Design and planning of the bioethanol supplyDesign and planning of the bioethanol supplyDesign and planning of the bioethanol supply chain via simulation-based optimization:

Th f A ti

Design and planning of the bioethanol supply chain via simulation-based optimization:

Th f A tiThe case of ArgentinaThe case of ArgentinaDr. Guillermo A. Durand1

D F d D M l 2Dr. Guillermo A. Durand1

D F d D M l 2Dr. Fernando D. Mele2

Dr. Gonzalo Guillén-Gosálbez3

Prof. Dr. J. Alberto Bandoni1

Dr. Fernando D. Mele2

Dr. Gonzalo Guillén-Gosálbez3

Prof. Dr. J. Alberto Bandoni1

41st JAIIO – SII28/08/2012

La Plata, Buenos Aires

41st JAIIO – SII28/08/2012

La Plata, Buenos Aires

1PLAPIQUI (Universidad Nacional del Sur-CONICET), Camino La Carrindanga km7, Bahía Blanca, Argentina2FACET (Universidad Nacional de Tucumán), SM de Tucumán, Argentina3Universitat Rovira i Virgili, Tarragona, Spain

IntroductionIntroduction

Agentina’s National Law #26093 (2006)

ProvidesProvides thethe frameworkframework forfor investment,investment, production,production, andand marketingmarketingofof biofuelsbiofuels..

EstablishesEstablishes aa minimumminimum contentcontent ofof biofuelsbiofuels inin gasolinegasoline andand dieseldiesel..EstablishesEstablishes aa minimumminimum contentcontent ofof biofuelsbiofuels inin gasolinegasoline andand dieseldiesel..

Expansion needed!Expansion needed!Current situationCurrent situation**

EthanolEthanol fromfrom sugarcanesugarcaneEthanolEthanol fromfrom sugarcanesugarcane2323 sugarsugar millsmills120000120000tn/dtn/d processprocess capacitycapacity

onlyonly 33%% ofof contentcontent

2/14

Perez et al, (2011), Biocombustibles en la Argentina y Tucumán, cifras de la industria en el período 2009-2011, Reporte Agroindustrial, 52.

The Biofuel Supply ChainThe Biofuel Supply ChainKostin et al, (2012), Design and planning of infrastructures for bioethanol and sugar production under demand uncertainty, ChemEngRes, 90, 359-376.

SSUGARUGAR CANECANE EELECTRICITYLECTRICITYEETHANOLTHANOLOOTHERTHER UUSESSESEETHANOLTHANOLSSUGARUGAR CANECANE

SSUPPLIERUPPLIER

SSUGARUGAR MMILLILL DDISTILLERYISTILLERY

OOTHERTHER UUSESSESEETHANOLTHANOLWWAREHOUSEAREHOUSE

DDISTILLERYISTILLERY

SSUGARUGAR CANECANESSUPPLIERUPPLIER

SSUGARUGAR MMILLILL

SSUGARUGAR DDISTRIBUTIONISTRIBUTIONCENTERCENTER FFUELUEL EETHANOLTHANOL

SSUGARUGAR MMILLILL

DDISTILLERYISTILLERY

EETHANOLTHANOLWWAREHOUSEAREHOUSE

SSUGARUGAR CANECANESSUPPLIERUPPLIER SSUGARUGAR DDISTRIBUTIONISTRIBUTION

CENTERCENTER

3/14

EELECTRICITYLECTRICITY

CENTERCENTER

SSUGARUGAR

Simulation-based Optimization (SbO)Simulation-based Optimization (SbO)

START

G tinonofinalizationcriterion

yesyes

GeneticAlgorithm

nono

expected value of anexpected value of anobjective functionobjective function

ENDfirst stagefirst stagevariablesvariables

nominal value of thenominal value of theuncertain parametersuncertain parameters

objective functionobjective function

OptimizationModel

first stage +first stage +second stagesecond stage

variablesvariables

nono

Monte Carlo

generatorsampled values of thesampled values of theuncertain parametersuncertain parameters

SimulationModel

finalizationcriterion

eses

4a/14

yesyes

Simulation-based Optimization (SbO) – The Biofuel Supply ChainSimulation-based Optimization (SbO) – The Biofuel Supply Chain

