A data-driven framework for modeling and optimization of industrial hydrocracking units

Rezaei, Arian (2025) A data-driven framework for modeling and optimization of industrial hydrocracking units. Masters thesis, Memorial University of Newfoundland.

[English] PDF - Accepted Version Available under License - The author retains copyright ownership and moral rights in this thesis. Neither the thesis nor substantial extracts from it may be printed or otherwise reproduced without the author's permission. Download (17MB)

Abstract

This thesis presents a comprehensive framework for modelling and multi-objective optimization of industrial hydrocracking and hydroprocessing units that e↵ectively addresses the challenges associated with complex process dynamics and data quality. The proposed framework combines advanced machine learning techniques, explainability methods and multi-objective optimization algorithms to improve the operational performance and efficiency of hydrocracking units. The framework begins with data preparation based on cycles and modes of operation. Data preprocessing involves addressing missing data using the iterative imputation method, followed by detection and removal of anomalies using the 3-! rule and Isolation Forest algorithm. An extensive comparative analysis was performed for product yield prediction using a diverse set of machine learning algorithms, including Decision Trees, Support Vector Regression, Random Forests, Deep Neural Networks, XGBoost, LightGBM and CatBoost. Through comprehensive performance assessment, CatBoost emerged as the most robust predictive algorithm, demonstrating exceptional potential in mitigating overfitting and significantly enhancing predictive accuracy across datasets. Subsequently, SHAP (SHapley Additive exPlanations) and LIME (Local Interpretable Model-agnostic Explanations) analyses were employed to explain model predictions and identify key influencing input variables to manipulate for optimization. The multi-objective optimization is then performed using the Non-dominated Sorting Genetic Algorithm II (NSGA-II), exploring various scenarios to maximize desired products’ yields or volume swell during cracking. The framework is implemented and validated through two case studies, achieving higher prediction accuracy and providing optimization results that significantly enhance operational performance and profitability. This research highlights the potential of integrating machine learning and optimization techniques to advance industrial processes. Future work could explore using reinforcement learning and digital twins for real-time optimization and control.

Actions (login required)

View Item