
Tribology,234(12):1846–1856
Nguyen,KLP,ChuangYH,ChenHW,ChangCC(2020).Impactsof
socioeconomicchangeson municipal solid wastecharacteristicsin
Taiwan, China. Resources, Conservation and Recycling, 161:
104931
NooriR,AbdoliMA, Ghasrodashti A A, Jalili Ghazizade M (2009).
Predictionofmunicipalsolidwastegenerationwithcombinationof
support vector machine and principal component analysis: a case
studyofMashhad.EnvironmentalProgress&SustainableEnergy,
28(2):249–258
Obileke K, Onyeaka H, Omoregbe O, Makaka G, Nwokolo N,
Mukumba P (2020). Bioenergy from bio-waste: a bibliometric
analysis of the trend in scientific research from 1998–2018.
BiomassConversionandBiorefinery,28(2):1–16
OzcanH,UcanO,SahinU,BoratM,BayatC(2006).Artificialneural
networkmodelingof methane emissions atIstanbulKemerburgaz-
Odayerilandfillsite.Journal of Scientific and Industrial Research,
65(2):128–134
Pan R, Duque J V F, Debenest G (2021). Investigating waste plastic
pyrolysis kinetic parameters by genetic algorithm coupled with
thermogravimetric analysis. Waste and Biomass Valorization,
12(5):2623–2637
Pan R, Duque J V F, Debenest G (2022). Waste plastic thermal
pyrolysisanalysisbya neural fuzzy model coupled withagenetic
algorithm.WasteandBiomassValorization,13(1):135–148
PandeyDS, Das S,PanI,Leahy J J,KwapinskiW(2016). Artificial
neuralnetworkbasedmodellingapproachformunicipalsolidwaste
gasification in a fluidized bed reactor. Waste Management (New
York,N.Y.),58:202–213
Park H I, Park B (2009). Prediction of MSW long-term settlement
induced by mechanical and decomposition-based compressions.
InternationalJournalofEnvironmentalResearch,3(3):335–348
Qu Y, Qian X, Song H, Xing Y, Li Z, Tan J (2018). Soil moisture
investigation utilizing machine learning approach based
experimental data and landsat5-TM images: a case study in the
MegaCityBeijing.Water,10(4):423
Rabl A, Spadaro J V, Mcgavran P D (1998). Health risks of air
pollution from incinerators: a perspective. Waste Management &
Research,16(4):365–388
Sabrin S, Nazari R, Karimi M, Fahad M G R, Everett J, Peters R
(2021). Development of a conceptual framework for risk
assessmentofelevatedinternaltemperaturesinlandfills.Scienceof
theTotalEnvironment,782:146831
SaghouriM, AbdiR,Ebrahimi-Nik M,RohaniA, MaysamiM(2020)
Modeling and optimization of biomethane production from solid-
state anaerobic co-digestion of organic fraction municipal solid
waste and other co-substrates. Energy Sources Part A. Recovery
UtilizationandEnvironmentalEffects:1–17
Shi M, Wang X, Shao M, Lu L, Ullah H, Zheng H, Li F (2023).
Resource utilization of typical biomass wastes as biochars in
removingplasticizerdiethylphthalatefromwater:characterization
andadsorption mechanisms.FrontiersofEnvironmentalScience &
Engineering,17(1):5
Simsek C, Kincal C, Gunduz O (2006). A solid waste disposal site
selection procedure based on groundwater vulnerability mapping.
EnvironmentalGeology,49(4):620–633
Singh D, Chavan D, Pandey A K, Periyaswami L, Kumar S (2021).
Determination of landfill gas generation potential from
lignocellulose biomass contents of municipal solid waste. Science
oftheTotalEnvironment,785:147243
Sun Y, Tao J, Chen G, Yan B, Cheng Z (2020). Distribution of Hg
during sewage sludge and municipal solid waste Co-pyrolysis:
influence of multiple factors. Waste Management (New York,
N.Y.),107:276–284
TaoJ,LiangR, Li J, Yan B, Chen G,ChengZ,LiW,Lin F, Hou L
(2020). Fast characterization of biomass and waste by infrared
spectra and machine learning models. Journal of Hazardous
Materials,387:121723
Turkdogan-Aydinol F I, Yetilmezsoy K (2010). A fuzzy-logic-based
model to predict biogas and methane production rates in a pilot-
scale mesophilic UASB reactor treating molasses wastewater.
