18th European Symposium on Computer Aided Process Engineering – ESCAPE 18 Bertrand Braunschweig and Xavier Joulia (Editors) © 2008 Elsevier B.V./Ltd. All rights reserved. A study on naphtha catalytic reforming reactor simulation and analysis Rita M. B. Alves, Fabio Menten, William S. Maejima, Roberto Guardani, Claudio A. O. Nascimento LSCP/CESQ - Department of Chemical Engineering, Polytechnic School University of São Paulo,Av. Prof. Luciano Gualberto, n. 380, trav. 3, CEP 05508-900 São Paulo, SP, Brazil. e-mail address: rita@lscp.pqi.ep.usp.br; oller@usp.br Abstract An industrial naphtha catalytic reforming unit with four fixed-bed reactors in series, in which a number of chemical reactions occur, is analyzed. Kinetics and thermodynamics equations are selected to describe the naphtha catalytic reforming reactions characteristics based on idealizing the complex naphtha mixture by representing the paraffin, naphthene, and aromatic compounds by lumps. Based on industrial plant information and the knowledge of the main reactions that occur in the process, all the lumps are defined in order to describe the phenomenological model. For each reaction, a kinetic expression is formulated as a function of product yield, reaction conditions and kinetic constants. All reactions are assumed to be pseudo-first order with regard to the hydrocarbon. The equations for all reaction steps are combined into a set of differential equations which comprising the kinetic model, which is incorporated into a onedimensional pseudo-homogeneous model for multiple reactions. The kinetic parameter values are estimated using industrial data obtained in a petroleum refinery. The process analysis was undertaken by using a one-year database. The data were collected every 10 minutes. Noise and gross errors, commonly found in industrial processes, could disturb the model fitting and lower its quality. Thus, they were detected and eliminated. Since the industrial data were not uniformly distributed over the ranges of the process variables included in the model, a mapping technique was used, which consists of fitting a neural network model to the industrial data, and using this model to generate new values of the process variables with adequate resolution, and uniformly distributed along their application range. These values were then used to fit phenomenological model parameters. The simulation results based on the proposed model agree very well with actual operating data of the industrial unit. Keywords: Naphtha catalytic reforming, Kinetics, Simulation and Analysis, Gross Errors, Neural Network. 1. Introduction Catalytic reforming of straight run naphtha is a very important process for producing high octane gasoline and aromatics in petroleum-refining and petrochemical industries. Hydrogen and lighter hydrocarbons are also obtained as by-products. Generally, the reforming is carried out in three or four adiabatic fixed bed reactors in series with intermediate preheaters. The fourth reactor is usually added to some units in order to allow an increase in severity of throughput while maintaining the same cycle length, which in the case of those plants operating with semiregenerative mode is about one year (Ancheyta et al., 2001). 2 R. M. B. Alves et al. The reforming feedstock is a complex mixture composed of normal and branched paraffins, five- and six-membered ring naphthenes, and single-ring aromatics, having carbon number ranging from 6 to 11. A large number of reactions occur in catalytic reforming, such as dehydrogenation and dehydroisomerization of naphthenes to aromatics, dehydrogenation of paraffins to olefins, dehydrocyclization of paraffins and olefins to aromatics, isomerization or hydroisomerization to isoparaffins, isomerization of alkylcyclopenatnes and susbstituted aromatics, and hydrocracking of paraffins and naphthenes to lower hydrocarbons. The catalytic reforming process is carried out at elevated temperatures, moderately high pressure in the presence of circulating hydrogen. A detailed kinetic model is very difficult to obtain due to complexity of the feedstock and reactions that take place. Thus, it is usual to assume that only three classes of hydrocarbons, the so called kinetic lumps, are present in naphtha (paraffins, naphthenes, and aromatics), and they are considered to have similar properties and kinetic behavior. Several models considering different levels of sophistication have been developed to represent the kinetics of catalytic reforming reactions (examples: Smith, 1959; Krane et al., 1959; Ramage et al., 1987; Padmavathi & Chaudhuri, 1997). Each of these models uses its own kinetic expressions or others already reported with or without some modifications. The main objective of all these models is to determine the operational conditions and to predict reformate yield and reactor temperature profile accurately. In this work, the simulation of the catalytic reforming process is carried out using the kinetic model proposed by Krane et al. (1959), which is one of the most elaborated models since it considers all possible reactions for each individual hydrocarbon. 2. Methodology The methodology for modeling naphtha catalytic reforming units based on historical plant data is schematically presented in the diagram in Figure 1. The industrial process studied is a Petrobras refinery catalytic reforming unit whose simplified process flowsheet is shown in Figure 2. 2.1. Kinetic Model The developed model is based on Krane’s kinetic scheme (Krane et al., 1959) and utilizes lumped mathematical representation of all possible reactions that take place. These representations are written in terms of isomers of the same nature (paraffins, naphthenes, and aromatics). Based on industrial plant information and the knowledge of the main reactions that occur in the process, all the lumps are defined in order to describe the phenomenological model. These groups range from 1 to 10 carbon atoms for paraffins, and 6 to 10 carbon atoms for naphthenes and aromatics. An additional lump representing one of the most important precursors of benzene, methyl cyclopentane, is added. The model developed in this work includes 20 lumps and 53 chemical reactions. For each reaction, a kinetic expression is formulated as a function of product yield, reaction conditions and kinetic constants. All reactions are assumed to be pseudo-first order with regard to the hydrocarbon. The equations for all reaction steps are combined into a set of differential equations which comprises the kinetic model. The kinetic parameter values are estimated