G ti

START

nonoMatlab R2007b/

GA Toolboxfinalizationcriterion

expected value of anexpected value of anobjective functionobjective function

GeneticAlgorithm

yesyes

nono GA Toolbox

SC design variables• No. and type of production

% demandsatisfaction

first stagefirst stagevariablesvariables

nominal value of thenominal value of theuncertain parametersuncertain parameters

objective functionobjective function

ENDMILP Model

GAMS/CPLEX 11 2

yp punits

• No. and type of storage units• No. and type of

transportation units• Capacities

sugar and ethanol demands

satisfaction

first stage +first stage +second stagesecond stage

variablesvariables

OptimizationModel

GAMS/CPLEX 11.2 • Capacities• Expansions

SC planning variables• Amount of products to be produced and

storedFl f t t d t i l

sampled values of thesampled values of theuncertain parametersuncertain parameters

Monte Carlo

generator nono

Matlab R2007b

Matlab R2007bsugar and ethanol

demands

• Flows of transported materials• Product sales

finalizationcriterion

SimulationModel

eses

Matlab R2007b

yesyes

4b/14

Supply Chain MILP ModelSupply Chain MILP Model

SetsSets VariablesVariablestt periodsperiods (year)(year) CFCFtt cashcash flow of period flow of period ttii materialsmaterials DTSDTSigtigt inventoryinventory of material of material ii on storage on storage ss on region on region gg on period on period

ttIMIM subset of subset of ii, raw materials, raw materialsgg regionsregions NPVNPV supply chain total net present valuesupply chain total net present valuegg regionsregions NPVNPV supply chain total net present valuesupply chain total net present valuess storagestorage technologiestechnologies PCapPCappgtpgt production capacity ofproduction capacity of technology technology pp on region on region gg on on

period period ttISIS subset ofsubset of ss, storage technologies for , storage technologies for

material material iiPCapEPCapEpgtpgt expansion of production capacity ofexpansion of production capacity of technology technology pp on on

region region gg on period on period ttpp prod ction technologiesprod ction technologies PEPE materialmaterial ii prod cedprod ced b technologb technolog pp on regionon region gg ononpp production technologiesproduction technologies PEPEipgtipgt material material ii producedproduced by technology by technology pp on region on region gg on on

period period ttii transportation typestransportation typesILIL subset ofsubset of ii, transportation types for , transportation types for

material material iiPTPTigtigt material material ii producedproduced on region on region gg on period on period tt

ParametersParameters PUPUigtigt material material ii purchasedpurchased on region on region gg on period on period ttρρpipi mass balance coefficientsmass balance coefficients QQilgg’tilgg’t flow of material flow of material ii from regionfrom region g to region g to region gg’ by transport ’ by transport ll

on period on period ttττpgpg minimumminimum desired fraction of desired fraction of

technology technology pp on region on region ggSCapSCapsgtsgt storage capacity ofstorage capacity of technology technology ss on region on region gg on period on period tt

CapCropCapCropgtgt maximun sugarcane’s production onmaximun sugarcane’s production on STSTisgtisgt inventoryinventory of materialof material ii on storageon storage ss on regionon region gg on periodon periodCapCropCapCropgtgt maximun sugarcane s production onmaximun sugarcane s production onregion region gg on period on period tt

STSTisgtisgt inventoryinventory of material of material ii on storage on storage ss on region on region gg on period on period tt

NPNPpgtpgt number of number of production units of production units of technology p on region g on period t technology p on region g on period t (first stage variable)(first stage variable)

SDSD productproduct i’i’s demand on regions demand on region gg ononSDSDigtigt productproduct ii s demand on region s demand on region gg on on period period t t (sampled)(sampled)

5/14

Supply Chain MILP ModelSupply Chain MILP Model

MM b lb l t i tt i tMassMass balancebalance constraintsconstraints

CapacityCapacity constraintsconstraints

6/14

SbO of the Argentina’s Biofuel Supply ChainSbO of the Argentina’s Biofuel Supply Chain

UncertainUncertain parametersparameters werewere sampledsampled consideringconsidering aa perturbationperturbation withwith aaprobabilityprobability distributiondistribution ofof NN((11,,3030%%)) forfor thethe sugarsugar demand,demand, andand NN((11,,55%%)) forforthethe ethanolethanol demanddemand..