JournalofHazardousMaterials,182(1–3):460–471
VazCAD,SamanamudGL,DaSilvaRS,FrancaAB,QuintaoCM
F,UrzedoAP,SilvaMB,BoschNetoJCB,AmaralMS,Loures
CCA,etal.(2021).Modelingandoptimizationofhybrid leachate
treatmentprocesses andscale-upof theprocess:review.Journalof
CleanerProduction,312:127732
ViottiP, PolettiniA,Pomi R,Innocenti,C (2003).Geneticalgorithms
asa promisingtoolfor optimisationoftheMSWcollection routes.
WasteManagement&Research,21(4):292–298
VuHL,Bolingbroke D, Ng K T W,FallahB(2019).Assessment of
waste characteristics and their impact on GIS vehicle collection
16 Front.Environ.Sci.Eng.2023,17(4):44
FSE-22087-LR2022-09-28 17:16:24
1
5
10
15
20
25
30
35
40
45
50
55
1
5
10
15
20
25
30
35
40
45
50
55
routeoptimizationusingANNwasteforecasts.WasteManagement
(NewYork,N.Y.),88:118–130
VuHL,NgKTW,FallahB, RichterA,KabirG(2020).Interactions
ofresidentialwaste composition and collectiontruckcompartment
designon GISrouteoptimization.WasteManagement(NewYork,
N.Y.),102:613–623
WanY,XiaoL,Wu C(2009).AnOptimumIntelligentAlgorithmand
its Application in Population Statistic and Forecast, 2009 WRI
GlobalCongressonIntelligentSystems,pp.40–44
Wang Y, Lai N, Zuo J, Chen G, Du H (2016). Characteristics and
trends of research on waste-to-energy incineration: a bibliometric
analysis, 1999–2015. Renewable & Sustainable Energy Reviews,
66:95–104
WangZ, PengX, XiaA,ShahAA, HuangY,ZhuX,ZhuX,Liao Q
(2022). The role of machine learning to boost the bioenergy and
biofuelsconversion.BioresourceTechnology,343:126099
Wen J, Yan J, Zhang D, Chi Y, Ni M, Cen K (2006). SO2 emission
characteristicsandBPneuralnetworkspredictioninMSW/coalco-
firedfluidizedbeds.JournalofThermalScience,15(3):281–288
YanB, Liang R,LiB, Tao J,ChenG, Cheng Z,ZhuZ, Li X(2021).
Fastidentificationandcharacterizationofresidualwastesvialaser-
inducedbreakdownspectroscopyandmachinelearning.Resources,
ConservationandRecycling,174:105851
Yan L, Li Y, Yang B, Farahani M R, Gao W (2018). Air-steam
gasification of municipal solid wastes (MSWs) for hydrogen
production. Energy Sources, Part A: recovery, Utilization, and
EnvironmentalEffects,40(5):538–543
YeG,LuoH,RenZ,AhmadMS,LiuCG,TawabA,Al-GhafariAB,
OmarU,GullM,Mehmood MA(2018).Evaluatingthebioenergy
potential of Chinese Liquor-industry waste through pyrolysis,
thermogravimetric, kinetics and evolved gas analyses. Energy
ConversionandManagement,163:13–21
YouH,MaZ,TangY,WangY,YanJ,NiM,CenK,HuangQ(2017).
ComparisonofANN(MLP),ANFIS, SVM,andRFmodelsforthe
online classification of heating value of burning municipal solid
wastein circulating fluidizedbedincinerators.Waste Management
(NewYork,N.Y.),68:186–197
ZhangH,YuS,ShaoL, He P (2019). Estimating source strengths of
HCl and SO2 emissions in the flue gas from waste incineration.
JournalofEnvironmentalSciences(China),75:370–377
ZhangW,LiJ,LiuT,LengS,YangL,PengH,JiangS,ZhouW,Leng
L, Li H (2021). Machine learning prediction and optimization of
bio-oil production from hydrothermal liquefaction of algae.
BioresourceTechnology,342:126011
Zheng H, Gu Y (2021). EnCNN-UPMWS: waste classification by a
CNN ensemble using the UPM weighting strategy. Electronics
(Basel),10(4):427
Zhong S, Zhang K, Bagheri M, Burken J G, Gu A, Li B, Ma X,
MarroneBL,RenZJ,SchrierJ,et al. (2021). Machine learning:
new ideas and tools in environmental science and engineering.
EnvironmentalScience&Technology,55(19):12741–12754
ZhangY, LiJ,Liu H,ZhaoG, TianY,Xie K(2021).Environmental,
social, and economic assessment of energy utilization of crop
residueinChina.FrontiersinEnergy,15(2):308–319
RuiLiangetal.Applicationsofmachinelearninginthelifecycleofsolidwaste 17
FSE-22087-LR2022-09-28 17:16:24
