2-Artificial Neural Network Models for Biomass

lss

b,

ns

cion

Article history:

modern use of biomass is considered a very promising clean

in isolated installations or in developing countries. In addi-

tion, it is the only renewable energy source that can directly

replace fossil fuels as it is widely available and allows con-

tinuous power generation and synthesis of different fuels and

chemicals.

oils). Air, steam or oxygen can be supplied to the reaction as

ed varies according

erating conditions

ulate biomass gas-

ification process for scale-up, industrial control strategies,

performance calculation after modifying the operating con-

ditions, etc. Mathematical models aim to study the thermo-

chemical processes during the gasification of the biomass and

to evaluate the influence of the main input variables on the

* Corresponding author. Tel.: 34 977257068; fax: 34 [email protected] (J.C. Bruno).

Available online at www.sciencedirect.com

.co

b i om a s s a n d b i o e n e r g y 4 9 ( 2 0 1 3 ) 2 7 9e2 8 9E-mail addresses: [email protected] (M. Puig-Arnavat), juancaenergy option for reducing energy dependency and green-

house gas emissions; biomass is considered to be CO2-neutral.

Biomass gasification can be considered in advanced applica-

tions in developed countries, and also for rural electrification

gasifying agents. The quality of gas produc

to the gasifying agent used and the op

selected.

Consequently, it is necessary to sim1. Introduction

Biomass gasification is a highly efficient and clean conversion

process that converts different biomass feedstocks to a wide

variety of products for various applications. In this context,

Gasification conversion process can be defined as a partial

thermal oxidation, which results in a great proportion of

gaseous products (carbon dioxide, hydrogen, carbon monox-

ide, water and other gaseous hydrocarbons), little quantities

of char, ash and several condensable compounds (tars andReceived 4 April 2012

Received in revised form

16 November 2012

Accepted 10 December 2012

Available online 28 January 2013

Keywords:

Biomass

Gasification

Artificial neural network

Simulation

Fluidized bed0961-9534/$ e see front matter 2012 Elsevhttp://dx.doi.org/10.1016/j.biombioe.2012.12.Artificial neural networks (ANNs) have been applied for modeling biomass gasification

process in fluidized bed reactors. Two architectures of ANNs models are presented; one for

circulating fluidized bed gasifiers (CFB) and the other for bubbling fluidized bed gasifiers

(BFB). Both models determine the producer gas composition (CO, CO2, H2, CH4) and gas

yield. Published experimental data from other authors has been used to train the ANNs.

The obtained results show that the percentage composition of the main four gas species in

producer gas (CO, CO2, H2, CH4) and producer gas yield for a biomass fluidized bed gasifier

can be successfully predicted by applying neural networks. ANNs models use in the input

layer the biomass composition and few operating parameters, two neurons in the hidden

layer and the backpropagation algorithm. The results obtained by these ANNs show high

agreement with published experimental data used R2 > 0.98. Furthermore a sensitivity

analysis has been applied in each ANN model showing that all studied input variables are

important.

2012 Elsevier Ltd. All rights reserved.a r t i c l e i n f o a b s t r a c tArtificial neural network modegasification in fluidized bed ga

Maria Puig-Arnavat a, J. Alfredo HernandezaUniversitat Rovira i Virgili, Dept. Eng. Meca`nica, Av. Pasos CatalabUniversidad Autonoma del Estado de Morelos, Centro de Investiga

1001 Col. Chamilpa, 62209 Cuernavaca, Mexico

http: / /www.elsevierier Ltd. All rights reserved012for biomassifiers

Joan Carles Bruno a,*, Alberto Coronas a

26, 43007 Tarragona, Spain

en Ingeniera y Ciencias Aplicadas (CIICAp), Av. Universidad No.

m/locate/biombioe.

residence time. Taking into account only these two input

Table 1 e Characteristics of input and output variables inthe ANN model for CFB gasifiers.

