FUEL ETHANOL FROM BRAZILIAN BIOMASS: ECONOMIC RISKANALYSIS BASED ON MONTE CARLO SIMULATION

TECHNIQUES

Nøvaes1, W. S.; Caixeta, S. H. M; Selvam, P. V. P.; Santos, H. P. A.; Costa, G. B.Universidade Federal do Rio Grande Norte – UFRN, Núcleo de Tecnologia – NT

Programa de Pós – Graduação em Engenharia Química – PPGEQGrupo de Pesquisa em Engenharia de Custos e Processos – GPEC

Campus Universitário – Lagoa Nova, CEP: 59072 – 970 – Natal/RNHome Page: www.ufrngpec.hpg.com.br; E – Mail: alquimista@eq.ufrn.br1

1 Author to whom correspondence should be addressed.

ABSTRACT

At the present the fuel ethanol fromBrazilian biomass is one of the most favorableenergetic project in development to the Brazilmainly with respect to the rural employment,environment and energetic safety. In thiscontext, a technological and economical studywas developed applied to fuel ethanolproduction from sugar cane based on dynamicmodeling, simulations and economical riskanalysis. The objective of this work concern todevelop an stochastic methodology and as wellas the implementation of the economic riskanalysis inherent to the ethanol processproduction by the changing of the processvariables and parameters based on MonteCarlo simulation method. The mass balanceand economic evaluation was obtained fromsimulations were carried out using theSuperPro DesigneTM v. 3.0 software. Inaddition to this, the economic risk analysis wascarried out based on changing in thedistribution of the selling price, yield offermentation and feed flow rate the processstochastic variables. The stochastic analysisusing Monte Carlo simulation wasimplemented in a spreadsheet model utilizing@Risk v. 4.5 software for Excel. Based ondeterministic and stochastic simulations of thefuel ethanol production the results saveraltechn economical parameters have beenobtained and analyzed. The importantadvantage of the proposed method is thepossibility to predict the economical risk

involved in the ethanol production take in toaccount the aleatory and complex natureinherent to the process, with precision,reliability and very rapidly.KEY WORDS: Ethanol, Simulation,Stochastic, Monte Carlo, Risk Analysis.

RESUMO

O etanol derivado da biomassa brasileiraconstitui hoje um dos projetos energéticos maispropícios em desenvolvimento para o Brasil,principalmente no que diz respeito ao empregorural, ao meio ambiente e a segurança energética.Pensando nisso, desenvolvemos um estudotecnológico e econômico sobre a produção deetanol derivado da cana – de – açúcar, utilizandomodelagem dinâmica, simulação e análise de riscoeconômico através de métodos estocásticos. Oobjetivo do presente trabalho consiste emdesenvolver um modelo estocástico para asimulação e também a implementação da análise derisco econômico envolvido no projeto, devido àvariação de parâmetros e variáveis de processoatravés do método Monte Carlo de simulaçãoestocástica. O balanço de massa e a avaliaçãoeconômica foram obtidos através de simulações,utilizando para isso o software SuperProTM

Designer v.3.0. Realizou-se, também, a análise derisco econômico variando o preço de venda,rendimento de fermentação e a capacidade deprodução, consideradas como variáveis estocásticasdo processo. Desenvolvemos a simulaçãoestocástica utilizando planilha eletrônica baseadaem métodos de distribuição de incerteza e osoftware @Risk, simulador Monte Carlo. Atravésda simulação do processo de produção de etanolcombustível, foram obtidos como resultados os

investimentos necessários, os lucros e o retornosobre os investimentos. Pelos resultados obtidos,podemos concluir que por meio de nossassimulações e análise de tais resultados, que atravésdo método Monte Carlo é possível realizar umamodelagem considerando a natureza aleatória ecomplexa inerente a cada processo químico, comrapidez, precisão e confiabilidade dos resultadosobtidos.

PALAVRAS CHAVES: Etanol combustível,Simulação, Estocástica, Monte Carlo, Análise deRisco.

