Abstract:Hydrotreating (HDT) is one of the most important p etroleum refining processes. The process consists of removing impurities such as sulfur, nit rogen and metals and reducing the aromatic content from petroleum fractions under a hydrogen pressure in the presence of a solid catalyst [1]. The HDT process can be used to enhance the quality of refin ed products so as to reach environmental constraint s and commercial specifications as for kerosenes and gas oils, but also to pretreat feedstocks to other processing units such as naphtha cuts and vacuum ga s oils. To accurately predict process performances, reliab l kinetic models are needed. Classic HDT kinetic models are based on a lumping strategy, in which molecular components are grouped into several chemical families, according to their globa properties (boiling point, solubility, etc.). Hen ce, the actual molecular information is hidden due to t he multicomponent nature of each lump. With environmental restrictions becoming more stringent, kinetic models have to be able to predict molecular properties of the products, a requirement tha can not be achieved by using a lumped modeling strategy. Hence, more detailed kinetic mod els that contain molecule-based reaction pathways need to be developed. However, developing moleculebased kinetic models for complex petroleum feedstocks is intricate, not only due to the large number of chemical species, reactions and associate d rate constants, but also due to difficulties arisin g in the characterization of these complex mixtures . The present work focuses on the development of a n vel two-step kinetic modeling strategy that retains the molecular detail throughout the method. In the proposed approach, the lack of molecular detail of the petroleum fractions is overcome by us ing a carefully selected synthetic mixture of representative molecules which has properties close t the available feedstock analyses. This mixture of molecules is generated in the first step of the methodology via a molecular reconstruction method. In the second step, a method called “kinetic Monte Car lo” (kMC) is utilized to simulate the reactions of this mixture of molecules. This method is based on a Markovian process, which transforms the molecule, event by event, by means of probability c onsiderations. In this way, the reaction network is generated “on-the-fly” thereby reducing the size of the reaction network to be managed along the simulations. The methodology has been applied to hydrotreating of two Light Cycle Oil (LCO) gas oils and the predicted effluents were favorably compared to the experimental yield pattern.

Reducing the number of experiments required for modelling the hydrocracking process with kriging through Bayesian transfer learning

Transfer learning for predicting source terms of principal component transport in chemically reactive flow

Probabilistic transfer learning methodology to expedite high fidelity simulation of reactive flows

Application of nonlinear partial least square in catalyst modeling

Improving Data-Driven Inferential Sensor Modeling by Industrial Knowledge: A Bayesian Perspective

Bayesian Optimization of Catalysts With In-context Learning

Data-efficient modeling of catalytic reactions via enhanced sampling and on-the-fly learning of machine learning potentials

MODELING OF HYDROTREATING PROCESSES VIA MOLECULAR F EED RECONSTRUCTION AND MOLECULE-BASED MONTE CARLO KINET ICS

Bayesian optimization with transfer learning: A study on spatial variability of rock properties using NMR relaxometry

An adaptive kriging method for solving nonlinear inverse statistical problems

Operation Optimization of Hydrocracking Process Based on Kriging Surrogate Model

A cost-effective strategy of enhancing machine learning potentials by transfer learning from a multicomponent dataset on ænet-PyTorch

Integrated Modeling of Transfer Learning and Intelligent Heuristic Optimization for a Steam Cracking Process

Integrated Modeling of Transfer Learning and Intelligent Heuristic Optimization for Steam Cracking Process

Learning reduced kinetic Monte Carlo models of complex chemistry from molecular dynamics

Hierarchical Bayesian Modelling for Knowledge Transfer Across Engineering Fleets via Multitask Learning

Optimization-Based Multi-Source Transfer Learning for Modeling of Nonlinear Processes

Kinetics parameter optimization of hydrocarbon fuels via neural ordinary differential equations

Agent-Based and Stochastic Optimization Incorporated with Machine Learning for Simulation of Postcombustion CO2 Capture Process

Unlocking Potential Catalysts: A Machine Learning Approach with Bayesian and Regression Models

Recursive co-kriging model for Design of Computer experiments with multiple levels of fidelity with an application to hydrodynamic