- : Ms Word, Ms Word Format
- : 100 Pages
- : ₦5000
- : 1-5 Chapters
- Click to DOWNLOAD Materials
FACTOROPTIMIZATIONFORANAEROBICBIOGASGENERATIONFROM PALMOILMILLEFFLUENTUSINGARTIFICIALNEURALNETWORK
ABSTRACT
A preliminary study of anaerobic digestion of Palm Oil Mill Effluent (POME) was undertaken using a 1L gasometric assembly. Biogas generation at varying pH values (6.0 to 9.0) and temperature range (30oC to 60oC) were investigated in 14days. The results revealed optimal biogas yields at pH of 7.0 and temperature of 50oC. An Artificial Neural Network (ANN) model was applied in other to optimize the yield in anaerobic digestion of POME using experimental data from an integrated Bioreactor. The integrated bioreactor system comprised of a standard up flow anaerobic sludge bed (UASB) (Module 1), a centrally pump- stirred twin vessel bioreactor (Module II) and a standard continuous stirred tank (CSTR) (Module III). Operational data (Module I, II and III), were collected and employed in the analysis using Neural Network toolbox embedded in MATLAB R2015a. The study considered the effect of digester operational parameters, such as temperature (T), pH, C:N ratio, HRT (days), % C, % N, % P, % S and % COD reduction on the cumulative methane yield. A Two-layer feed forward neural network was trained with Levenberg-Marquardt Back Propagation training algorithm (trainlm) to simulate the digester operations. This was capable of predicting the yields of the methane production of the anaerobic digestion process as factors were varied simultaneously. The performance of the ANN model was validated and demonstrated the effectiveness of the model to predict the methane production accurately. The model with neural network structure 9-5-1 produced the best performance with the generated models showing a high correlation coefficient (R2-value). The linear regression between the network outputs and the corresponding targets showed a good correlation indicating that the outputs matched the targets closely based on the obtained ANN regression R2 values which ranged between 0.98262 to 0.99719. Application of the functional relationship of the methane process generated a set of model equations, which revealed that the methane yield can be predicted by a family of power series functions of the type:
y = 2.7 x 10-1x4 – 94x3 + 1.2 x 104x2 – 7 x 105x + 1.5 x 107 (R2 = 0.9976), where y is the
methane yield and R2 expresses agreement between observed values of methane yield and theoretical constructs. These equations are novel and have high predictive and analytical capabilities that can be used to predict methane generation as dependent on factors such as % COD reduction, pH and HRT. Finally, the proposed model has an acceptable generalization and can be used for further predictions for data sets with similar experimental conditions.
Keywords: Anaerobic digestion, artificial neural network, integrated bioreactor, POME, Optimization and model.
TABLEOFCONTENTS
Page
Titlepage i
Certification ii
Dedication iii
Acknowledgements iv
Abstract vi
TableofContents vii
ListofTables xi
ListofFigures xiii
ListofAbbreviations xxi
CHAPTERONE: INTRODUCTION 1
1.1 Background to Study 1
1.1.1 Alternative Energy Sources 2
1.1.2 Biomass Energy 3
1.1.3 Biogas Generations as a Waste Treatment Technology 4
1.1.4 Waste Generation in Palm Oil Mills 5
1.1.4.1 Gaseous Emission 6
1.1.4.2 Environmental Regulations of Effluent Discharge 6
1.1.4.3 Energy Sources 7
1.1.4.4 Palm Oil Mill Effluent Treatment System 8
1.1.4.5 Optimization Methods 8
1.1.5.1 Simplex Methods 9
1.1.5.2 Optimization Methods In Anaerobic Digestion Processes 10
1.1.6 Optimization Using Artificial Neural Network 12
1.2 Statement of Problem 13
1.3 Research Aim and Objectives 15 1.4 Hypothesis of The Study 15
1.5 Significance of Study 16
1.6 Scope of The Study 17
CHAPTERTWO:LITERATUREREVIEW 18 2.1 Anaerobic Digestion of POME 18
2.1.1 POME Treatment 19
2.2 Anaerobic Digestion Compared with Alternative Methods 20
2.2.1 Conventional Treatment Systems 21
2.2.2 Anaerobic Filtration 22
2.2.3 Fluidized Bed Reactor 24
2.2.4 Up-Flow Anaerobic Sludge Blanket (UASB) Reactor 26
2.2.5 Up-Flow Anaerobic Sludge Fixed-Film (UASFF) Reactor 28
2.2.6 Expanded Granular Sludge Bed (EGSB) 30
2.2.7 Continuous Stirred Tank Reactor (CSTR) 31
VI
2.2.8 Anaerobic Contact Digestion 32
2.3 Comparison of Various Anaerobic Treatment Methods of POME 33
2.4 Factors Affecting Anaerobic Digester Performance 36
2.4.1 Effect of pH 37
2.4.2 Effect of Mixing 38
2.4.3 Effect of Temperature 39
2.4.4 Effect of Organic Loading Rates 41
2.4.5 Effect of Hydraulic Retention Time 41
2.4.6 Effect of Carbon: Nitrogen Ratio 42
2.4.7 Effect of Nutrients for Bacteria 42
2.4.8 Effect of Chemical and Physical Pre-Treatment 43
2.5 Mechanism of Anaerobic Digestion 45
2.5.1 Hydrolysis 45
2.5.2 Acidogenesis 45
2.5.3 Acetogenesis 46
2.5.4 Methanogenesis 46
2.5.5 Biochemical Mechanism of Anaerobic Digestion 47
2.5.6 Comparative Reactor Performance 48
2.5.7 Identification of Responsive Microorganisms 52 2.5.8 POME Stabilization in terms of COD removal 53