INTRODUCTION

The increasing capability of computerprocessing has facilitated the development and useof software to simulate chemical and othersindustrial process. The procedure of processsimulation consist in the resolution of the set ofmathematical relations that describe systembehaviour, for example, mass and energy balance,phase equilibrium equations, physical propertiescalculations, etc. Where as such mathematicalmodel contains input variables, process parametersand output variables. The traditional simulationapproach involve definition on input variables andprocess parameters, thus output variables areevaluated by use of the model, which producesdeterministic (point estimate) results for umparticular set of assumptions. (MIZUTANI et al,2000)

Fuel ethanol production industries areusually faced with uncertain conditions during theiroperation. These uncertain can arise from variationseither in external parameters, such as the quantityof feed streams, temperature and concentration, orinternal process parameters such transfercoefficients, reactions constants, physical propertiesand as well as price of fuel ethanol because ofvariation in fuel price market. If the technology isnew, there are additional uncertainties due tolimited performance data. Thus, utilisation of thedeterministic simulation results can be limited dueto the uncertainties associated with the variablesand parameters in the real process and the solutionof the model can be significant from the realprocess response. Therefore, an efficient andsystematic method is required to evaluateuncertainties and investigate the effect of theuncertainties in chemical process simulation.

An approach tool widely used foruncertainty analysis employs the concept ofstochastic analysis simulation (DIWEKAR andRUBIN, 1991). This approach is based on a largenumber of simulation runs, in each run, inputvariables and process parameters are randomlyselected according to adequate probability densityfunctions. The set of out put variables evaluated

gives important information about the behaviour ofthe statistical process, indicating the most probablevariable range.

In short, the process analysis of realsystems requires both stochastic and deterministicmodelling capabilities. Yet, not much works havebeen realised in Brazil about stochastic modellingapplied to fuel ethanol production even thoughBrazil is the leader in this field.

Thus, the objective of this paper is todevelop a generalised stochastic modellingmethodology applied to uncertainties analysisinherent to the fuel ethanol process productionaided for deterministic simulator and stochasticspreadsheet models based on Monte Carlosimulation techniques.

DETERMINISTIC ANALYSIS

The modelling tools in current simulatorsmay roughly classified into two groups, block-oriented (or modular) and equations-orientedapproaches. (MARQUARDT, 1996)

Block-oriented approach mainly addressesmodelling on the flowsheet level. Every process isabstracted by a block diagram consisting ofstandardised blocks, which model behaviour of aprocess unit or part of it. All the blocks are linkedby single-like connections representing the flow ofinformation, materials and energy employingstandardised interface and stream formats.

Models of process unit are preceded byone model expert and incorporated in a modellibrary for later use. Modelling on the flowsheetlevel is either supported by one modelling languageor by a graphical editor. In both cases, the end userselect the models from the library, provides themodel parameters and connects them to the plantmodel. The incorporated chemical engineeringknowledge as well as the model structure arelargely fixed and not accessible. Commonexceptions are physical property models, which canbe selected independently of the process unitmodel.

Equation-oriented modelling tool supportthe implementation of unit models and theirincorporation in a model library by means ofdeclarative modelling language or by providing aset of subroutine templates to be complete directoryin a procedural programming language. There areno different tools for modelling expert or for theend user. Hence, modelling on the unit levelrequires profound knowledge in such diverse areasas chemical engineering, modelling and simulation,numerical mathematics, and computer science. Thedevelopment of novel process unit models istherefore often restrict to a small group of expert.

Theses later approaches do not manyadvantage compared to the modular modelling toolmeanly in the implementation complexity because

the equation-modular approaches requires a highcomputational and technical knowledge.

The SuperPro DesignerTM v.3.0 softwarehas been the computational tool used in this study.The SuperPro Designer is part of the block-oriented simulator group. It is comprehensivecomputer simulations package the fate of chemicalsin individual and combinations of unit operation. Alist of these chemicals that is global to the processflowsheet is specified from an extensive databaseprovided with the software. User-defined chemicalscan be added of this database. The process diagramis built in a “point and click” CAD environment byadding unit operation to a flowsheet andestablishing the connectivity between these unitsusing streams. The chemical components of thestreams that serve as influent to the whole processare defined. (INTELLIGEN, 1994 and 1998)

The consecutive unit operations are solvedusing sequential mass balances where the effluentstreams from one unit operation serves as theinfluent streams to another unit operation.(INTELLIGEN, 1994 and 1998)

Results of the simulation can be viewed byclicking directly on stream or unit operation on thescreen. Comprehensive output can be obtained bygenerating stream and input data report that can beviewed in separating windows. (INTELLIGEN,1994 and 1998)

Beyond mass balance, SuperPro Designerhas the capacity to perform economical analysis toevaluate the capital and operating costs of differentflow diagrams, and the capacity to schedule batchoperations. The steps to take place onedeterministic simulation approach using theSuperPro Designer are the following.(INTELLIGEN, 1998)

1. Draw out the flowsheet;2. Define the components;3. Select the thermodynamic calculation method;4. Define the feed streams;5. Provide process conditions for the unit

operation;6. Run the simulation.

