Optimizing propane–propylene separation via Aspen Plus simulation and machine learning-based predictive modeling

The separation of propane–propylene mixture is an important process in the petrochemical industry as propylene is a key component for various industrial processes. The separation process with conventional methods is energy intensive, costly, and challenging due to the similar physical properties of both components. Extractive distillation (ED) using aqueous N-methyl-2-pyrrolidone (NMP) as extracting agent can overcome these challenges. However, optimizing process conditions to achieve high purity with minimal energy consumption remains a complex challenge. The aim of the study is to make the separation process efficient and cost effective by combining the extractive distillation (ED) simulation with machine learning (ML) algorithms and optimization. In this work, comprehensive data from simulation of ED using Aspen Plus V11 was used to train ML models i.e. Random Forest, XGBoost, and KNN. The models were used to predict propylene purity, condenser and reboiler duties for extractive and recovery columns. RF, XGBoost, and KNN performed well on the data and obtained R2 scores higher than 0.99. The RF model was then integrated into Genetic Algorithm (GA) to optimize the operating conditions that maximize the propylene purity and minimize the total energy consumption. The GA successfully optimized the separation process with a 99.66% propylene purity and total energy consumption less than the initial ED simulation. The ED process with the aid of ML optimization reduced the total annual cost (TAC) by significant amounts compared to conventional high-pressure binary distillation and has very promising potential to yield significant economic benefits in industry.