(189c) Modeling of Hydrotreating Processes Via Molecular Feed Reconstruction and Molecule-Based Monte Carlo Kinetics

Over the last years, both the quality and the maximum allowable impurities content of refined products have been subject to increasingly more severe environmental constraints. This has driven research to develop improved hydrotreating processes. In order to correctly predict the process performances, the accuracy and reliability of the kinetic models needs to be improved. Classically, kinetic models for transformation of complex hydrocarbon mixtures are based on a lumped kinetic model, in which molecular components are grouped into several chemical families, according to some global physical properties (boiling point, solubility, …). However, the precision and especially the robustness of a lumped model strongly depend on the range of operating conditions and feedstock compositions used to fit the model parameters. Moreover, such models assume that similar physical properties result in similar reactivities, and that the properties of the lumps do not change during the reaction, which is not true.

The limitations of lumped models motivated the development of more detailed kinetic models containing molecule-based reaction pathways. Such models expect a molecular description of the feedstock, however. Unfortunately, even though the most advanced analytical techniques allow identifying a large number of compounds and classes of chemical families, the complete and quantitative molecular detail of process feedstocks still remains unknown.

In the present work, a novel methodology for the kinetic modeling of refining processes is presented. The methodology models both the feedstock composition and the process reactions at a molecular level.

The composition modeling [1-16] consists of generating a set of molecules whose properties are close to those obtained from the process feedstock analyses. In this work, the set of molecules is generated using the SR-REM molecular reconstruction algorithm. This approach results from the coupling of two methods, stochastic reconstruction (SR) and reconstruction by entropy maximization (REM).

The SR method generates the synthetic mixture of molecules from a set of probability distribution functions (pdf) for molecular structural attributes [3,4,8,9,11,16]. To this aim, the pdf's are sampled via a Monte Carlo procedure to define the type and the number of structural blocks (aromatic rings, alkyl chains, etc) of a molecule. The structural blocks are then assembled to obtain the structure of the molecule. The construction of a molecule is repeated N times so as to obtain a mixture of molecules. The pure component properties of each molecule are calculated either by direct inspection of its structure or by group contribution methods. Then, the mixture properties are compared to the available analyses through an objective function. Finally, the value of the objective function is minimized by means of a genetic algorithm that modifies the parameters of the pdf's for the structural attributes.

The Reconstruction by Entropy Maximization (REM) method [8,10-12] consists of adjusting the molar fractions of the molecules to achieve a close correspondence between the mixture properties of the set of molecules and those of the feedstock analysis. The adjustment is performed by maximizing an information entropy criterion, based on the Shannon's criterion [17].

After generating this synthetic set of molecules, the hydrotreating process is simulated by applying, event by event, its main reactions to the set of molecules using a variable time step kinetic Monte Carlo simulation method [18]. In this method, all potential reactions are identified from the structure of the molecules and their associated rate constants are determined. The rate constants are normalized to obtain the probability that a reaction occurs. Based on the current reaction network and the individual reaction probabilities, a cumulative probability distribution function accounting for the entire reaction system is constructed. At this point, two random numbers are drawn. The first random number determines the time delay before the next reaction occurs. In order to select the next reaction, the second random number samples the cumulative probability distribution function of the overall reactivity. The selected reaction is then executed by transforming the reactant molecule. Finally, the simulation time is updated by incrementing the time step between the previous reaction and the selected reaction. This procedure is repeated until the final simulation time is reached and all molecules are reacted.

This modeling methodology has been applied to the hydrotreating (HDT) of Light Cycle Oil (LCO) gas oils. The experiments were carried out in an isothermal fixed-bed up-flow reactor containing 200 ml of a sulfided commercial NiMo/Al₂O₃ catalyst [19,20]. The operating conditions were varied over a large range: temperature was varied between 320°C and 390°C, total pressure ranged from 20 to 110 bar, while the range for LHSV was varied between 0.5 and 4 h^-1. The LCO gas oils were characterized through elemental analysis (C, H, S), mass spectrometry, NMR ¹³C and simulated distillation [19-21]. Each LCO gas oil was represented by 5000 molecules that were reconstructed from these analyses. A molecule has either zero cores (Paraffins) or one core with a maximum of 4 rings (benzenes, cyclohexanes, or thiophenes). Since the LCO gas oil is drawn from the fractionator of a catalytic cracking unit, the maximum length of the alkyl chains on these rings is limited to 3 carbon atoms. Finally, the paraffinic molecules are considered to be linear.

