Evaluation of ANN modeling for prediction of crude oil fouling behavior

In this research, artificial neural network (ANN) modeling for prediction of crude oil fouling behavior in preheat exchangers of crude distillation units has been evaluated. Outputs of the ANN model have been compared with appropriate sets of experimental data in order to compute overall mean relative error (OMRE). The value of OMRE was also computed for three different threshold fouling models. The OMRE of ANN model was 26.23% whereas the lowest OMRE obtained using threshold fouling models was 47.9%. In order to identify relative significance of the governing variables, a sensitivity analysis, named sequential zeroing of weights (SZWs), has also been performed. This analysis showed that the influence of crude velocity and tube diameter on the fouling rate is higher than tube surface temperature.