Range

Input variables for the ANNs

Ash content of dry biomass (g kg1) 4e33.4Moisture content of wet biomass (g kg1) 35e220Carbon content of dry biomass (g kg1) 476.6e529.9Oxygen content of dry biomass (g kg1) 383.8e435.5Hydrogen content of dry biomass (g kg1) 54.3e78.6Equivalence ratio (ER) () 0.19e0.64Gasification temperature (Tg) (C) 701e861Output variables for the various ANNs

Producer gas yield (at 298 K, 103 kPa), (m3 kg1) 1.72e3.30Gas composition (volume fraction, dry basis)

H2 content (%) 3.00e7.30

CH4 content (%) 1.20e4.60

CO2 content (%) 13.94e18.30

CO content (%) 6.90e21.40

Moisture

Ash

C

O

H

ER

Tg

Input layer

( i)Hidden layer Output layer

i=1

i=7

j=1

j=2

k=1

IWj,i

LWk,j

Weights

b1j

b2k

biases

Output

(CO, CO2, H

2, CH

4

orGas yield)

Fig. 1 e ANN model structure to predict producer gas

composition and gas yield from biomass gasification in

a CFB gasifier.

b i om a s s an d b i o e n e r g y 4 9 ( 2 0 1 3 ) 2 7 9e2 8 9280producer gas composition and calorific value. However, the

operation of a biomass gasifier depends on several complex

chemical reactions, including several steps like: pyrolysis,

thermal cracking of vapors to gas and char, gasification of

char, and partial oxidation of combustible gas, vapors and

char. Due to the complexity of the gasification process coupled

with the sensitivity of the products distribution to the oper-

ating conditions; many idealized assumptions have to be

made in the development of these models.

Different kinds of models have been implemented for

gasification systems, including equilibrium, kinetic and arti-

ficial neural networks. According to Villanueva et al. [1],

equilibrium models are considered a good approach when

simulating entrained-flow gasifiers in chemical process sim-

ulators or for downdraft fixed-bed gasifiers, as long as high

temperature and high gas residence time are achieved in the

throat. By contrast, updraft fixed-bed, dual fluidized-bed and

stand-alone fluidized-bed gasifiers should be modeled by

revised equilibrium models or, in some extreme cases, by

detailed rate-flowmodels. A detailed review of recent biomass

gasification models is available elsewhere [2,3].Table 2 e Characteristics of input and output variables inthe ANN model for BFB gasifiers.

Range

Input variables for the ANNs

Ash content of dry biomass (g kg1) 5.5e11.0Moisture content of wet biomass (g kg1) 62.8e250Carbon content of dry biomass (g kg1) 458.9e505.4Oxygen content of dry biomass (g kg1) 411.1e471.8Hydrogen content of dry biomass (g kg1) 56.4e70.8Equivalence ratio (ER) () 0.19e0.47Gasification temperature (Tg) (C) 700e900Steam to dry biomass ratio (VB) (kg kg1) 0e0.04Output variables for the various ANNs

Producer gas yield (at 298 K, 103 kPa), (m3 kg1) 1.17e3.42Gas composition (volume fraction, dry basis)

H2 content (%) 4.97e26.17

CH4 content (%) 2.40e6.07

CO2 content (%) 9.82e18.60

CO content (%) 10e29.47Artificial neural networks (ANNs) have been extensively

used in the field of pattern recognition; signal processing,

function approximation and process simulation. However,

they almost have not been used in the field of biomass gas-

ification modeling. Only few references can be found in the

literature covering this field [4e6]. ANNs are useful when the

primary goal is outcome prediction and important in-

teractions of complex nonlinearities exist in a data set like for

biomass gasification, because they can approximate arbitrary

nonlinear functions. One of the characteristics of modeling

based on artificial neural networks is that it does not require

the mathematical description of the phenomena involved in

the process, and might therefore prove useful in simulating

and up-scaling complex biomass gasification process. Guo

et al. [4] developed a hybrid neural network model to predict

the product yield and gas composition of biomass gasification

in an atmospheric pressure steam fluidized bed gasifier. They

used as input variables the bed temperature and the stockFig. 2 e ANN model structure to predict producer gas

composition and gas yield from biomass gasification in

a BFB gasifier.

