A Hybrid Machine Learning and Mathematical Optimization Framework for Modeling, Simulation, and Global Sensitivity Analysis in Petrochemical Processes

The increasing complexity of petrochemical processes necessitates the development of advanced modeling and optimization techniques. This paper proposes a hybrid framework integrating Extreme Gradient Boosting (XGBoost), mathematical optimization using Pyomo, and global sensitivity analysis based on Sobol indices. The objective is to enhance predictive accuracy, optimize operational decisions, and improve system interpretability. A synthetic dataset mimicking real petrochemical processes is generated, incorporating nonlinear relationships among operational variables such as temperature, pressure, flow rate, catalyst concentration, and feed composition. The predictive model demonstrates high accuracy, with significant improvements over traditional regression approaches. The optimization model identifies optimal operating conditions under nonlinear constraints, while the sensitivity analysis reveals key influencing parameters. The results highlight the effectiveness of integrating machine learning with optimization in complex industrial systems.

کلیدواژه ها:

Machine Learning ، XGBoost ، Pyomo ، Petrochemical Process ، Optimization ، Sensitivity Analysis ، Simulation