The hydrotreating of LCO gas oil was simulated by accounting for the hydrogenation of aromatic rings, the dehydrogenation of saturated rings and the hydrodesulfurization of thiophene rings. The hydrogenation / dehydrogenation reactivities are calculated by using three Quantitative Structure / Reactivity Correlations (QS/RCs) [22-23]. These three correlations estimate the reactivity from the reaction stoichiometry with respect to hydrogen molecule, the heat of hydrogenation reaction at 25°C (kJ/mol) and the number of aromatic, saturated and thiophene rings in the molecule. The hydrodesulfurization reactions are classified into five families according to the molecular structure of the reacting sulfur compound and each reaction family is assigned by an adjustable rate parameter.

The simulation results for LCO gas oil hydrotreating show a good agreement with the experimental data. These results are even more remarkable by the fact that the LCO gas oils present different molecular compositions, which remains problematic to reproduce in traditional lumping kinetic models. The proposed approach has even more advantages: as the complex feedstock is represented by means of a set of molecules, this molecular description can be retained throughout the entire reactor simulation. Moreover, the effect of the reactions can be simulated without a pre-defined reaction network, since the stochastic simulation algorithm generates the reaction network "on-the-fly", as the reactions proceed.

References

1. Liguras D.K., Allen D.T. (1989); Industrial & Engineering Chemistry Research, 28, 674-683.
2. Quann R.J., Jaffe S.B. (1992); Industrial & Engineering Chemistry Research, 31, 2483-2497.
3. Neurock M., Nigam A., Trauth D.M., Klein M.T. (1994); Chemical Engineering Science, 49, 4153-4177.
4. Trauth D.M., Stark S.M., Petti T.F., Neurock M., Klein M.T. (1994); Energy & Fuels, 8, 576-580.
5. Quann R.J., Jaffe S.B. (1996); Chemical Engineering Science, 51, 1615-1635.
6. Khorasheh F., Khaledi R., Gray M.R. (1998); Fuel, 77, 247-253.
7. Zhang Y. (1999); Ph.D. thesis, University of Manchester.
8. Hudebine D., Verstraete J.J. (2004), Chemical Engineering Science, 59, 4755-4763.
9. Verstraete J.J., Revellin N., Dulot H. (2004); Preprints of Papers - American Chemical Society, Division of Fuel Chemistry, 49, 20-21.
10. Van Geem K.M., Hudebine D., Reyniers M.-F., Wahl F., Verstraete J.J., Marin G.B. (2007); Computers & Chemical Engineering, 31, 1020-1034.
11. Verstraete J.J., Schnongs P., Dulot H., Hudebine D. (2010); Chemical Engineering Science, 65, 304-312.
12. Hudebine D., Verstraete J.J (2011); Oil & Gas Science and Technology – Revue d’IFP Energies Nouvelles, 66, 437-460.
13. Hudebine D., Verstraete J.J., Chapus T. (2011); Oil & Gas Science and Technology – Revue d’IFP Energies Nouvelles, 66, 461-477.
14. Pyl S.P., Hou Z., Van Geem K.M., Reyniers M.-F., Marin G.B., Klein M.T. (2011); Industrial & Engineering Chemistry Research, 50 (18), 10850-10858.
15. Pereira de Oliveira L., Verstraete J.J., Kolb M. (2012); Chemical Engineering Journal. 207-208, 94-102.
16. Pereira de Oliveira L., Trujillo Vazquez A., Verstraete J.J, Kolb M. (2013), Energy & Fuels. http://dx.doi.org/10.1021/ef300768u.
17. Shannon C.E. (1948); The Bell System Technical Journal, 27, 379-423, 623-656.
18. Gillespie D.T. (1976); Journal of Computational Physics, 22, 403-434.
19. López-García C. (2000); Ph.D. thesis, Université Claude Bernard - Lyon I.
20. López-García C., Hudebine D., Schweitzer J.-M., Verstraete J.J, Ferré D. (2010), Catalysis Today, 150, 279-299.
21. López-García C., Becchi M., Grenier-Loustalot M.-F., Païsse O., Szymanski R. (2002); Analytical Chemistry, 74, 3849-3857.
22. Korre S.C., Neurock M., Klein M.T., Quann R.J. (1994); Chemical Engineering Science, 49, 4191-4210.
23. Korre S.C., Klein M.T. (1995); Preprints of Papers - American Chemical Society, Division of Fuel Chemistry, 956-961.