Fig. 3 e Comparison of the experimental results with the results calculated by ANN for CFB gasifiers.

b i om a s s a n d b i o e n e r g y 4 9 ( 2 0 1 3 ) 2 7 9e2 8 9 281

variables, forced the authors to develop four ANNs, one for

each biomass feedstock considered. Even the results showed

that the ANNs developed could reflect the real gasification

process; it would have been more interesting to develop just

one but more general model for the biomass gasifier in study

and accounting for different biomass feedstocks.

Brown et al. [5] developed a reaction model for computa-

tion of products compositions of biomass gasification in an

atmospheric air gasification fluidized bed reactor. They com-

bined the use of an equilibrium model and ANN regressions

for modeling the biomass gasification process. Their objective

was to improve the accuracy of equilibrium calculations

and prevent the ANN model from learning mass and energy

balances, thereby minimizing the experimental data re-

quirements. As a result, a complete stoichiometry was for-

mulated, and corresponding reaction temperature difference

parameters computed under the constraint of the non-

equilibrium distribution of gasification products determined

bymass balance and data reconciliation. The ANN regressions

related temperature differences to fuel composition and gas-

ifier operating conditions. This combination of equilibrium

model and ANN was further investigated and improved by

the same authors [6]. Even though the model incorporates

ANNs, it cannot be considered a pure ANNmodel for biomass

gasification process because the most important part of the

model is a stoichiometric equilibrium model.

In this study, two feed-forward ANNs models have been

developed to simulate the biomass gasification process in

bubbling and circulating fluidized bed gasifiers, respectively.

The aim is to obtain twomodels that can predict the producer

gas composition and the gas yield from biomass composition

and few operating parameters, like thermodynamic equilib-

rium models do, but avoiding the high complexity of kinetic

models. The experimental data reported and published by

other authors has been used here to train the ANNs. The

resulting model predictions for different types of biomass,

given by the neural networks, are investigated in detail.

2. Methods

2.1. Experimental data selection

Since different kinds of biomass and different gasifiers have

different gasification behavior, two ANNmodels are presented

in this work. The first one applies for circulating fluidized bed

(CFB) gasifiers and the second one for bubbling fluidized bed

(BFB) gasifiers.

Table 3 eWeights and biases of the ANNs designed for the four major gas species of producer gas (CO, CO2, H2, CH4) andproducer gas yield for ANN model for CFB gasifiers.

CO

IWi,j3.2006 0.0722 0.5638 5.3061 3.7749 0.9014 0.96321.1408 1.8333 0.3493 0.1148 0.4085 3.8072 0.8495LW1,j b1j b2

0.8706 1.9226 2.5402

33b1j3.9

17.

b i om a s s an d b i o e n e r g y 4 9 ( 2 0 1 3 ) 2 7 9e2 8 92825.4159 10.0337 3.60630.0732

CO2IWi,j1.7859 3.1087 4.0413

9.8078 9.1839 12.4537LW1,j4.6685 3.9112

CH4IWi,j1.1889 2.4613 1.7017 4.40291.4276 3.9629 2.6406 3.0161

LW1,j b1j1.2490 6.3563 3.8959

5.2290H2IWi,j1.7403 3.4878 0.818516.8436 24.2709 1.2959LW1,j3.0137 1.9792

Producer gas yield

IWi,j6.8841 6.4443 2.34345.7169 1.6951 1.5775LW1,j0.5083 0.3425b2

8.5215

.3632 4.0765 7.8066 0.7735

.6059 6.6673 20.5250 18.7020b2

094 7.8781

6810

1.3813 3.7339 9.9848 1.42793.2875 3.6455 19.7080 6.0075b1j b2

14.6688 2.449712.9445

5.0279 1.9819 1.5078 0.6350

2.3948 15.3984 9.5719 4.0890b1j b2

7.6506 10.28002.1235

0.2984 2.6040 2.73150.8342

The selection of an appropriate set of variables for inclu-

sion as inputs to the model is a crucial step in model devel-

opment, as the performance of the final model is heavily

dependent on the input variables used.