STOCHASTIC ANALYSIS

In general, uncertainty analysis requiresstochastic modelling environments. Some papers(LEE et al, 1996) classify the Monte Carlo (MC)analysis like one of the most widely used approach,which implies two steps of applications. One isdistribution sampling from one distribution toestimate other distribution. The other is using thedistribution sampling technique to solve aprobabilistic problem, which is generated from adeterministic problem. This approach has beensuccessfully applied to calculation of complexintegrals, solutions of equations, etc. the advantage

of this approach is flexibility. First, itsimplementation is very easy, and modification ofmodel is not necessary. Second, variousindependent and dependent variables are consideredsimultaneously. Third, various numerical andgraphical technique can be used for analysis ofrelationship between the model inputs and outputs.

Generally, simulation based on MonteCarlo simulation analysis is performed through thefollowing six steps.

1. Functional or mathematical model is defined;2. Probability distribution function are allocated

to design and/or input variables;3. Random samples are generated for each

uncertain variable;4. Sets of data on the system output are obtained

by repeated model calculations;5. Statistical estimates are calculated from the

data;6. Sensitivity analysis of input and output

variables is performed.

In this work the stochastic analysis basedon Monte Carlo simulation was carried out in aspreadsheet model created in MS Excel 2002environment utilising @risk software.

STOCHASTIC SIMULATIONMETHODOLOGY

The proposed simulation method usesdeterministic models without modification. In thiswork, models in a general-purpose processsimulator, SuperPro DesignerTM v. 3.0, weredirectly used.

The first step of stochastic simulation is todetermine the range and distribution of theuncertain variables. This step consists of thefollowing procedures.

1. Basic assessment of uncertainties of theprocess;

2. Selection of uncertain variables;3. Gathering information for selected variables;4. Estimation of uncertainties based on available

information;

Various probability distributions can beused to represent uncertainties of the variables.Selection of a probability function for an uncertainvariable depends on the characteristics of thevariables and the amount of variable information onthe uncertainties of the variable. The mostfrequently distribution function are uniformdistribution, normal distribution, and triangulardistribution. Uniform distribution is need wheninformation is poor and only the limiting values areknown. As it has no central tendency, theuncertainties result in broad distribution of the

values of the output variables. Normal distributionis known as a widely applicable distribution instatistics. Triangular distribution is used when hecentral and limiting values are known.

In this work uniform distribution was usedbased on factorial experimental planning techniques(PANNIR e al, 1998). With these experiments wascarried out eight simulation utilising SuperProDesigner. The Table 1. shows the factorialexperimental planning utilised.

Table 1 – factorial experimental panning techniques.

Run V1 V2 V3 VOutput

1º Run + + + ?2º Run + + - ?3º Run + - + ?4º Run + - - ?5º Run - + + ?6º Run - + - ?7º Run - - + ?8º Run - - - ?

V is an input variable or parameter and the pluscodification represent the maximum value and theminus codification represent minimum valueassumed for the variable or parameters.

After carrying out the set of simulationproposed in the Table 1, the second step is toconduct a regression analysis to determine a modelwhich is the best fit to the range of input parametersand variables, and output variables. In the present

study the equation 1 was used. The regressionanalysis is the most important step in order tointegrate the deterministic flowsheet model to thespreadsheet model.

)1(........239

228

217326315214

3322110

VbVb

VbVVbVVbVVb

VbVbVbbVOutput

++

+++++

++++=

Where,

b0, b1, … b9 are regression parameters;V1, V2 and V3 are independent variables;VOutput is dependent variable.

The third step is to generate samplesaccording to the selected probability distributionfunction

The fourth step is the propagation ofuncertainties through the model. This stepcomprises samples allocation to the model input,model calculation, and storage of the calculationresults.

