ANN Based Surrogate Model for Exergy Efficiency Optimization of Heat Exchanger Network of a Crude Distillation Unit Under Uncertainty

Abstract

In this work, a combined framework of the Artificial Neural Network and Genetic Algorithm was developed to realize higher exergy efficiency of Heat Exchanger Network of a Crude Distillation Unit, under uncertainty in process conditions. Initially, the steadystate exergy analysis was carried out using an Aspen HYSYS model to quantify the exergy destructions and exergy efficiencies of all the individual heat exchangers and the overall Heat Exchanger Network. Then the Aspen HYSYS model was changed to dynamic mode by introducing uncertainty of ±5% in various process parameters, i.e., temperatures, pressures, and mass flow rates of different process streams, to produce a dataset of 200 samples for HEN of CDU. Using the dataset, an ANN model was generated for the prediction of overall exergy efficiency of HEN. The trained ANN model was then used as a surrogate in the GA environment to attain improved overall exergy efficiency of the HEN in the presence of uncertainty. Using a GA-based approach, the optimized process conditions were found and then put into the Aspen HYSYS model for the purpose of crossvalidation. The overall exergy destruction and exergy efficiency of the HEN were 17611.21 kW and 63.34%, respectively. The trained ANN model had a correlation coefficient (R) of 0.9996 and an RMSE of 0.0097 for overall exergy efficiency of HEN. The performance of the GA-based approach was good enough, and it significantly enhanced the overall exergy efficiency of HEN when compared to standalone Aspen HYSYS model of the process.