2.6 Kinetic Studies 54
2.6.1 Theoritical Mathematical Model 56
2.6.2 Advanced Model for Biogas Yield 58
2.6.3 Simple Model for Biogas Yield 59
2.6.4 Summary of Bio-Chemical Kinetics Models 60
2.7 Empirical Optimization Conditions 63
2.7.1 Reactor Optimization by application of Simplex Models 63
2.8 Computer Assisted Modelling 64
2.8.1 Data Simulation 64
2.8.2 Predictive Analytics 64
2.8.3 Multilayer Perceptron 65
2.8.4 Artificial Neural Networks 65 2.8.5 Uses of Artificial Neural Nets 67 2.8.6 Types of Artificial Neural Networks 69
2.8.7 Algorithms and Logic Systems Associated with ANN 70
2.8.8 MATLAB 70
2.8.9 Curve Fitting Tool Box 71
2.9 Functions Used to Process Data in ANN 71
2.9.1 Step Function 72
2.9.2 Radial Basic Function 73
2.9.3 Linear Combination and Continuous Log-Sigmoid Function 73
2.9.4 Learning Paradigms 75 2.9.5 Gradient Descent 76 2.9.6 Backpropagation 76
2.9.7 Boltzmann Learning 78
2.9.8 Hebbian Learning 79
2.10 Application of ANN in Environmental Waste Management System
Modelling 79
2.10.1 ANN for Modeling of COD Removal from Waste Aqueous Antibiotic
Solutions 80
2.10.2 Modelling with ANN for Dosage Control and Color Removal from
Waste Textile Water 82
2.10.3 Application of ANN for Prediction of Methyl Tert-Butyl Ether (MTBE)
Degradation 83
2.10.4 Application of ANN for Predicting Nitrogen Oxides Removal Efficiency 85
2.10.5 Application of ANN in Air Quality Prediction 85
2.10.6 Simulation of Process Control using nonlinear Systems Control 87
2.11 Application of ANN in Experimental Data Analysis of Biogas Production 88
2.11.1 Application of ANN in Biogas Production from Sago Wastewater using
Anaerobic Tapered Fluidized Bed Reactor 88
2.11.2 Application of ANN in Biogas Production from fresh Organic Waste 88
2.11.3 Application of ANN in Biogas Production 89
2.11.4 Application of ANN in Biogas Production from Field-Scale Landfill
Bioreactors 90
2.11.5 Application of ANN in Biogas Production from Anaerobic Co-Digestion
of Leachate with Pineapple Peel 91
CHAPTERTHREE: METHODOLOGY 94
3.1 Research Materials 94
3.1.1 Laboratory Equipment Used 94
3.1.2 Laboratory Reagents Used 94
3.2 Reactor Rig 95
3.2.1 Definition of Bioreactor Modules 98
3.2.2 Biogas Collector 98
3.3 Selection of Substrate and its Volume 99
3.4 Laboratory Analysis 99
3.4.1 Characterization of the Feed 99
3.4.2 pH Determination 100
3.4.3 Nitrogen/Crude Protein Determination 100 3.4.4 Determination of Total Organic Carbon (TOC) 101 3.4.5 Determination of Volatile Solids 102
3.4.6 Determination of Biochemical Oxygen Demand (BOD) 103 3.4.7 Determination of Chemical Oxygen Demand (COD) 104
3.4.8 Determination of Alkalinity 106
3.4.9 Determination of Density 106
3.5 Preliminary Studies (Gasometric Assembly) 106 3.5.1 Experimental Variables (Preliminary Studies) 106 3.5.2 Preparation of Feed Slurry for the Variable Determination 108 3.5.3 Analysis of Results of Preliminary Studies 108
3.5.4 Gas Chromatography (GC) Analysis of Product Mixture 109
3.6 Digestion of POME Using Integrated Reactors 109
3.6.1 Experimental Initialization of the Integrated Bioreactor 109
3.6.2 Maintenance of Anaerobic Conditions 110
3.7 Modeling and Optimization of Biogas Production Using Artificial Neural
Network 110
3.7.1 MATLAB Neural Network Toolbox 110
3.7.2 Neural Network Fitting Tool Box 111
3.7.3 Neural Network Architecture 112 3.7.4 Neural Network Size 113
3.7.5 Neural Network Training 114
3.7.6 Model Validation 117
3.7.7 Artificial Neural Network Model for Biogas Yields Obtained 117
CHAPTERFOUR:RESULTSANDDISCUSSION 119
4.1 Results on Preliminary Studies 119
4.1.1 Data on Variable Factor Determination 122
4.1.2 ANOVA based on Preliminary Analysis 123
4.1.3 Methane Yield from Integrated Bioreactors 126
4.2 Artificial Neural Network Model 130
4.2.1 Graphical Representations of the ANN Training Results 131
4.2.3 Interpretation of Figures and Graph from ANN 155
4.2.4 Data Mining and Normalization 156
4.2.5 Mathematical Modelling to Circumvent Kinetics of Biodegradation 156
4.3 MATLAB Code for Neural Networks Model 158
4.4 Functional Relationship Between Percentage COD Reduction and the
Biogas Yields 160
4.5 Functional Relationship Between pH and the Biogas Yields 166
4.6 Functional Relationship Between Time (HRT) and the Biogas Yields 174
4.7 Chromatographic Analysis of the Biogas Produced 181
4.8 Feed Stock/ Waste (Spent Slurry) Characterization 181
4.9 Economic Analysis of Biomethane Processing 182
CHAPTERFIVE: CONCLUSIONANDRECOMMENDATION 185
5.1 Conclusion 185
5.2 Contribution to Knowledge 186
5.3 Recommendations 187
REFERENCE 188
CHAPTER ONEINTRODUCTION
1.1 BACKGROUNDTOSTUDY
Theglobeisfacedwiththetwincrisesofsafeenergygenerationandclimatechange.Global
warming as a result of CO2increaseintheatmosphereduetotheuseoffossilfuelscauses
greenhouseeffect(Courtney&Dorman,2003;Ejike&Anyile,2011;Ramadhas,Jayaraj,&
Muraleedharan,2004).Hence,theneedtoexploreandexploitnewenergysourcesthatare
renewableandecologicallysustainableispertinent(Dow&Downing,2006;Mills,2009).
