Evaluation of neural networks for simulation of three-phase bubble column reactors

The use of a neural network model (NNM) to simulate the performance of a three-phase slurry bubble-column reactor for Fischer-Tropsch synthesis is investigated. The learning set needed to generate the NNM is obtained from a cell-type model where the number of cells relates to the degree of backmixing. To develop the neural network and to perform the required learning, model-predicted output responses are generated from the cell model by using all possible combinations of six key input parameters. The axial variation of the output responses is represented by a recurrent NNM. The NNM parameters are then identified using a special-purpose package that implements both training and analysis. To simulate the behaviour of an actual reactor, data used for training are corrupted with random noise. The NNM obtained from noisy data exhibits substantial filtering capability.