A Trustworthy Artificial Intelligence Framework for Predicting Gasoline Octane Loss Using Sparse Autoencoder and Stacking Ensemble Learning in Petrochemical Processes

In response to the increasing demand for reliable and robust artificial intelligence (AI) applications in petrochemical and process industries, this study proposes an intelligent prediction framework for estimating Research Octane Number (RON) loss during gasoline refining. The approach integrates a Sparse Autoencoder (SAE) for feature extraction and a Stacking Ensemble Learning (StackingEL) model for predictive regression, thereby enhancing performance in high-dimensional and noisy industrial datasets. Real-world process data obtained from a petrochemical enterprise were utilized for model training and evaluation. After comprehensive data preprocessing, the SAE effectively captured latent representations of complex process variables, which were then used to train twelve regression models including Lasso and advanced ensemble techniques. Experimental results indicate that the proposed SAE+StackingEL framework outperforms conventional methods in prediction accuracy, robustness, and generalization ability. This AI-assisted process modeling strategy contributes to optimizing gasoline production, reducing environmental emissions, and supporting cleaner and more sustainable industrial practices. The proposed method demonstrates significant potential for integration into Industry 4.0 systems and petrochemical process improvement.