Furthermore,ithasbeendemonstratedthatenergysupplyandutilizationcanbeoptimized
byincorporatingdifferentfactorsthatenhancebestmanagementpracticesintheIndustry
(Ejike&Anyile,2011).Biogasproductionhasemergedasoneofthepromisingalternatives
tofossilfuelsources,andthedevelopmentofthistechnologywillhelphumanityto
overcomethecurrentenergycrisisandalsoprovideacleansourceofenergytocombatthe
phenomenonofglobalwarming(Levin,Pitt,&Love,2004;Nath&Das,2011).
Anaerobicdigestion(AD)isaversatilebiochemicalprocess,whichiscapableofconverting
mosttypesofbiodegradableorganicmatterunderanaerobicconditionintoamixtureof
gasesconsistingofmethaneandcarbondioxide(Blonskaja&Vaalu,2006).ADprocessis
controlledbyseveralprocessparametersandthedeterminationoftheoptimumvaluesof
theseparametersarecrucialforbioprocessdevelopmentandscale-up(Wang&Wan,
2009b).ADprocessesaredescribedtobenon-linear,complexandunsteady;thus,itis
challengingtodevelopaprecisephysical-basedformulatocharacterizetheirphysical
performance.Similarly,thedevelopmentofaccurateADprocessmodelscontinuestobaffle
expertsduetothenon-linearnatureofthebiochemicalnetworkinteractionsthatoccur
duringbioprocesses(Franco-Lara,Link,&Weuster-Botz,2006).Modellingand
optimizationofbioprocessescontributestoincreasedunderstandingoftheprocessinputsfor
1
optimumyieldandproductionrate.Variousmodellingalgorithmshavebeenappliedin
biogasproductionprocesses,andresultshaveshownthatmodellingandoptimizationcan
enhancebiogasyields(Sewsynker-Sukai,Faloye,&Kana,2017).
However,theimprovementofbiogasproductionrequiresarobustprocessmodelthat
accuratelyrelatestheeffectofinputvariablestotheprocessoutput.Artificialneural
networks(ANNs)haveemergedasoneofthelatesttoolsforoptimizationandmodelling
complex,non-linearprocesses.ANNsareappliedtopredictvariousprocessesthatareuseful
forvirtualexperimentationsandcanpotentiallyenhancebioprocessresearchand
development(Sewsynker-Sukaietal.,2017).
1.1.1 AlternativeEnergySources
Duetotoday’senergydemandandthepossibilityofdepletionofnon-renewableenergy
resources,theneedtoexploreandexploitnewenergysourcesthatarerenewable,secureand
ecologicallysustainableispertinent.Alternativeenergysourcesevolvinginthelasttwo
decades are gaining attention all over the world because they do not cause increase in CO2
compositionoftheatmosphere.Theyarealsoknowntoenhanceenergyconservationand
providenewjobopportunities(Ejike&Anyile,2011;Mshandete,Björnsson,Kivaisi,
Rubindamayugi,&Mattiasson,2006).Inordertoencouragethediversificationofenergy
generation,variousprotocolshavebeendevelopedtogiveassistancetoenergygeneration
methodswhichdonotdirectlydependonpetroleumandnaturalgasreserves(Wayman&
Parekh,1990).Variousalternativeenergysourcesinclude;solarenergy,geothermalenergy,
nuclearenergy,hydro-power,windenergyandbiomassenergy.
Thedifferentalternativeenergysourceshavepotentialnottoincreasetheconcentrationof
carbondioxide (CO2) in the atmosphere. However due to the contribution of CO2byfossil
fuels,theyarenolongerregardedasthebestoptioninmodernenergysupply(Ejike&
Anyile,2011;Ejike,Chinedu&Egbujor,2012).ThepredictionmadebyDowandDowning
(2006) has led to several protocols on the management of CO2emissionandother
greenhousegases.Themanagementsystemhasledtowhatisreferredtoas“CarbonCredit”.
Acarboncreditisagenerictermforanytradablecertificateorpermitrepresentingtheright
toemitonetoneofcarbondioxideormassofanothergreenhousegaswithacarbondioxide
equivalent.Carboncreditandcarbonmarketarecomponentofnationalandinternational
attemptstomitigatethegrowthinconcentrationofgreenhousegases(ClimateChange
Glossary,2008).Biomassenergybeinglessenvironmentallydemandingisadaptableto
operationindevelopingcountries.
1.1.2BiomassEnergy
Thisinvolvesplantremainsbeingconvertedintosmallermoleculeswhichcouldbeburntto
generateenergy.Twoprocessesarewidelyknown.First,thecelluloseofplantcellwallcan
be hydrolyzed to glucose which in turn is fermented into ethanol and CO2(Singh,2001).
Secondly,anaerobicdigestionprocessofplantandotherorganicmatterbyaconcertiaof
microorganismstogivebiogas(Ghosh,Samanta,&Ray,2011).Theuseofbiogasfrom
biomass as an energy source is regarded as CO2 neutral, because CO2releasedduring
combustion of the biogas is the CO2thattheplanthasassimilatedduringphotosynthesisto
createorganicbiomass(Jewel,1974).Thisisincontrasttoburningfossilfuelswhich
resultedfromplantthatlivedover20millionyearsbefore.