Finally, the fifth step is the uncertaintyanalysis of the simulation results. Uncertainties ofthe output variables are represented by moments ofdistribution. The distribution of the outputvariables can be described by various uncertaintydisplays such as cumulative distribution function,probability density function, box plot, bar graph,etc. The figure 1 shows the basic operation of thestochastic simulation method.

Figure 1 – Basic operation of the stochastic simulation method.

Flowsheet (Deterministic Level)

OutputVariables and

Parameters

FactorialExperimental

Planning

MathematicalModel

Spreadsheet Model (Stochastic Level)

Output UncertaintiesStochastic Simulation(Monte Carlo Method)

InputUncertainties

CASE STUDY

Based on data obtained in each step ofethanol process production and, physical andthermodynamic properties of the input and outputstream simulations was carried out for Brazilianmedium scale distillery with conventional processand the performance and applicability of thedeveloped stochastic simulation methodology wastested. The figure 2 shows the simplified schematicof the conventional process of the fuel ethanolproduction from sugar cane (JOSHI and PANDEY,1999). The objective is to conduct an economicalrisk analysis evaluation due of the effect ofuncertainties of the input variables. The simulatedsituation is to predict the behaviour of the annualprofit (VOutput) based on change of the distributionof feed flow rate (V1), yield of fermentation (V2)and selling price (V3), base on uniform distributionwith minimum 38,460 kg/h and maximum 153,846kg/h, minimum 84% and maximum 95%, minimum0.16 R$ and 0.41 R$, respectively. Thisexperimental factorial planning is illustrated in thetable 2 represented in the columns 2, 3 and 4.

Base on simplified process flow diagram

of fuel ethanol production proposed by (JOSHI andPANDEY, 1999) out lined in the figure 2 thedeterministic simulation using SuperPro DesigneTM

v. 3.0 software has been carried out. Where,

1. First utilising the software databank, thecomponents used in the process was defined;

2. The equipment required was selected and Drawout the flowsheet (Figure 3.);

3. The input streams and process parametersinformation have been specified;

4. After the conduct the three steps above, theprocess simulation was conducted and themass Balance, energy balance and outputstreams information was automatically solved;

5. Now, with the changing of the stochasticparameters based on the experimental factorialplanning out lined in the table 2 the eightsimulations runs proposed was take placed, thetable 2 in the columns 5 shows the economicalresult;

6. Finally, costs, mass and energy balancesreporters, equipment sizing, output flowstreams and composition, and final processflowsheet have been obtained.

.

Figure 2 – Simplified process flow diagram of fuel ethanol production (JOSHI and PANDEY, 1999).

Sugar Cane Milling

Bagasse

Pre

Treatment

ResiduesSterilization

FermentationCentrifugation

Yeast

Distillation

Yeast CO2

Storage

Stillage

Storage Fuel Ethanol

Simplified Process Flow Diagram of Fuel Ethanol Production

Figure 3 – Ethanol production flowsheet.

Table 2 – Results from deterministic simulation utilisingexperimental factorial planning.

Run V1 V2 V3 VOutput

1º Run 153,846 95 0.41 24,9352º Run 153,846 95 0.16 22,8243º Run 153,846 84 0.41 15,8174º Run 153,846 84 0.16 14,7385º Run 38,460 95 0.41 4,5636º Run 38,460 95 0.16 4,1927º Run 38,460 84 0.41 1,9788º Run 38,460 84 0.16 1,799

Where,

V1: feed flow rate, (103)kg/h;V2: yield fermentation, %;V3: selling price, R$;VOutput: annual net profit, R$.

Based on data eight data point from table 2was determined the set of regression parameterswhich better fit to the equation 1. The table 3 showsresults from non-linear regression analysis.

Table 3 – Set of parameters model

b0 = -463065,48 b5 = 45,758605b1 = -652,84371 b6 = 40800106b2 = -1142522,9 b7 = 0,0000735b3 = -27282863 b8 = -1701621,5b4 = 882,97289 b9 = -15566709

Based on the Monte Carlo method andstatistical distribution of the input variables aspreadsheet model was developed using @Risk v.5.0 software and the uncertainties were propagatedthrough the model obtained from regressionanalysis utilising Statistics v. 5.0 software. In this

case study, ten thousand simulation runs werecarried out and for each groups sampling wasrealised one hundred iteration runs.

Stochastic Simulation

0

5000000

10000000

15000000

20000000

25000000

30000000

0 50 100 150 200

Runs

An

nu

al n

et P

rofi

t (R

$)

Figure 4 – Stochastic simulation of ethanol production fromsugar cane.

