Computational Modeling and Machine Learning for Predicting the Volumetric Flows in Crude Distillation Units: A Detailed Simulation and Validation Approach

[EN] This research presents a predictive model based on Artificial Neural Networks (ANNs) for the prediction of molar flows in Crude Distillation Units (CDUs). Through rigorous simulation in DWSIM, a database of 350 points was generated, correlating the True Boiling Point (TBP) distillation temperat...

Descripción completa

Detalles Bibliográficos
Autores: Chuquin-Vasco, Daniel, Osorio-Getial, Julian, Chuquin-Vasco, Nelson, Chuquin-Vasco, Juan, Aguirre-Ruiz, Diana, Mejía-Peñafiel, Fernando
Tipo de recurso: artículo
Fecha de publicación:2025
País:España
Institución:Universitat Politècnica de València (UPV)
Repositorio:RiuNet. Repositorio Institucional de la Universitat Politécnica de Valéncia
Idioma:inglés
OAI Identifier:oai:riunet.upv.es:10251/225947
Acceso en línea:https://riunet.upv.es/handle/10251/225947
Access Level:acceso abierto
Palabra clave:Artificial Neuronal Networks
Crude
DWSIM
MATLAB
Naphta
Descripción
Sumario:[EN] This research presents a predictive model based on Artificial Neural Networks (ANNs) for the prediction of molar flows in Crude Distillation Units (CDUs). Through rigorous simulation in DWSIM, a database of 350 points was generated, correlating the True Boiling Point (TBP) distillation temperatures of crude oil with the volumetric flows of light and heavy naphtha, distillates, and residue. An ANN with 10 inputs, 20 hidden neurons, and 5 outputs was trained using LevenbergMarquardt (LM), Bayesian Regularization (BR), and Scaled Conjugate Gradient (SCG) algorithms. The BR algorithm demonstrated superior performance, achieving a mean squared error (MSE) of 2.6904E-04 and a regression coefficient (R) of 0.9971 during the testing phase. Validation with experimental data confirmed the accuracy of the model, with average percentage errors of less than 0.68% for all products except residue (6.4%). ANOVA analysis (95% confidence) corroborated the statistical robustness of the ANN. This predictive tool will allow for the optimization of CDU design and operation, with a focus on energy efficiency and minimizing environmental impact. The study discusses the implications for realtime integration with control systems.