Biogasgenerationisthereforeoneofthemostdesirableoptionsamongthevariousother
alternativesourcesofrenewableenergycurrentlyavailable,asitrequireslesscapital
investmentperunitproductioncost(Ghoshetal.,2011).Ithasbeendemonstratedearlier
thatbiogasproductionfromcropresiduesiseconomicallyfeasibleonafarmscalelevel50–
500KW (Wayman & Parekh, 1990). Fermenting 8000 m3ofslurryformedfromorganic
fractionsofsolidwasteandanimalwastewithadepressionof10%generallygives
electricityat16Ev/KWh(Hoffman,2001).
Introductionoftheuseofrenewableenergyisusuallyapoliticallydrivendecision.The
reasonisthatrenewableenergyismoreexpensivebutitsusebylargepopulationsdecreases
thecost.Europeancommunityhasagreedonatarget,forconversionofenergybased
sourcesandtheirstandardwillbe20%renewableenergybytheyear2020(European
EnergyCommission2010).
1.1.3BiogasGenerationsAsAWasteTreatmentTechnology
Biogasaregeneratedfromfoodindustrywastesuchasdairywaste,chocolate,sugarcane,
beefhoof,potatoes,oilseed(e.g.olive,palmfruit,coconutfruit),brewerywaste,vegetable,
suchastomatoes,mangoandguavajuices.Othersubstratesforbiogasgenerationinclude
municipalorganicwaste(MOW),sewage,manurepoultry,piggeryandcattlefarms
(Baharuddimetal.,2010;Demirel,&Schere,2008;Fang,Angelidaki,&Boe,2010).
Biogastechnologyhasbeenshowntobeasustainablealternativeforwastetreatmentin
municipalandindustrialwastedisposalsystem.Indeed,thetownofLinkopinginSweden
obtained90%oftheirrequiredenergyneedfromprocessinganimalmanureparticularlypig
manureimportedfromtheNetherland.Ninety-eightpercentorganicwastesfromhousehold
inthecityarerecycledthroughprocessingtobiogasandusedforelectricitygeneration
(Philipsson,2001).InLinkopingCity,70%ofthecityvehiclesrunonbiogasfrom
municipalorganicandanimalwastes.
LinkopingCitysupportstheslightlyhighercostofbiogasgeneratedenergythroughcarbon
tradeexchangewhichisastockexchangeforcarboncredit.HenceLinkopinghashighly
validatedandsophisticatedprocesses,whichallowstheirassessmentasacleanenergy
developmentenvironmentintheinter-governmentalpanelonclimatechange(IPCC,2012)
(KyotoWorkDataPoliciesInfrastructures,2007).Thisisinlinewiththerequirementofthe
KyotoProtocolonthecapsandquotaforgreenhousegas(KyotoProtocolTargets,2010).
1.1.4WasteGenerationInPalmOilMills
The main produce of a palm oil mill is the crude palm oil (CPO). Other products are liquid
effluent and solid wastes (empty fruit bunches, palm kernel, fibre and shell), which may
haveasignificantimpactontheenvironmentiftheyarenotdealtwithproperly.The
productionofpalmoilresultsinthegenerationoflargequantitiesofwastewatercommonly
referredtoaspalmoilmilleffluent(POME).Typically,1tonneofcrudepalmoil
productionrequires5-7.5tonnesofwater;over50%ofwhichendsupasPOME(Ma,
1999).ThePOMEcomprisesacombinationofwastewaterfromthreemainsourcesviz.
clarification(60%),sterilization(36%)andhydrocyclone(4%)units(Ma,1999).
POMEcontainsvarioussuspendedcomponentsincludingcellwalls,organelles,shortfibres,
aspectrumofcarbohydratesrangingfromhemicellulosestosimplesugars,arangeof
nitrogenouscompoundsfromfreshfruitbunch.Fromenvironmentalperspective,fresh
POMEisahotandacidicbrownishcolloidalsuspension,characterizedbyhighamountsof
totalsolids(40,500mg/l),oilandgrease(4000mg/l),COD(50,000mg/l)andBOD(25,000
mg/l)(Ejike,Chinedu&Egbujor,2012;Singh,Ghumen,&Grewal1999;Ma,1999).The
chemicalcharacteristicofatypicalPOMEissummarizedinTable1.1.
Table 1.1 Typical characteristics of POME (Ma, 1999).
Parameter pH | Average | Metal | *Average |
4.7 | Phosphorous | 180 | |
Oil and Grease | 4000 | Potassium | 2270 |
Biochemical Oxygen | 25000 | Magnesium | 615 |
Demand (BOD5)
Chemical Oxygen Demand (COD) |
50000 | Calcium | 439 |
Total Solids | 40500 | Boron | 7.6 |
Suspended Solids | 18000 | Iron | 46.5 |
Total Volatile Solids | 34000 | Manganese | 2.0 |
Ammonical Nitrogen | 35 | Copper | 0.89 |
Total Nitrogen | 750 | Zinc | 2.3 |
* All in mg/l except pH. |
*
1.1.4.1 GaseousEmission
Palm oil mills are generally self-sufficient in terms of energy requirements due to the
availability of adequate quantities of fiber and shell materials that are used as solid fuel in
the boiler. There are two sources of air pollution in the mills, the boiler and incinerator that
arecausedbyincompletecombustionofthesolidwastematerials;wastefiber,shell
materials and empty fruit bunch (Thani, Hussin, Wan Ramlah, & Sanusi.,1999).Themain
practice of treating POME is by using ponding and/or open digesting tank systems (Ma,
1999). The emission of greenhouse gases (CH4 and CO2)fromtheseopensystemstothe
atmospherehasbeenrecentlyreportedasasourceofairpollutionfromthepalmoilmills
(Yacob, Hassan, Shirai, Wakisaka, & Subash,2005).
