Multi Objective Optimization of Sulfur Recovery Unit for Minimization of Energy Consumption and Cost

Abstract

Crude oil remains a vital energy source; nevertheless, its production faces increasing challenges due to climate change concerns and stricter environmental regulations. Ensuring its sustainable utilization requires prioritizing practices that minimize environmental impact. Optimizing the performance of existing infrastructure, such as Sulfur Recovery Units (SRUs), plays a crucial role in this endeavor. SRUs play a vital role in mitigating the environmental footprint of oil processing by capturing and converting hazardous gases like hydrogen sulfide (H2S) into elemental sulfur. However, achieving high Sulfur Recovery Efficiency (SRE) conflicts with economic considerations, as optimal performance requires precise control of process parameters. This research investigates the application of machine learning algorithms to predict SRE and optimize process parameters in SRUs. An industrial SRU model was simulated in Aspen HYSYS and validated with industrial data. This simulated model was then used to generate diverse datasets by incorporating artificial uncertainties of ±5% in key process parameters like air/gas flow rates, temperatures, and Claus reactor inlet temperature. These datasets were used to train three machine learning models: Gaussian Process Regression (GPR), Artificial Neural Network (ANN), and Regression Trees (RT). The evaluation criteria for these models were R-squared (R²) and Root Mean Squared Error (RMSE) values of the validation step. All three models achieved high prediction accuracy, with R² values exceeding 0.97. Furthermore, the most accurate models, GPR and ANN, were employed as surrogate models within the fitness function evaluation of two optimization frameworks: a Genetic Algorithm (GA) and a Particle Swarm Optimization (PSO) approach. Both algorithms demonstrated effectiveness, but PSO exhibited marginally superior performance in optimizing the process parameters. This proposed methodology, utilizing surrogate-based optimization, establishes a platform for practical implementation at the plant level. It paves the way for the realization of Industry 4.0 principles within the refinery sector by enabling the optimization of existing infrastructure for improved sustainability and environmental performance.