In this study, an extensive literature review was done to

obtain experimental data that could be used to develop the

ANNs models. Due to the different properties and behavior of

different biomasses, and to have more homogeneous data,

only experimental data for wood gasification in atmospheric

pressure and inert bed reactors was considered. Data for cir-

culating fluidized bed ANN model was obtained for air gas-

ification of wood from Li et al. [7] (cypress, hemlock andmixed

sawdust) and van der Drift et al. [8] (mixed wood). Published

experimental data for bubbling fluidized bed reactors was

found in the studies of Narvaez et al. [9] (pine sawdust),

Campoy [10] (pellets), Kaewluan and Pipatmanomai [11]

(rubber wood chips) and Lv et al. [12] (pine sawdust) for air

and airesteam gasification.

In both ANNs models, the data sets containing the infor-

mation (the values of input and output variables) of different

biomass gasification tests are small. The data sets for CFB and

BFB gasifiers contain the results of 18 and36 tests, respectively.

Due to the small size of the data sets and after some pre-

liminary validation tests and results from the literature [5,6];

the number of input variables was reduced compared to the

initial available ones. Fixed carbon (FC) and volatile matter

(VM) were considered as dependent variables because the FC

ratio is proportional to both the H/C and O/C ratios [5,13,14].

Considering that the gas species to be determined are CO, CO2,

H2 andCH4;nitrogenandsulphurwerenot consideredeitheras

input variables. In addition, their amount in wood is very low

and, in some cases, almost negligible compared with the con-

tent of carbon (C), hydrogen (H) andoxygen (O). For this reason,

the input layer for the CFB ANN model consists of seven vari-

ables: biomassmoisture (MC), biomass content of ash, C,H and

O, gasification temperature (Tg) and equivalence ratio (ER). In

the case of BFBmodel, the operational variables considered for

the input layer were the same than those for CFB gasifier plus

another variable that stands for the ratio between the amount

of steam injected and the biomass flowrate (VB). The charac-

teristics of these input and output variables, obtained from

published experimental data, are shown in Table 1 for CFB

gasifiers and in Table 2 for BFB gasifiers.

2.2. Artificial neural networks topology

An artificial neural network is a systembased on the operation

of biological neural networks, a computationalmodel inspired

b i om a s s a n d b i o e n e r g y 4 9 ( 2 0 1 3 ) 2 7 9e2 8 9 283Fig. 4 e Relative impact (%) of input variables on the different o

producer gas yield of the ANN model for CFB gasifiers.utputs for the four main producer gas components and

Matlab environment using the Neural Network Toolbox [15].

Fig. 1 and Fig. 2 illustrate the architecture of the models for

CFB and BFB gasifiers, respectively. Since there is no explicit

rule to determine either the number of neurons in the hidden

layer or the number of hidden layers, the trial and error

method was applied to find the best solution by minimizing

the Root Mean Square Error (RMSE). In this step of training,

a studywas carried out to determine the number of neurons in

hidden layer which was considered to one and two neurons

for both ANNs models. The best obtained results (data not

show) were considering two neurons in hidden layer (see Figs.

measured by RMSE and regression coefficient (R2), which were

calculated with the experimental values and networks

predictions.

3. Results and discussion

3.1. Proposed ANN model for circulating fluidized bedgasifiers

Five neural networks with seven inputs, two neurons in the

37775

b i om a s s an d b i o e n e r g y 4 9 ( 2 0 1 3 ) 2 7 9e2 8 92841 and 2).

The ANNsmodels proposed in the present study consist in:

- CFB gasifier model: five ANNs, one for each output (CO, CO2,

H2, CH4 and gas yield). Each ANN has one input layer with

seven variables (biomass moisture (MC), biomass content of

ash, C, H and O, gasification temperature (Tg) and equiv-

alence ratio (ER)), one hidden layer with two neurons and

one output.