1.1.4.2 EnvironmentalRegulationsofEffluentDischarge
Theenvironmentalcontrolinpalmoilindustrywasgrantedalicensedapproachthatwould
permitintimatecontrolofindividualfactories.Thisapproachalsoprovidesamechanismfor
permittingvariableeffluentstandardstobeappliedbasedonthedemandsofprevailing
environmentalcircumstances.Theenvironmentalqualityregulationsforthecrudepalmoil
industrywerethefirstsetofregulationspromulgatedundertheEnvironmentalQualityAct
(EQA), 1977, for control of industrial pollution sources (Thani et al.,1999).Theeffluent
dischargestandardsapplicabletocrudepalmoilmillsarepresentedintheTable1.2.
Table 1.2 Effluent discharge standards for crude palm oil mills (EQA 1974, 2005).
Parameter | unit | Parameter limits | Remarks |
BiochemicalOxygenDemand | mg/l | 100 | |
ChemicalOxygenDemand | mg/l | * | |
TotalSolids | mg/l | * | |
SuspendedSolids | mg/l | 400 | |
OilandGrease | mg/l | 50 | |
AmmoniacalNitrogen | mg/l | 150 | Valueoffilteredsample |
TotalNitrogen | mg/l | 200 | Valueoffilteredsample |
pH | – | 5-9 | |
Temperature | °C | 45 |
Note:*Nodischargestandardafter1984.
1.1.4.3 EnergySources
Duetotheincreaseindemandforenergy,costsavingandtheprotectionoftheenvironment,
otherenergysourcesthatprimarilycompriseofoil,naturalgas,hydropower,coal,solar
powerandbiomassarenowbeingexploited.AspresentedinTable1.3,naturalgas,
hydropower,andbiomassenergyresourcesinMalaysiaaregenerallyabundant.
Table 1.3Energy resource potential in Malaysia (ASEAN, 2003).
Energy resources | Amount | Unit |
Oilreserve | 5.0 | Billionbarrels |
Gasestimatereserve | 2402 | Billioncubicmeters |
Coalprovenreserve | – | Milliontonnes |
Hydropowertechnically | 72 | Twh/y |
FeasibleBiomass | 665 | MW |
Geothermalpotential | – | MW |
Windenergypotential | – | MW |
InMalaysia,themostextensivestudyontheuseofbiomasshasbeenonpalmoilwastes,
whichcanbeutilizedtomeettheenergyrequirementofthepalmoilmillsandtheelectricity
needsoftheworkers.Thetotalenergypotentialofthebiomassisestimatedtobeabout5%
ofMalaysianelectricitydemand.Therefore,renewableenergyhasbeenidentifiedby
Malaysian government as the 5thfuelunder‘TheNewFive-FuelDiversificationStrategy’
(EuropeanEnergyCommission,2002).Therearedifferenttypesofbiomassgeneratedbya
palmoilmill.Fromthefourbiomasssources,threeofthem(emptyfruitbunch,fiberand
shell)canbedirectlyburnedasfuelwhilePOMEmustfirstbeanaerobicallyconvertedto
methane.Therefore,itisessentialforahighrateanaerobicbioreactortobeappliedasitcan
servedual-functioni.e.wastewatertreatmentandenergygeneration(organicconversionto
methane).
1.1.4.4PalmOilMillEffluentTreatmentSystem
Palm oil industries are facing tremendous challenges of meeting the increasingly stringent
environmental regulations. Over the past decades, several cost-effective treatment
technologies comprising of anaerobic, aerobic and facultative processes have been
developed for the treatment of POME (Ma,1999).Morethan85%ofpalmoilmillsuse
solely ponding systems due to their low costs (Thani et al.,1999).Ithasbeenreportedthat
onlyafewmillsareequippedwithbiogasrecoverysystems(Yeoh,2004).Longhydraulic
retentiontimes(HRT),lowtreatmentefficiency,highsludgeproduction,extensivelandarea
requirement and emission of large amount of greenhouse gases (CO2 and CH4)aresomeof
thedrawbacksofconventionalPOMEtreatmentmethods.However,theoptimizationand
applicationofefficient,stableandeconomichighrateanaerobictreatmentsystemsarebeing
investigated (Yacob et al.,2005).
1.1.5 OptimizationMethods
Optimizationwithmodellingandsimulationhasbeenanintegralpartofmoderndesign
practice.Themainaimofanydesignproblemistofindasetofgood,feasibleandideally
bestsolutionstoagivenproblem,whichmeasuresuptodesigncriteriasuchastheminimum
costs,highperformance,sustainability,recyclabilityandenergyefficiency.Thestringent
requirementsofminimizingenvironmentalimpactandcarbonfootprintrequireaparadigm
shiftinscientificthinkinganddesignpractice.Despitethesignificantprogressmadeinthe
lastfewdecades,therearestillchallengesrelatedtononlinearity,scaleoftheproblem,time
constraintandthecomplexityofthesystem(Yang,Koziel,&Leifsson,2013).Mathematical
modelinghasmostmodelsbasedonpartialdifferentialequationsthatarenotanalytically
solvable,hence,approximatemethodsandnumericalmethodsarethealternative.For
complex tasks in which the actual system behaviors have no explicit or closed expression
(equations), approximate learning-based models such as neural networks are very useful and
effective in practice (Yang et al., 2013). An algorithm development is one of the main activities in optimization. An algorithm is an iterative procedure for solving a class of
problemsandmustbefinite(optimalsolutionsarefoundinafinitenumberofiteration).
Inmathematics,optimizationisthedisciplineconcernedwithfindinginputsofafunction
thatminimizeormaximizeitsvalue,whichmaybesubjectedtoconstraints.Inrecenttimes,
researchershavesolutionstoreal-lifeproblemsasaresultofnewtechnological
developmentsinalgorithmsandcomputerhardware(Ba˜nos,Manzano-Agugliaro,Montoya,
Gil,Alcayde,&Gomez,2011;Yangetal.,2013;Yu,Wensel,Ma,&Chen,2013).