TheThe timetime horizonhorizon forfor thethe designdesign andand planningplanning waswas setset atat 33 yearsyears..

TheThe finalizationfinalization criterioncriterion forfor thethe innerinner looploop waswas forfor thethe MonteMonte CarloCarlogeneratorgenerator toto taketake 100100 samplessamples..

TheThe finalizationfinalization criterioncriterion forfor thethe outerouter looploop waswas toto testtest 4040 generationsgenerations ofof100100 individualsindividuals eacheach..

Obj tiObj ti f tif ti ff thth tt llObjectiveObjective functionfunction ofof thethe outerouter looploop::

IndividualsIndividuals withwith aa negativenegative objectiveobjective functionfunction werewere assignedassigned aa valuevalue ofof 00..

7/14

SbO solution statisticsSbO solution statistics

PopulationPopulation 100100

GenerationsGenerations 4040

Objective function (best individual)Objective function (best individual) 82.17%82.17%

NPV (best individual)NPV (best individual) $227 7 millions$227 7 millionsNPV (best individual)NPV (best individual) $227.7 millions$227.7 millions

Best individualBest individual found at generationfound at generation 3838

TriedTried combinationscombinations 100x41 = 4100100x41 = 4100

Unique combinationsUnique combinations 3975 (16273975 (1627 not valid)not valid)

Average CPU time per MILP solvingAverage CPU time per MILP solving** 0.8460.846 secondsseconds

Total CPU timeTotal CPU time** 61 minutes 1.47 seconds61 minutes 1.47 seconds* SbO* SbO run on a Pentium D 945 desktop PC with 1GB of RAMrun on a Pentium D 945 desktop PC with 1GB of RAM

Average number of uncertain parameters per runAverage number of uncertain parameters per run 7070Average number of uncertain parameters per runAverage number of uncertain parameters per run 7070

8/14

SbO solution statisticsSbO solution statisticsBest and worst individual and average of each generationBest and worst individual and average of each generation NPV of the best individual of each generationNPV of the best individual of each generation

60

70

80

90

on

Best and worst individual, and average of each generation

60

70

80

90

on

Best and worst individual, and average of each generation

200

250NPV of the best individual of each generation

200

250NPV of the best individual of each generation

30

40

50

60

% o

f de

man

d sa

tifac

tio

30

40

50

60

% o

f de

man

d sa

tifac

tio

100

150

Mill

ions

of

$

100

150

Mill

ions

of

$

0 5 10 15 20 25 30 35 400

10

20

%

Best

AverageWorst

0 5 10 15 20 25 30 35 400

10

20

%

Best

AverageWorst

0 5 10 15 20 25 30 35 400

50

0 5 10 15 20 25 30 35 400

50

GenerationGeneration GenerationGeneration

90

100Number of non-feasible individuals of each generation

90

100Number of non-feasible individuals of each generation

50

60

70

80

50

60

70

80

10

20

30

40

10

20

30

40

9/14

0 5 10 15 20 25 30 35 400

Generation0 5 10 15 20 25 30 35 40

0

Generation

SbO Results – Best solution – Year 1SbO Results – Best solution – Year 1

22

11

22 1

7

8

9

ear]

Installed Production Capacity

7

8

9

ear]

Installed Production Capacity

12

2

11

2

22

3

4

5

6

al capacity [105Tn/ye

Mill T1

Mill T2

Mill T3

Dist. T13

4

5

6

al capacity [105Tn/ye

Mill T1

Mill T2

Mill T3

Dist. T1

1

2 1

1

11

0

1

2

1 2 3

Tota

Year

Dist. T2

0

1

2

1 2 3

Tota

Year

Dist. T2

5

6

Installed Storage Capacity

5

6

Installed Storage Capacity

12

2

3

4

l capacity [104Tn]

Storage T1

Storage T22

3

4

l capacity [104Tn]