- BFB gasifier model: five ANNs with eight variables in the

input layer (biomassmoisture (MC), biomass content of ash,

C, H and O, gasification temperature (Tg), equivalence ratio

(ER) and injected steam ratio (VB)), one hidden layer with

two neurons and one output each ANN.

To test the robustness and predict the ability of themodels,

in both ANNs models, the data sets were divided into training

(80%) and validation-test subsets (20%), randomly selected

from the available database. Due to the small size of the

database, validation and test sets were the same.

In all models, a hyperbolic tangent sigmoid function (tan-

sig) was used in the hidden layer and the linear transfer

function ( purelin) was used in the output layer. The input

parameters were normalized in the range of 0.2e0.8. So, any

samples from the training and validation-test sets ( pi) were

scaled to a new value pi using Eq. (1) [19]:

pi 0:2

0:6$pi min

pi

maxpiminpi (1)

aoutput Xj2j1

26664LW1;j$

0BBB@

2

1 exp 2$

Pi7i1

IWj;i$pi

b1j

11CCCAin the natural neurons. An ANN is composed of a large num-

ber of highly interconnected processing elements (neurons or

nodes) working in unison to solve specific problems. The

neurons are grouped into distinct layers and interconnected

according to a given architecture. Each layer has a weight

matrix, a bias vector and an output vector.

In this study, two ANNs models were developed in thewhere pi is the normalized input variable and pi is the input

variable.To assess the relative importance of the input variables,

the evaluation process based on the neural net weight matrix

and Garson equation [18] was used [17,19]. Garson proposed

an equation based on the partitioning of connection weights.

The numerator describes the sums of absolute products of

weights for each input while the denominator represents the

sum of all weights feeding into hidden unit, taking the abso-hidden layer and one output each, was found to be efficient in

predicting producer gas composition as well as gas yield for

CFB gasifiers.

Experimental and simulated values for CO, CO2, H2, CH4,

and gas yield were compared satisfactorily through a linear

regression model ( y a$x b) for each. The obtainedregression coefficients (R2) are presented in Fig. 3. It can be

seen how all R2 values are higher than 0.99 except for the case

of H2 composition that it is 0.98.

According to Verma et al. [16] and El Hamzaoui et al. [17] to

satisfy the statistical test of intercept and slope; the interval

between the highest and lowest values of the intercept must

contain zero and the interval between the highest and lowest

values of the slope must contain one. The proposed ANNs

passed the test with 99.8% of confidence level. This test

guarantees that whole ANNmodel, containing five ANNs, has

a satisfactory level of confidence.

Table 3 gives the obtained parameters (IWj,i, LW1,j, b1j, b2)

of the best fit for 2 neurons in the hidden layer for each of the

five ANN developed in the CFB model. These parameters were

used in the proposed model to simulate the output values. In

consequence, the proposed ANN model follows Eq. (2):

b2 (2)The outputs of each ANNwere comparedwith targets from

experimental data reported by other authors. Tominimize the

error, the LavenbergeMarquardt backpropagation algorithm

was used. The system adjusted the weights of the internal

connections to minimize errors between the network output

and target output.

The performance of the different ANNs was statisticallylute values. The proposed equation, adapted to the present

ANN topology, is as presented in Eq. (3):

Fig. 5 e Comparison of the experimental results with the results calculated by ANN for BFB gasifiers.

b i om a s s a n d b i o e n e r g y 4 9 ( 2 0 1 3 ) 2 7 9e2 8 9 285

Ii

Pj2j1

0BBB@

0BBB@

IWj;iPi7i1IWj;i

1CCCA$LW1;j

1CCCA

Pi7i1

8>>>>>:Pj2

j1

0BBB@

0BBB@

IWj;iPi7i1IWj;i

1CCCA$LW1;j

1CCCA

9>>>=>>>;

(3)

where Ii is the relative influence of the ith input variable on

the output variable. The relative importance of the different

input variables, for each ANN, calculated using Eq. (3) is

shown in Fig. 4. As it can be observed, all variables have

a strong effect on the different outputs (CO, CO2, H2, CH4and producer gas yield). It can be seen how variables that

account for biomass composition (C, H, O) represent be-

tween 31.7% and 54.1% of the importance on CO, CO2, H2and CH4 prediction. However, this importance is reduced to

25% for producer gas yield. On the other hand ER is the

most important variable for producer gas yield prediction

(37.6%) while it is also important for CO and H2 (31.2 and

30.2%) and less important for CO2 (11.5%) and CH4 (12.6%).

