Fast response distributed parameter fluidized bed reactor model for propylene partial oxidation using feed-forward neural network methods

The use of a neural network model (NNM) to simulate the performance of a fluidized-bed reactor for the partial oxidation of propylene to acrolein is investigated. The training set needed to generate the NNM is obtained from a two-phase cell model of the fluidized-bed where the flow patterns for the bubble and emulsion phases in each cell are assumed to be plug-flow and perfectly mixed, respectively. The intrinsic kinetics, which are taken from the literature, are based upon a single site redox type model that exhibits a nonlinear dependence on both molecular oxygen and propylene. The formation of acrolein, acetaldehyde, and total combustion products is described by a series-parallel reaction network. The fluidized bed model accounts for variable gas velocity as well as finite transport resistance between the bubble and emulsion phases. To perform the required NNM training, output responses predicted from the cell model are first generated by using all possible combinations of eleven key input parameters varied over practical ranges of interest. The axial variation of the nine output responses is represented by a recurrent NNM. The NNM parameters are then identified using a special-purpose computer software package that implements both training and analysis of the input data and corresponding output responses. To simulate the behavior of a real reactor, the output responses are corrupted with random noise. Comparisons between the output responses obtained from the NNM trained to noisy data to those from the cell model with no noise indicate that the NNM is capable of providing filtering Furthermore, a sensitivity analysis indicates that the NNM captures the dependence of the output variables on the input ones.