Data-Driven Optimization of C4 Olefin Synthesis: A Gradient Boosting and Particle Swarm Optimization Framework

Abstract

Thesynthesis of C₄ olefins from ethanol presents a promising sustainable pathway, yet its industrial application is hindered by the challenge of optimizing complex, nonlinear interactions among reaction parameters. To address this, we developed a hybrid machine learning framework combining a Gradient Boosting Decision Tree (GBDT) for predictive modeling with Particle Swarm Optimization (PSO) for parameter optimization. The integrated GBDT-PSO model accurately captured the process dynamics, achieving a prediction accuracy of 92.5%. Our results revealed a critical reaction temperature threshold of 350^◦C, above which C₄ olefin yields increase significantly. Optimal conditions were identified at 400^◦C with a 2 wt% Co/SiO₂ + Hydroxyapatite (HAP) catalyst, achieving a maximum yield of 44.73%. The proposed framework markedly outperformed conventional optimization methods, including Genetic Algorithm (88.2%) and Differential Evolution (86.7%). This study not only provides a robust, data-driven methodology for process optimization but also delivers actionable insights that can guide the scale-up and industrial implementation of ethanol-to olefin technologies.