Gasification temperature has a relative constant importance

in all cases (around 10%) except for CO2 where it is lower

(4.9%).

3.2. Proposed ANN model for bubbling fluidized bedgasifiers

In this model, the same procedure than that applied for

CFB gasifiers has been followed. The topology of the five

ANNs integrated in the model is the same than in the pre-

vious case. However, here, eight input variables are consid-

ered instead of seven because the model also accounts for

airesteam gasification and not only for air gasification like in

CFB gasifiers.

The obtained regression coefficients (R2) when comparing

experimental and simulated values for CO, CO2, H2, CH4,

and gas yield are presented in Fig. 5. All R2 values are higher

than 0.99 except for the case of CO2 composition that it is

0.98.

The limits for the statistical test of intercept and slopewere

calculated. In all cases, the slope contained one and the

intercept contained zero. Consequently, the proposed ANNs

also passed the test with 99.8% of confidence level.

Table 4 eWeights and biases of the ANNs for the fourmajor producer gas species (CO, CO2, H2, CH4) and producer gas yieldfor the ANN model for BFB gasifiers.

CO

IWi,j0.9005 22.8979 0.3383 10.2693 13.9051 0.5125 1.2177 1.61454.0218 2.0805 0.6249 1.9391 1.0988 0.6812 0.1740 0.5222LW1,j b1j b2

33.7782 39.6833 15.6788 12.35243.6788

19.3959 10.3177 4.3555 12.2481

b i om a s s an d b i o e n e r g y 4 9 ( 2 0 1 3 ) 2 7 9e2 8 9286CO2IWi,j8.6144 1.1591 9.1504 4.13210.4782 3.9688 5.2829 1.2131LW1,j b1j3.5726 2.6414 4.4389

5.3372CH4IWi,j27.6038 30.0594 31.5068 31.934456.8348 245.3845 194.6359 29.1672LW1,j b1j0.4665 1.0988 28.9205

79.8145H2IWi,j2.6766 3.3581 1.7070 0.71231.0173 0.0697 3.1264 1.8738

LW1,j b1j13.8413 8.0323 1.3738

0.7616Producer gas yield

IWi,j5.3707 31.8927 4.4783 23.24724.1585 10.9772 2.1819 5.8447LW1,j b1j

0.5422 1.2019 22.8709

6.01266.7403 4.6368 4.3425 1.4914b2

1.75170.7413 12.6004 1.6067 4.854718.4774 3.4298 6.4298 7.5909

b2

13.4535

49.1297 85.5683 10.8387 1.0029

243.0979 158.8235 82.2433 103.3151b2

4.2972

1.0042 1.4738 0.0854 2.39630.1026 1.6956 5.1339 6.0746

b2

13.6191

Table 4 shows the obtained parameters (IWj,i, LW1,j, b1j, b2)

of the best fit for 2 neurons in the hidden layer for each of the

five ANN developed in the BFB model. The proposed ANN

model follows the same expression than the previous case

but it is necessary to take into account that in this case eight

inputs are considered as shown in Eq. (4):

The relative influence of the input variables was also

evaluated using Eq. (3) as in the CFB gasifiers model. The

relative importance of the different input variables for each

ANN is shown in Fig. 6. As can be seen in the previous

model, in this case, all of the variables also have a strong

effect on the different outputs (CO, CO2, H2, CH4 and

producer gas yield). Variables that account for biomass

composition (C, H, O) always represent, like in CFB model,

more than 25% of the importance of all studied outputs. The

importance of ER is reduced in all cases. However, ER and VB

together represent around 20% of importance in all cases

except for CO.