Optimizationdoesnotessentiallymeanfindingtheoptimumsolutiontoaproblem,sinceit
maybeunworkableduetothecharacteristicsoftheproblem,butinrecentdecadesmany
authorshaveproposedapproximatemethods,includingheuristicapproachesandartificial
neuralnetworks(ANN),tosolvetheseproblemsinsteadofusingtraditionaloptimization
methods,suchaslinear-programming(LP),SteepestAscent,Simplexmethod,quadratic
programming(QP),etc.(Yuetal.,2013).However,tilldatemostcomputational
optimizationmethodshavefocusedonsolvingbothsingle-objectiveandmulti-objective
problems.Thisapproachhasseveraldrawbacksasitonlyreturnsasinglesolutionasaresult
ofthesearchprocess,whichbecomesanimportantlimitationinthedecision-making
process,wherethedecisionmakermustselectonesolutionfromseveralalternatives(Ba˜nos
etal.,2011).
1.1.5.1SimplexMethods
TheSimplexmethodsarewell-knownalgorithmusedforlinearprogramming.Itpresentsan
organizedstrategyforevaluatingafeasibleregionsverticesandthishelpstofigureout
optimalvalueoftheobjectivefunctioninalinearprogram.Itisclearthatonefactoris
crucial to the method: which variable should replace which. Simplex method is very
efficient in practice. However, its worst-case complexity is exponential as demonstrated
with carefully constructed examples (Carreira-Perpinan, 2017). There is also a complexity of polynomial functions for both average and worst cases in linear programming. Linear
Programming Models can be used easily to solve two variables, but more than three to five
variables involves the simplex method as it is an iterative method which by repeated use
gives the solution to any n variable. Simplex Methods deals with standard form
(standardization)andthereisneedtosolveafewsetofequations.InSimplexMethod,the
primaryalgebraictaskofiterationistotransform,usingGauss-Jordanelimination,andthe
constraintequationsfromagivenconfigurationtoanewconfigurationthatcorrespondsto
thenextbasicfeasiblesolution.Suchtransformationsarerepeatedmanytimesinthecourse
ofthesolutionofaproblem(Carreira-Perpinan,2017).
1.1.5.2OptimizationMethodsInAnaerobicDigestionProcesses
AnaerobicDigestion(AD)processesaredescribedtobenon-linear,complexandunsteady;
thus,itischallengingtodevelopaprecisephysical-basedformulatocharacterizetheir
physicalperformance.ThedevelopmentofaccurateADprocessmodelscontinuestobaffle
expertsduetothenon-linearnatureofthebiochemicalnetworkinteractionsthatoccur
duringbioprocesses(Franco-Laraetal.,2006).Modellingandoptimizationofbioprocesses
contributestoincreasedunderstandingoftheprocessinputsforoptimumyieldand
productionrate.Variousmodellingalgorithmshavebeenappliedinbioprocesses,and
resultshaveshownthatmodellingandoptimizationcanenhancebiogasyields(Sewsynker-
Sukaietal.,2017).
Mathematicalandstatistical-basedmodelsprovidevitalinformationfortheunderstanding,
analysisandpredictionofbioprocessesandtheyarerequiredfortheoptimizationofkey
parametersinordertoimprovetheprocessoutput(Escamilla-Alvarado,Rios-Leal,&
Ponce-Noyola,2012;Yuetal.,2013).Thesebioprocessmodelscanprovideinsightonthe
individual as well as the interactive effect of the various input parameters on the target
output. Nevertheless, the non-linearities associated with microbial fermentations have
limited the use of these bioprocess models. Non-linear systems as opposed to linear systems are not standardized, which leads to deviations between the results obtained. The
implementation of bioprocess models that are able to efficiently encapsulate these non-
linearities is of paramount importance for optimization and scale-up of the bioprocess
(Ahmadian-Moghadam,Elegado,&Nayve,2013).
Traditionally,modellingandoptimizationofbioprocesseshavebeencarriedoutusingthe
one-variable-ata-timeapproach(OVAT),factorialDesignofExperiment(DOE)and
responsesurfacemethodologyRSM(Nath&Das,2011;Whiteman&GueguimKana,
2013).Theseapproacheshavebeenextensivelyusedandtheirconceptsaswellas
limitationsarewellknown.Forexample,OVATdoesnotconsidertheinteractiveeffectof
parametersontheprocessand,therefore,theoptimumsetpointsmaybecompletelyignored
(GueguimKanaetal.,2012;Wang&Lu,2005).Moreover,itisunfeasibleforthesearchto
accomplishanappropriateoptimuminarestrictedamountofexperimentalset-ups(Lotfy,
Ghanem,&El-Helow,2007).ThefactorialDOEhasbeenshowntobeunappealing,sinceit
istime-consuming,resourcedemandingandlabourintensivewhenthenumbersofinput
factorsareincreased(Wang&Wan,2009a).Ontheotherhand,theRSMdisregardsthe
‘lessimportant’parameterswithalimitedunderstandingoftheirpossibleinteractiveeffects
onthebioprocessoutput(GueguimKanaetal.,2012;Desaietal.,2008).
Artificialintelligencetoolshaveemergedasapromisingmethodformodellingand
optimizationofbioprocesses.Someoftheseincludeartificialneuralnetwork(ANN),
geneticalgorithm(GA),fuzzylogic,antalgorithmandparticleswarmoptimization,allof
whichareconsideredsuitableinthedesignofbioprocessesforresearchanddevelopment
(Sewsynker-Sukaietal.,2017).Inthelastdecade,ANNhasbeenappliedinmultivariate
non-linearbioprocessresearchanddevelopment.Theyareefficientforthedevelopmentof
bioprocessmodelsdevoidofpreviousinformationwithregardtothekineticsandmetabolic
fluxesthatoccurwithinthecellsandcellsurroundings(GueguimKanaetal.,2012).ANN
models simulate the linkage that exists in biological neurons with extraordinary capability for learning, analysis, association and adaptation. Some recent studies have shown that the
analysis using ANN yielded better results, in terms of R2 values, than the linear regression
method. Furthermore in the area of predictive accuracy test, the ANN has a higher accuracy
than multiple regression analysis. Other comparative studies show that ANN has higher
correlationcoefficientthantheresponsesurfacemethodology(RSM),henceabettermodel
forpredictionoftheexperimentaldata(Sewsynker-Sukaietal.,2017).
