Neural Computing and Applications

, Volume 19, Issue 1, pp 67–75 | Cite as

The prediction of SO2 removal using statistical methods and artificial neural network

Original Article


Artificial neural network and a statistical model have been applied in a laboratory scale trickle bed reactor (TBR) to investigate the SO2 removal efficiency of activated carbon. The performance of artificial neural network (ANN) model has been compared with the statistical model based on central composite experimental design. Two independent variables, which affect the amount of SO2 removal by the liquid phase in the TBR, were selected; namely liquid flow rate and gas flow rate. Amount of SO2 removal was chosen as the dependent variable (target data). A second order statistical model has been considered to show the dependence of the amount of SO2 removal on the operating parameters. A back-propagation ANN has been used to develop a model relating to the amount of SO2 removal. A series of experiments have been conducted on the basis of the statistics-based design of experimental method. It is observed that a neural network architecture having one input layer with two neurons, one hidden layer with three neurons, one output layer with one neuron and an epoch size of 20 gives better prediction. The predictions are more accurate than those obtained from regression models.


Artificial neural network Experimental design Trickle bed reactor SO2 removal 

List of symbols


Statistical model coefficients


Real value of the parameters


Average values of the parameters


Average value of the independent variables at centre points


Incremental value of the parameters


Liquid flow rate


Gas flow rate


Coded value of the liquid flow rate


Coded value of the gas flow rate


Error square integral


ith experimental value of the amount of SO2 removal


Trickle bed reactor

Prob > F

Probability of seeing the observed F value if the null hypothesis is true. Small probability values call for rejection of null hypothesis. The probability equals the proportion of the area under the curve of the F distribution that lie beyond the observed F value. The F distribution itself is determined by the degrees of freedom associated with the variances being compared

R squared

A measure of the amount of deviation around the mean explained by the model

Mean squared

Sum of squares divided by DF

Model F value

A test for comparing model variance with residual variance. If the variances are close to the same the radio will be close to one and it is less likely that any of the factors have a significant effect on the response calculated by model mean square divided by residual mean square


Degrees of freedom


  1. 1.
    Castellari AT, Haure PM (1995) Experimental study of the periodic operation of a trickle bed reactor. AIChE J 41(6):1593–1597. doi: 10.1002/aic.690410624 CrossRefGoogle Scholar
  2. 2.
    Coşkun N, Yıldırım T (2003) The effects of training algorithms in MLP network on image classification, vol 2. In: International joint conference on neural networks (IJCNN), Portland, Oregon, USA, pp 1223–1226Google Scholar
  3. 3.
    Dernuth H, Beale M (1992) Matlab neural network toolbox. The Math Works Inc., Natick, pp 6.2–6.3Google Scholar
  4. 4.
    Dudukovic MP, Larachi F, And Mills PL (2002) Multiphase catalytic reactors: a perspective on current knowledge and future trends. Catal Rev 44:123–246. doi: 10.1081/CR-120001460 CrossRefGoogle Scholar
  5. 5.
    Fausett L (1994) Fundamental of neural networks—architectures, algorithms, and applications. Prentice-Hall, Englewood cliffsGoogle Scholar
  6. 6.
    Fernandes FAN, Lona LMF (2005) Neural network applications in polymerization processes. Brazilian J Chem Eng 22(03):401–418Google Scholar
  7. 7.
    Lee KT, Bhatia S, Mohamed AR, Chu KH (2006) Optimizing the specific surface area of fly ash-based sorbents for flue gas desulfurization. Chemosphere 62(1):89–96CrossRefGoogle Scholar
  8. 8.
    MATLAB (2002) Version 6.5.0 help files. Math Works IncGoogle Scholar
  9. 9.
    Montgomery DC (2005) Design and analysis of experiments, 6th edn. Wiley, HobokenGoogle Scholar
  10. 10.
    Nunnari G, Dorling S, Schlink U, Cawley G, Foxall R, Chatterton T (2004) Modelling SO2 concentration at a point with statistical approaches. Environ Modell Softw 19(10):887–905CrossRefGoogle Scholar
  11. 11.
    Rubio JJ, Yu W (2006) A new discrete-time sliding—mode control with time-varying gain and neural identification. Int J Control 79(4):338–438MATHCrossRefGoogle Scholar
  12. 12.
    Rubio JJ, Yu W (2007) Stability analysis of nonlinear system identification vis delayed neural networks. IEEE Trans Circuits Syst II 54(2):161–165CrossRefGoogle Scholar
  13. 13.
    Silveston PL, Harika J (2004) Periodic operation of three-phase catalytic reactors. Can J Chem Eng 82(6):1105–1142CrossRefGoogle Scholar
  14. 14.
    Simpson PK (1990) Artificial neural system—foundation, paradigm, application and implementations. Pergamon Press, New YorkGoogle Scholar
  15. 15.
    Tamura S, Tateishi M (1997) Capabilities of a four-layered feedforward neural network: four layers versus three. IEEE Trans Neural Netw 8:251CrossRefGoogle Scholar
  16. 16.
    Uçan HL (2001) The sorption of sulfur dioxide in a trickle bed reactor with periodic liquid flow and activated carbon bed. MS thesis, Gazi University, AnkaraGoogle Scholar
  17. 17.
    Uçan HL, Özkan G, Biçer A, Pamuk V (2005) Removal of SO2 in a periodically operating TBR with activated carbon bed. Process Saf Environ Prot 83(B1):47–49Google Scholar

Copyright information

© Springer-Verlag London Limited 2009

Authors and Affiliations

  1. 1.Department of Chemical Engineering, Faculty of EngineeringAnkara UniversityAnkaraTurkey
  2. 2.Department of Chemical EngineeringGazi UniversityAnkaraTurkey

Personalised recommendations