Results presented in this section and in Section 3.1 show

how the percentage composition of the main four gas species

in producer gas and producer gas yield for a biomass CFB or

BFB gasifier can be successfully predicted by applying a neural

network with two hidden neurons in the hidden layer and

using backpropagation algorithm. The results obtained by

aoutput Xj2j1

26664LW1;j$

0BBB@

2

1 exp 2$

Pi8i1

IWj;i$pi

b1j

11CCCA

37775 b2 (4)

b i om a s s a n d b i o e n e r g y 4 9 ( 2 0 1 3 ) 2 7 9e2 8 9 287Fig. 6 e Relative impact (%) of input variables on the different o

producer gas yield of the ANN model for BFB gasifiers.utputs for the four main producer gas components and

Very few references can be found in the field of biomass

[6] Brown D, Fuchino T, Marechal F. Stoichiometric equilibriummodelling of biomass gasification: validation of artificial

b i om a s s an d b i o e n e r g y 4 9 ( 2 0 1 3 ) 2 7 9e2 8 9288gasification modeling. The two ANN models developed in the

present study for CFB and BFB gasifiers have shown the pos-

sibility that ANN may offer some contribution to research in

this field.

Results presented show how the percentage composition

of the main four gas species in producer gas and producer gas

yield for a biomass CFB or BFB gasifier can be successfully

predicted by applying a neural network with two hidden

neurons in the hidden layer and using backpropagation

algorithm. The results obtained by these ANNs show high

agreement with published experimental data used: very good

correlations (R2 > 0.98) in almost all cases and small RMSEs.

According to analysis, all of the variables have a strong

effect on the different outputs (CO, CO2, H2, CH4 and producer

gas yield) for all ANNmodels. Biomass composition (C, H, O) in

CFB represents between 31.7% and 54.1% of the importance on

CO, CO2, H2 and CH4 prediction and in BFB between 28.9% and

52.3%. In the case of producer gas yield prediction, in CFB, the

ER input is the most important variable (37.6%) while in BFB

model decreases down to 10.8%.

This study is a first step and provides a good approach of

the great potential of this kind of models in this field. How-

ever, further additional experimental data to enlarge the

database would be useful for further ANN training and

improve the developed models. Finally, these proposed ANNs

models can be used to optimize and control the process.

Acknowledgments

The authorswould like to thank the European Commission for

the financial support received as part of the European Project

Polycity (Energy networks in sustainable communities) (TREN/

05FP6EN/S07.43964/51381)

Nomenclature

ANN artificial neural network

BFB bubbling fluidized bed

b1, b2 biases

CFB circulating fluidized bed

ER equivalence ratio ()FC mass fraction% of fixed carbon in dry biomass

IW, LW matrix weightthese ANNs show high agreement with published exper-

imental data used: very good correlations (R2 > 0.98) in almost

all cases and small RMSEs. However, it is necessary to have in

mind that ANN models are limited to a specified range of

operating conditions for which they have been trained. For

this reason, a larger experimental databasewould be desirable

to get improved models.

4. ConclusionsMC mass fraction% of H2O

VM mass fraction% of volatile matter in dry biomassneural network temperature difference parameterregressions. J Chem Eng Jpn 2007;40(3):244e54.

[7] Li XT, Grace JR, Lim CJ, Watkinson AP, Chen HP, Kim JR.Biomass gasification in a circulating fluidized bed. BiomassBioenergy 2004;26(2):171e93.

[8] van der Drift A, Van Doorn J, Vermeulen JW. Ten residualbiomass fuels for circulating fluidized-bed gasification.Biomass Bioenergy 2001;20(1):45e6.

[9] Narvaez I, Oro A, Aznar MP, Corella J. Biomass gasificationwith air in an atmospheric bubbling fluidized bed. Effect ofsix operational variables on the quality of the produced rawgas. Ind Eng Chem Res 1996;35(7):2110e20.

[10] Campoy M. Gasificacion de biomasa y residuos en lechofluidizado: estudios en planta piloto [PhD thesis]. Universityof Seville; 2009.