1.1.6 OptimizationUsingArtificialNeuralNetwork
ArtificialNeuralNetworksareabstractcomputationalmodels,roughlybasedonthe
organizationalstructureofthehumanbrain.TheattractivenessofANNscomesfromtheir
remarkableinformationprocessingcharacteristicsthatismainlypertinenttononlinearity,
highparallelism,faultandnoisetolerance,aswellastheirlearningandgeneralization
abilities,(Strukov,Snider,Stewart,&Williams,2008).ANNconsistsofaninputlayer,one
ormorehiddenlayersandanoutputlayer.Theneuronsofthehiddenlayerassistthe
networkinestablishingthecomplexassociationsthatsubsistbetweentheinputandoutput
parameters.Asshownschematicallyinfigure1.1,eachcircularnoderepresentsanartificial
neuronandanarrowrepresentsaconnectionfromtheoutputofoneneurontotheinputof
another(Graves,&Schmidhuber,2009b).
Figure1.1AschematicdiagramofanArtificialNeuralNetwork
An ANN is able to learn process input and output without underlying assumption about the
distribution of data. They are powerful in data processing and analysis and can be used for
optimization of processes, which are highly complex and nonlinear. ANN can be trained to
solve certain problems using a learning method and sample data. In this way, identically
constructedANNcanbeusedtoperformdifferenttasksdependingonthetrainingreceived.
Withpropertraining,ANNiscapableofgeneralization,andhastheabilitytopredict
analysis(Graves,&Schmidhuber,2009b;Kurzweil,2012).OptimizationsusingANNsare
concernedwiththeminimizationofaparticularcostfunctionwithrespecttocertain
constraints.ANNisshowntobecapableofhighlyefficientoptimizationduetoitsabilityto
capturenonlinearbehaviour,andthusprovidesamodelthatlinkstheprocessinputstothe
correspondingoutputparameters.(Ciresan,Meier,Masci,&Schmidhuber,2012b).
1.2 STATEMENTOFPROBLEM
Biogasreactorsarecomplexinstallations.AlmostallbiogasreactorsinNigeriahavefailed.
HumanwastebiogasplantatZariaprisonwasabandonedbecauseofthereactorfailure.The
reactorusedwasIMHOFFtank,whichisanimprovedversionofseptictankandisreported
tohaveexcessivesedimentationintheinletchannels.Italsohasexcessivefoulingofsurface
andfloatingofsludge.Acowdungbasedbiogasplantatthefodderfarmofthenational
animalproductionresearchinstitute,Zaria(NAPRI)isreportedtohaveexperiencedlower
gasyield.Thereactorusedistheupflowanaerobicsludgeblanket(UASB).TheUASBhave
inherentproblemwhenoperatedunderhydrodynamicconditionsandthereispossibilityof
ioninhibitionfromtheun-ionisedvolatilefattyacids(UVFA’s)whichmayleadto
disintegrationofthesubstrategranuleleadingtothedeathofmethanogensunderhydro
dynamic condition resulting to incomplete mixing between the granule and block waste. The
utilization of POME from palm oil mills for biogas production can prevent the harmful
environmental impacts associated with the discharge into the ecosystem. Several studies on POME treatment have been carried out using various high rate anaerobic reactors but these
reactor designs have at some point experienced their own fate of failures. Such failures are
due to complex biochemical processes which are often poorly understood and difficult to
control and run. However, as an important technology, the optimization of anaerobic
digestionprocessescanassistitsoperationandprocesscontrolaswellasmaximizing
methaneproduction.
Traditionally,theoptimizationofexperimentalprocessesisachievedbyrepetitionsof
experimentswhilevaryingonefactorafteranother.Modernmethodsforoptimizationapply
thesimplexproceduresandoftengivegoodresults.Theuseofsimulationasanoptimization
toolismoreadvantageous,asvariousmodellingformatsapplyanumberofexpressions
whichrequiresolutionstopolynomialequations.Thesemethodsareallstochasticbecause
theresultsaredeterminedabinitio.However,theuseofANNforsimulationandmodelling
arenon-stochastic;thisisbecausetheprogramusedallowstheusertopreselectthe
distributionwhichcanbemodifiedbythesystemdynamicstoreachatarget.Thistargetis
whatthesimulatorregardsastheoptimumdistribution.Inaddition,mathematicalmodelling
ofanaerobicdigestioncanbeverycomplexandbecauseofthecomplexitiesinvolved,the
digestionprocessisoftenmodelledasablack-box.ANNmodelscanbeusedtoovercome
someoftheproblemsofcomprehendingtheanaerobicdigestionprocesses.Theapproachof
ANNisparticularlyeffectiveformodellingprocesseswheretheunderlyingrelationships
betweeninputandoutputdataareunknown,complexandnonlinear.Otheroptimization
modellingsystemslikeregressionmodelling,linearprogramming,andgenericalgorithm,
requiresequations(mathematicalmodelling)butANNgivesadatadrivenmodelwhichis
much preferred because of its high degree of accuracy that is the closeness of the simulated
orpredictedoutputtothetargetwhichistheexperimentaloutput.
In view of these considerations, this research’s concern, which has been the challenge of
many researchers and other relevant stakeholders is, how can we model an experimental data
in order to remove their complexity, contain their non-linearity and unsteadiness in the
anaerobicbiogasgenerationfrompalmoilmilleffluentforanoptimizedbiogasyield?
