Tuesday , October 23 2018

Neural and Hybrid Modelling of Biotechnological Process

Elena CHORUKOVA, Ivan SIMEONOV
The Stephan Angeloff Institute of Microbiology, Bulgarian Academy of Sciences
Acad. G. Bonchev St., Bl. 26, 1113 Sofia, Bulgaria

Abstract: The paper deals with neural and hybrid modelling of the three main types of biotechnological processes – batch and fed-batch processes for the enzyme superoxide dismutase (SOD) production and continuous process of anaerobic digestion (AD) of organic wastes. Neural models for batch SOD production with and without dissolved oxygen regulation have been developed. A hybrid model for fed-batch process for SOD production and neural model for continuous AD process have been developed as well. The most appropriate network architecture in all cases is with one hidden layer with hyperbolic tangent activation function and output layer with linear activation function.

Keywords: SOD production, anaerobic digestion, laboratory experiments, neural models, hybrid models.

Elena Chorukova received her MSc degree in Bioengineering from the Technical University of Sofia, Bulgaria, in 1994. Since 2000 she has been a Research Associate at the Bulgarian Academy of Sciences, in the research group “Mathematical Modeling and Computer Science” of the Stephan Angeloff Institute of Microbiology. She is the author/co-author of about 30 scientific papers published in international journals and conference proceedings.

Ivan Simeonov received his MSc and PhD degrees in Control Engineering from the Technical University of Sofia, Bulgaria, in 1971 and 1981, respectively. Since 1986 he has been a Senior Research Associate with the Bulgarian Academy of Sciences, where he is presently the Head of the research group “Mathematical Modeling and Computer Science” at the Institute of Microbiology. He has been an Associate Professor at the Technical University of Sofia (1988-1999), Technical University of Gabrovo (2000-2005) and New Bulgarian University (since 2005). He is the author/co-author of over 160 scientific papers published in international journals and conference proceedings.

>>Full text
CITE THIS PAPER AS:
Elena CHORUKOVA, Ivan SIMEONOV, Neural and Hybrid Modelling of Biotechnological Process, Studies in Informatics and Control, ISSN 1220-1766, vol. 17 (3), pp. 305-314, 2008.

1. Introduction

The complexity and the non-linear dynamics of biotechnological processes require sophisticated methods for their modelling [1, 12]. Such methods should be able to update their knowledge and refine the model through interaction with the environment, that means the systems should have an ability of learning for a short time.

Theoretical results indicate clearly the advantages of Artificial Neural Networks (ANNs) when used especially for real data learning applications. Neural models are useful for many biotechnological applications [1, 2, 3].

On the basis of laboratory experiments and knowledge concerning the process for biosynthesis of the enzyme SOD, deterministic mathematical models have been developed [4]. However, it is very difficult to choose an appropriate non-linear structure for deterministic models including dissolved oxygen (DO). The ANN approach may be more useful in this case.

The anaerobic digestion (AD) has been long an object of mathematical modelling with deterministic approach [5], however parameter estimation of these non-linear models is a very hard problem [6]. That is why recently the instrument of ANNs has been used for AD modelling [7, 8, 11].

The objective of this study is to investigate the ability of the ANN approach for modelling of the three main types of biotechnological processes – batch, fed-batch and continuous. The studied processes have been as follows: batch and fed-batch processes for biosynthesis of the enzyme superoxide dismutase (SOD) and continuous process of AD of organic wastes.

5. Conclusion

Neural networks with different architectures have been studied during the modelling of all processes. The most appropriate network architecture in all cases is with one hidden layer with hyperbolic tangent activation function and output layer with linear activation function. Automatic software pruning did not improve the models.

Deterministic and neural models of the main variables (protein concentration, biomass concentration and enzyme activity) of batch process for the enzyme SOD production have been developed. The neural model predict better than the deterministic one.

Neural models for batch cultivation without DO regulation have been developed. They could be used for one-step-ahead prediction of the real time immeasurable state variables (protein concentration, biomass concentration and enzyme activity).

Neural models for batch cultivation with DO regulation using dissolved oxygen as a control input have been developed. They represent a good basis for developing of optimal control strategies of the process.

A hybrid model for fed-batch process for the enzyme SOD production has been developed, in which specific growth rates (for biomass growth, substrate consumption and enzyme formation) have been modelled using the ANN approach. They represent a good basis for developing of optimal control strategies of the process.

Neural models of the AD of organic wastes have been developed. They could be used as one-step-ahead predictor of the biogas flow rates for different dilution rates and for process control design.

References:

  1. EIKENS, B., KARIM, M., SIMON, L., Neural networks and first principle models for bioprocesses, Proc. of the 14th IFAC congress, 1999, pp. 367-372.
  2. NUCCI, E.R. et al., Comparing the performance of multilayer perceptrons networks and neuro-fuzzy systems for on-line inference of Bacillus megaterium cellular concentrations, Bioprocess Biosyst Eng., 30, 2007, pp. 429-438.
  3. TE BRAAKE, H., Neural Control of Biotechnological Processes, PhD thesis, Technical University of Delft, 1997.
  4. SIMEONOV, I., CHORUKOVA, E., DOLASHKA, P., ANGELOVA, M., Modelling of the growth kinetics and the biosynthesis of the enzyme SOD in batch cultivation, Automatics and Informatics 2000, Sofia, Oct. 24-26, 2000, pp. 77-79 (in Bulgarian).
  5. SIMEONOV, I., GALABOVA, D., QUEINNEC, I., Investigations and mathematical modelling of the anaerobic digestion of organic wastes with addition of electron acceptors, 9th World Congress “Anaerobic Digestion”, Antverpen, Sept. 2-5, 2001, pp. 381-383.
  6. SIMEONOV, I., Mathematical modelling and parameters estimation of anaerobic fermentation processes, Bioprocess Eng., 21, 1999, pp. 377-381.
  7. STEYER, J.P. et al., Neural Network modelling of a depollution process, Bioprocess Engineering, No 6, vol. 23, 2000, pp. 727-730.
  8. CAKMAKCI, M., Adaptive neuro-fuzzy modelling of anaerobic digestion of primary sedimentation sludge, Bioprocess Biosyst. Eng., 30, 2007, pp. 349-357.
  9. ANGELOVA, M., DOLASHKA-ANGELOVA, P., IVANOVA, E., SERKEDJIEVA, J., SLOKOSKA, L., PASHOVA, S., TOSHKOVA, R., VASSILEV, S., SIMEONOV, I., HARTMAN, H.-J., STOEVA, S., WESER, U., VOELTER, W., A novel glycosylated Cu/Zn-containing superoxide dismutase: production and potential therapeutic effect, Microbiology UK, vol. 147, 2001, pp. 1641-1650.
  10. NORGAARD, M., RAVN, O., POULSEN, N. K., NNSYSID-Toolbox for system identification with Neural Networks, Mathematical and Computer Modelling of Dynamical Systems, Vol. 8, No 1, 2002, pp. 1-20.
  11. PREMIER, G.C. et al., A comparison of the ability of black box and neural network models of ARX structure to represent a fluidized bed anaerobic digestion process, Wat. Res., Vol.33. No 4, 1999, pp. 1027-1037.
  12. VAN IMPE, J., et al., Advanced Instrumentation, Data Interpretation and Control of Biotechnological processes, Kluwer Acad. Publ., Dordrecht, 1998.