[11] Kaewluan S, Pipatmanomai S. Potential of synthesis gasproduction from rubber wood chip gasification in a bubblingfluidized bed gasifier. Energy Convers Manage 2011;52(1):75e84.

[12] Lv P, Xiong ZH, Chang J, Wu C, Chen Y, Zhu J. AnH mass fraction% of hydrogen content in dry biomass

I relative influence of an input variable on the output

variable (%)

O mass fraction% of oxygen content in dry biomass

C mass fraction% of carbon content in dry biomass

p input to the ANN model

p

normalized input to the ANN model

R2 correlation coefficient

RMSE root mean square error

Tg gasification temperature (C)VB steam to dry biomass mass ratio (kg kg1)

Subscripts

i number of neurons in the input layer

j number of neurons in the hidden layer

k number of neurons in the output layer

r e f e r e n c e s

[1] Villanueva AL, Gomez-Barea A, Revuelta E, Campoy M,Ollero P. Guidelines for selection of gasifiers modellingstrategies. In: Proceedings of the 16th European BiomassConference and Exhibition; 2008 June 2e6, Valencia, Spain.ETA-Florence Renewable Energies; 2008. p. 980e6.

[2] Puig-Arnavat M, Bruno JC, Coronas A. Review and analysis ofbiomass gasification models. Renew Sustain Energ Rev 2010;14(9):2841e51.

[3] Gomez-Barea A, Leckner B. Modeling of biomass gasificationin fluidized bed. Prog Energy Combust Sci 2010;36(4):444e509.

[4] Guo B, Li D, Cheng C, Lu Z, Shen Y. Simulation of biomassgasification with a hybrid neural network model. BioresourTechnol 2001;76(2):77e83.

[5] Brown D, Fuchino T, Marechal F. Solid fuel decompositionmodelling for the design of biomass gasification systems. In:Marquardt W, Pantelides C, editors. Proceedings of the 16thEuropean Symposium on Computer Aided ProcessEngineering and 9th International Symposium on ProcessSystems Engineering, July 9e13, 2006; Garmisch-Partenkirchen, Germany. p. 1661e1666.experimental study on biomass airesteam gasification ina fluidized bed. Bioresour Technol 2004;95(1):95e101.

[13] van KrevelenDW. Graphical-statisticalmethod for the study ofstructure and reaction processes of coal. Fuel 1950;29:269e84.

[14] Jenkins BM, Baxter LL, Miles TR, Miles TR. Combustionproperties of biomass. Fuel Process Technol 1998;54(1):17e46.

[15] Demuth H, Beale M, Hagan M. Neural network toolboxTM 6users guide. Natick MA: The Mathworks Inc; 2010.

[16] Verma SP, Andaverde J, Santoyo E. Application of the errorpropagation theory in estimates of static formationtemperatures in geothermal and petroleum boreholes.Energy Convers Manage 2006;47(20):3659e71.

[17] El Hamzaoui Y, Hernandez JA, Silva-Martinez S, Bassam A,Alvarez A, Lizama-Bahena C. Optimal performance of CODremoval during aqueous treatment of alazine and gesaprimcommercial herbicides by direct and inverse neural network.Desalination 2011;227(1e3):325e37.

[18] Garson GD. Interpreting neural-network connection weights.AI Expert 1991;6:47e51.

[19] Khataee AR, Mirzajani O. UV/peroxydisulfate oxidation of C.I. Basic Blue 3: modeling of key factors by artificial neuralnetwork. Desalination 2010;251(1e3):64e9.

b i om a s s a n d b i o e n e r g y 4 9 ( 2 0 1 3 ) 2 7 9e2 8 9 289

Artificial neural network models for biomass gasification in fluidized bed gasifiers1. Introduction2. Methods2.1. Experimental data selection2.2. Artificial neural networks topology

3. Results and discussion3.1. Proposed ANN model for circulating fluidized bed gasifiers3.2. Proposed ANN model for bubbling fluidized bed gasifiers

4. ConclusionsAcknowledgmentsNomenclatureReferences