The analysis of the results show that thesimulation was successful indicating the extent ofuncertainties of the output variable with in tolerablelimits of real values as shown the first two hundredsimulations runs in the figure 4 and the trend of thechange in these uncertainties using normaldistribution was also identified. The nature of theoutput variable studied was observed and theirprobabilistic distribution are listed in the table 4.

Table 4 – Stochastic simulation analysis.

Profit Range(R$/year) Probability (%)0,0000 < Profit ≤ 5x106 10,925x106 < Profit ≤ 1x107 30,931x107 < Profit ≤ 15x106 28,5315x106 < Profit ≤ 2x107 22,882x107 < Profit ≤ 25x106 6,73025x106 < Profit ≤ 3x107 0,010

CONCLUSIONS

A methodology for stochastic analysis,using statistical properties of process variables andparameters, is presented. A case study of anindustrial chemical process application wasconducted successfully and analysed. Theconclusions are summarised as follows.

1. Conventional simulators do not take intoaccount the uncertainties of the real process.Thus, there is no way to investigate the effectof the uncertainties in this chemical processsimulation;

2. The uncertainties analysis based on stochasticsimulation have been discussed in this workusing a model to integrate both deterministic

and stochastic simulation. This model isimplemented for fuel ethanol production.

3. In stochastic analysis, the distribution used inMonte Carlo simulation is derived from inputvariables and parameters, therefore, the resultsdepend only on process behaviour instead ofthe subjective judgement;

4. The results of the case studies indicate thatstochastic simulation is a powerful tool toevaluate the behaviour of the fuel ethanolproduction which has various changinguncertainties for several variables;

5. The proposed method can be effectivelyextended to treat uncertainties of fuel ethanolprocess in small, medium end large scale ofproduction;

6. This simulation work developed in the presentstudy can predict the probability of theviability of the fuel ethanol process fromBrazilian biomass and its survive withdifferent level of fuel ethanol selling price.

REFERENCES

[1]CHANG, C. P.; Lin, Z. S.; Stochastic Analysisof Decline Data for Production Predictionand Reserves Estimation, Journal ofPetroleum Science and Engineering, v. 23, p.149 – 160, 1999.

[2]DIWEKAR, U.M.; RUBIN, E.S. StochasticModelling of Chemical Processes,Computers Chem. Engng, v. 15, No. 2, p. 105– 114, 1991.

[3]INTELLIGEN, INC.; Bio Pro Designer: AnAdvanced Computing Environment forModelling and design of IntegratedBiochemical Process, Computers Chem.Engng, v. 18, Suppl, p. S621 – S625, 1994

[4]INTELLIGEN, INC.; Users Guide’s for the Pro-Designer (BioPro, EnviroPro, and SuperProDesigner) Family of Simulation andDesigner Tools for the Process andEnvironment Industries, Intelligen Inc., ScottPlains, NJ, 1998.

[5]JOSHI, V. K.; PANDEY, A.; Biotechnology:Food Fermentation, Vol. II: Applied,Educational Publishers & Distributors, NewDelhi – India, p. 1201 – 1230, 1999.

[6]LEE, K. L.; LEE, K. J.; CHOI, S. H.; YOON, E.S. Stochastic Dynamic Simulation ofChemical Process with ChangingUncertainties, Computers Chem. Engng, v.20, Suppl, p. S557 – S562, 1996.

[7]MARQUARDT, W. Trends in Computer-AidedProcess Modelling, Computers Chem.Engng, v. 20, No. 6/7, p. 591-609, 1996.

[8]MIZUTANI, F. T.; COSTA, A. L. H.; PESSOA,F. L. P. Stochastic Simulation of

supercritical Fluid Extraction Processes,Braz. J. Chem. Eng, v. 17, No. 3, 2000.

[9]PANNIR P. V.; WOLF, D. M. B.; MELO,H.N.S. Process, Cost Modelling andSimulation for Integrated ProjectDevelopment of Biomass for Fuel and rotein,J. Scien. Indus. Res., v.57, p. 567-574, 1998.

ACKNOWLEDGMENTS

The authors would like to thanks fromCNPq/PIBIC, PPGEQ/DEQ/CT/UFRN andANP/PRH – 14 for the financial support andlaboratories facilities.