1.3 RESEARCHAIMANDOBJECTIVES
Thisresearchaimsatdesigningamodelfortheoptimizationoffactorsaffectinganaerobic
biogasgenerationfrompalmoilmilleffluent(POME)usingartificialneuralnetwork.
Thespecificobjectivesforwhichtheresearchisbeingcarriedoutare:
- Toexaminerelationshipbetweenparameterssuchas%C,%N,CODreduction,
hydraulicretentiontime(HRT),pH,%S,%P,C/NRatio,temperature,organic
loadingrate(OLR)intermsofreactorefficiency.
- TodevelopamodelbasedonArtificialNeuralNetwork(ANN)foroptimizationof
biogasinanintegratedbioreactor.
- ComparisonofthepredictiveabilityusingANN,ofsimulatedmethaneyieldswith
experimentalmethaneyieldsoftheintegratedbioreactor.
- Todeterminethefunctionalrelationshipbetween%CODreduction,HRTandpHon
methaneyieldrespectively.
1.4 HYPOTHESIS OF THE STUDY.
Thehypothesisofthestudyisthetheoreticalbasiswhichhasmotivatedanddriventhis
research.ANNisoneoftheworld’sfastgrowingmodernmethodsusedintheanalysisof
data.Itisnon-stochasticandassuchcanbeusedincomplexcomputationalmodeltopredict
various experimental processes which are hard to solve by various conventional computer
based and programming techniques. The use of ANN as an optimization tool for biogas
generation is relatively new in Nigeria. The theory underlying this study is to apply ANN in modelling, simulation, prediction and interpreting of experimental data in order to obtain
useful information and enhance decision making. This work was designed to optimize the
methaneyieldofanintegratedbioreactorusingNeuralNetworkToolboxofMATLAB
R2015a and to determine the functional relationship between the % COD reduction and
biogasyield.
Secondly,themulti-layeredfeedforwardarchitectureusedtodesigntheANNmodelis
decidedbytestingdifferentnumbersofhiddenneuronsinthehiddenlayer.Levenberg-
MarquardtBackPropagationtrainingalgorithm,trainstheneuralnetworksgeneratinga
modelinformofMATLABcodeforprediction.Optimizationofatargetclosetotheoutput
validatestheperformanceoftheneuralnetworkmodel(Graves,&Schmidhuber,2009b;
Hinton,Osindero,&Teh,2006;Strukovetal.,2008;Ciresanet.al2012b).
Thirdly,anoptimizedbiogasgenerationsystemsupportstheefforttoassessthecarbontrade
exchangeasoperatedintheinter-governmentalpanelonclimatechange(IPCC,2012)since
NigeriaisasignatorytotheKyotoprotocols.Thereforeestablishingasystemofconversion
ofpalmoilmilleffluenttobiogasenergyeitherforheatingorelectricitygenerationwill
enablethecityorindustriesinvolvedtoearncarboncreditwhichinturncanactuallyoffset
anyextracostinvolvedinchangingfromfossilfuelenergygenerationtoaclean
developmentmechanism(Kyotoworkdatapoliciesinfrastructures,2007).
Fourthly,energyresourcesbasesuchaspalmoilmilleffluent,municipalorganicwaste,
domesticsewage,essentiallywillre-educatethesectionofthepopulationtoadapttocleaner
environmentandalsogeneratesustainablejobs.
1.5 SIGNIFICANCEOFSTUDY
Thestudypresentsanopportunitythathelpstofostereffectiveandefficientenvironmental
wastemanagementofPOME(Ejike,Chinedu&Egbujor,2012).Thereisneedtofinda
bettermethodforoptimizingchemicalexperimentsefficientlythroughtheoreticalbasis.The
core of a control/optimization system is the model describing the process. Despite several attempts to model the processes that occur in a digester, up to now classical mathematical
modeling was only possible when severe simplifications of the process representation were
performed (Sulaiman, Nikbakht, Tabatabaei, Khatamifar, & Hassan, 2010 ; Holubar et al.,
2003). The main reason for this situation is that the mechanisms ruling this processes are not
sufficientlywellunderstoodtoformulatereliablenon-linearmathematicalmodels.Asan
alternative,artificialneuralnetworkshaveadistinctiveadvantageoversomeothernon-
linearestimationmethodsusedforbio-processes,becausetheydonotrequireanyprior
knowledgeaboutthestructureoftherelationshipsthatexistbetweenimportantvariables
(Holubar et al., 2003).
Optimizationinvolvesvaryingmorethantwofactorssimultaneouslytherebyminimizingthe
errorontheyieldwhileidentifyingvalidrelationshipsamongthevariables.Consideringthe
hugeinvestmentcostrequiredforthesetupofbiogasplant,coupledwiththecomplexityof
workandtimespanningintomonths/yearsneededforestablishment,thedeploymentof
ANNhasbecomeaneffectiveandefficienttooltopromotethepredictionofproportionof
allneededparameters(factors)foroptimalyieldortarget.
1.6SCOPEOFTHESTUDY
Thestudyinvolvesthefollowingareasofactivity:
- PreliminaryStudiesonbiogasgenerationfrom
- DevelopingamodelbasedonArtificialNeuralNetwork(ANN)foroptimizationof
biogasinanintegratedbioreactor.
- Optimizingtheyieldofbiogasinanintegratedbioreactor,byapplying
- Determinefunctionalrelationshipbetweenthe%CODreduction,pHandHRTon
Biogasyieldrespectively.
These were performed by the application of ANN system embedded in MATLAB
software R2015amodel.
FACTOROPTIMIZATIONFORANAEROBICBIOGASGENERATIONFROM PALMOILMILLEFFLUENTUSINGARTIFICIALNEURALNETWORK