Dynamic Optimisation of Batch Processes Using Neural Network Modelling Methods (1997 - 2000)

Student name: Dr. Yuan (Michael) Tian

Degree and date obtained: PhD, December 2001

Career after study: currently employed by Panasonic Technology Inc., Princeton, NJ, USA

Thesis Abstract: Batch processes, considering their non-linearity and process mechanism, are much more complicated than continuous processes. In this contribution, several novel neural network model based optimal control schemes have been proposed and applied to the batch processes. Several approaches have been initiated in the modelling process and in the application of optimisation techniques. In terms of modelling, the traditional mathematical modelling method has faced increasing difficulties because of many complex chemical reactions. In this work, several techniques using neural networks modelling methods are experimented, in order to seek the best way to represent the real process. The basic neural network, which is used in this work, is recurrent neural network. With feed back functionality, recurrent neural network shows very powerful generalisation capability. By extending the application of recurrent neural network, hybrid neural network models and stacked neural network models are also investigated, of which the first one combines the recurrent neural network model and basic first principle model, while the latter combines multiple neural networks to generate more accurate predictions than that of any of the individual networks.

Based on the neural network models, optimal control of the batch processes can be realised using the Sequential Quadratic Programming method. From off-line optimisation to on-line optimisation, and furthermore, to re-optimisation by tackling the unknown disturbances such as the reactive impurities and reactor fouling during the batch process, they have all been discussed and applied to achieve the best process performance and reliability. With various combinations and comparisons of the modelling and optimisation methods, this contribution endeavours to find the most practical and profitable solution for the current chemical processes especially polymerisation processes.

Tian, Y., Zhang, J. and Morris, A. J., "Modelling and optimal control of a batch polymerisation reactor using a hybrid stacked neural network model", Industrial and Engineering Chemistry Research, Vol. 40, pp. 4525-4535, 2001.

Tian, Y., Zhang, J. and Morris, A. J., "Optimal control of a fed-batch bioreactor based upon an augmented recurrent neural network model", Neurocomputing, 2002, in press.

Tian, Y., Zhang, J. and Morris, A. J., "Optimal control of a batch emulsion copolymerisation reactor based on recurrent neural network models", Chemical Engineering and Processing, 2002, in press.

Tian, Y., Zhang, J. and Morris, A. J., "Dynamic on-line re-optimisation control of a batch MMA polymerisation reactor using hybrid neural network models", to be submitted to Chemical Engineering Research and Design, 2001.

Tian, Y., Zhang, J. and Morris, A. J., "On-line re-optimisation control of a batch MMA polymerisation reactor based on a hybrid neural network model", Proceedings of American Control Conference, Arlington, VA, USA, 25-27 June 2001, Vol. 1, pp. 350-355.

Tian, Y., Zhang, J. and Morris, A. J., "On-line re-optimisation of a batch polymerisation reactor using neural network models", Proceedings of CACUK2000, Loughbough, U.K., 23-24 September 2000, pp. 45-50.

Tian, Y., Zhang, J. and Morris, A. J., "Recurrent neural network model based optimal control of a batch emulsion copolymerisation reactor", Proceedings of CACUK99, Derby, U.K., 25-26 September 1999, pp. 85-90.

Tian, Y., Zhang, J. and Morris, A. J., "Optimal control of a batch emulsion copolymerisation reactor based on recurrent neural network models", Proceedings of EUFIT'99, Aachen, Germany, 13-16 September, 1999, CD-ROM, Paper No. BD2-4.

Tian, Y., Zhang, J. and Morris, A. J., "Neural network based optimal control of a fed-batch reactor", Proceedings of EUFIT'98, Aachen, Germany, 7-10 September, 1998, Vol. 1, pp. 308-312.