Process Simulation and Optimization of a Crude Distillation Unit Using Aspen Plus and MATLAB

Abstract

The crude distillation unit (CDU) is a primary refinery process where operating decisions impact both yield and energy requirements. This work presents an integrated Aspen Plus-MATLAB framework for CDU simulation and optimization. An Aspen Plus model was first established to simulate the base case, focusing on kerosene yield and energy consumption. Two surrogate modelling approaches, namely an autoencoder (AENN) and feed-forward (FFNN) neural networks were trained to predict CDU performance. FFNN outperformed AENN in predicting kerosene yield (R_FFNN² = 0.8467 vs. R_AENN² = 0.7834) and energy consumption (R_FFNN² = 0.9132 vs. R_AENN² = 0.8639), confirming its suitability as a reliable surrogate. The FFNN model was subsequently coupled with a genetic algorithm (GA) to identify optimal operating parameters for minimizing energy use. The FFNN-GA converged smoothly, recommending a CDU pressure of 2.58 bar and feed rate of 400 ton/h. Despite acceptable prediction error between FFNN/GA and Aspen Plus simulation below 20%, the optimized case yielded no net energy savings due to yield-energy trade-offs. The results demonstrate the viability of surrogate-based CDU optimization while emphasizing the need for multi-objective formulations and model refinement.