Storage T1

Storage T2

2

292 Heavy trucks(18980Tn)

221 Medium trucks

0

1

2

1 2 3

Tota

Year0

1

2

1 2 3

Tota

Year

221 Medium trucks(5525Tn)

68 Tank trucks(1904Tn)

10/14

SbO Results – Best solution – Year 2SbO Results – Best solution – Year 2

7

8

9

ear]

Installed Production Capacity

7

8

9

ear]

Installed Production Capacity

22

11

22 2

3

4

5

6

al capacity [105Tn/ye

Mill T1

Mill T2

Mill T3

Dist. T13

4

5

6

al capacity [105Tn/ye

Mill T1

Mill T2

Mill T3

Dist. T1

22

2

11

2

22

0

1

2

1 2 3

Tota

Year

Dist. T2

0

1

2

1 2 3

Tota

Year

Dist. T2

1

2 1

1

22

12

5

6

Installed Storage Capacity

5

6

Installed Storage Capacity

2

292 Heavy trucks(18980Tn)

221 Medium trucks 2

3

4

l capacity [104Tn]

Storage T1

Storage T22

3

4

l capacity [104Tn]

Storage T1

Storage T2221 Medium trucks

(5525Tn)69 Tank trucks

(1932Tn)0

1

2

1 2 3

Tota

Year0

1

2

1 2 3

Tota

Year

11/14

SbO Results – Best solution – Year 3SbO Results – Best solution – Year 3

7

8

9

ear]

Installed Production Capacity

7

8

9

ear]

Installed Production Capacity

22

11

22 2

3

4

5

6

al capacity [105Tn/ye

Mill T1

Mill T2

Mill T3

Dist. T13

4

5

6

al capacity [105Tn/ye

Mill T1

Mill T2

Mill T3

Dist. T1

22

2

11

2

22

0

1

2

1 2 3

Tota

Year

Dist. T2

0

1

2

1 2 3

Tota

Year

Dist. T2

1

2 1

1

22

12

5

6

Installed Storage Capacity

5

6

Installed Storage Capacity

2

292 Heavy trucks(18980Tn)

221 Medium trucks 2

3

4

l capacity [104Tn]

Storage T1

Storage T22

3

4

l capacity [104Tn]

Storage T1

Storage T2221 Medium trucks

(5525Tn)69 Tank trucks

(1932Tn)0

1

2

1 2 3

Tota

Year0

1

2

1 2 3

Tota

Year

12/14

ConclusionsConclusions

AA SbOSbO strategystrategy hashas beenbeen implementedimplemented toto solvesolve thethe problemproblem ofof optimaloptimal designdesignofof thethe sugar/ethanolsugar/ethanol SCSC inin ArgentinaArgentina underunder parametricparametric uncertaintyuncertainty..TheThe strategystrategy isis aa twotwo levellevel optimizationoptimization frameworkframework thatthat combinescombines MILPMILP solvingsolvingwithwith MonteMonte CarloCarlo simulationsimulation andand GeneticGenetic AlgorithmsAlgorithms..ThTh dd f kf k h dlh dl dd 7070 t it i tt (th(thTheThe proposedproposed frameworkframework handleshandles aroundaround 7070 uncertainuncertain parametersparameters (the(theproductsproducts demandsdemands forfor eacheach regionregion andand timetime period)period) andand geographicallygeographicallydistributesdistributes production/production/ storage/storage/ distributiondistribution nodesnodes inin ArAr--gentina,gentina, consideringconsideringdifferentdifferent technologiestechnologiesdifferentdifferent technologiestechnologies..TheThe modelmodel decidesdecides thethe production/storageproduction/storage capacitiescapacities ofof thethe nodes,nodes, thethe quantityquantityofof transporttransport unitsunits andand thethe periodperiod inin whichwhich eacheach installationinstallation and/orand/or expansionexpansionshouldshould bebe mademadeshouldshould bebe mademade..TheThe proposedproposed SbOSbO strategystrategy combinescombines twotwo objectiveobjective functions,functions, CSatCSat andand NPV,NPV,thusthus allowingallowing increasingincreasing bothboth althoughalthough itit waswas expectedexpected themthem toto bebe oppositeopposite..

13/14