using industrial data obtained from industrial operation. The analysis of the process was based on a one-year database. The data were collected every 10 minutes. Noise and gross errors, commonly found in industrial processes, could disturb the model fitting and lower its quality and were previously detected and eliminated. This step is extremely important since the quality of the data has direct influence on the quality of the parameters estimation. Knowledge of the A study on naphtha catalytic reforming reactor simulation and analysis 3 process, statistical procedures and first principles equations are used (Alves and Nascimento, 2007). Industrial Process Analysis Main Process Variables Data Collection Historical Plant Data Process Modelling and Simulation Data Treatment Statistical Tools Kinetic Parameters Fitting First Principles Model Treated data Kinetic model Reactor model Temperature profile Reformate composition Fig. 1- Methodology for modeling the naphtha catalytic reforming unit Hydrogen Recycle Reactor 1 Heater Gas Separator Reactor 2 Heater Heater Reactor 3 Reactor 4 Heater pre-treated Naphtha Heat Exchanger Cooler Separator Fig. 2 – Naphtha Catalytic Reforming Process Diagram to Reformate Stabilizer 4 R. M. B. Alves et al. Since the industrial data were not uniformly distributed over the ranges of the process variables included in the model, a mapping technique was used, which consists of fitting a neural network model to the industrial data. This model is a sequential modular simulator able to evaluate operational conditions, feed and product streams quality, catalyst performance, product yield and cycle length (Silva, 2002). Each module is represented by a specific neural network simulating the features to be considered in the catalytic reforming process. A three-layer feed-forward neural network was adopted. The fitting was based on the back-propagation algorithm. The model was then used to generate new values of the process with adequate resolution and uniformly distributed along the range of the variables. These values were then used in the fitting of the phenomenological model parameters. The minimization of the objective function, based on the sum of square errors between experimental and calculated yields, was applied to select the best set of kinetic parameters. This objective function was solved using the least squares criterion with a non-linear regression procedure based on Marquardt’s algorithm. 2.2. Process Model The kinetic model was incorporated into a physical model proposed to describe the reactor. A fixed-bed unidimensional pseudo-homogeneous adiabatic reactor model was considered. It asssumes concentration and temperature gradients in the axial direction only, and the only transport mechanism that takes place is the uniform flow due to the fluid flux. Mass and energy balance equations were solved assuming steady state operation. A set of ordinary differential equations was established. When integrated, these equations provide the composition over the whole reaction zone and temperature profile along the length of the reactors. The fourth-order Runge-Kutta method was used to solve the model. This process model is able to successfully simulate commercial semiregenerative reformer operations. 3. Results and Discussion Figures 3a and 3b show examples of the comparison between measured and calculated output variable values using the neural network model. Good agreement between modeled and measured values can be observed and the uniform distribution indicates absence of any tendency in fitting. (a) Fig.3a-b - Neural Network Simulation Results (b) A study on naphtha catalytic reforming reactor simulation and analysis The simulated process variables are then used for fitting the phenomenological model parameters. Since these data represent accurately the industrial unit and the neural network model into takes account all important features such as catalyst activity, cycle length, and pressure and temperature profiles, the fit kinetic constants include the effects of these variables. Concerning the simulation using the proposed kinetic model, the developed program is evaluated first by using data from literature (Krane et al., 1959). Figure 4 shows the comparison between experimental data and calculated conversions of whole naphtha, considering Krane’s information. Solid lines represent the calculated values and the symbols represent the experimental data; the maximum deviations occur in predicting aromatics with nine and more carbon atoms (A9+). It can be seen that, as the naphtha passes through the catalyst bed, the concentration of all aromatic compounds increases. The concentration of heavy paraffins decreases as they undergo conversion. Heavy Paraffins Concentration 16 P7 P8 P9 P10 (%) 12 8 4 0 0 10 20 30 40 50 60 Volume Reactor 70 80 90 100 Aromatics Concentration 30 (%) 22.5 15 A6 A7 A8 A9+ 7.5 0 0 10 20 30 40 50 60 Volume Reactor 70 80 90 100 Fig. 4 – Concentration profile over the reactor volume Table 1 shows the reformate composition predictions obtained from the proposed model. Industrial data are also included for comparison. It can be observed that both values agree well, with less than 1.0 mol% average absolute deviation. 4. Conclusion The results of the present study indicate that the proposed model is able to accurately simulate the operation of the naphta reforming industrial unit considered. The model has important features, since it is based on first principle phenomena, with parameters obtained by fitting the model to industrial operating data. Thus, the model embraces specific and not explicit effects on conversion caused by factors associated with the industrial operation itself, like geometric peculiarities of the industrial equipment or catalyst properties, among others. Moreover, the commonly encountered problem of 5 6 R. M. B. Alves et al. maldistribution of process variable values over the range of interest was overcome by fitting a neural network to the original data, and then using the neural network model to generate uniformly distributed values of the process variables over the range. This method resulted in an accurate fitting of the reactor model parameters, with the generation of an accurate model that can be used in studies aimed at optimizing the operation of the industrial unit. The method adopted in this study is innovative and can be recommended for the fitting of hybrid models adapted to existing industrial units. Table 1- Reformate Composition (mol%) – Model vs. Experimental Data Reformate Component P5 P6 P7 P8 P9 P10 N6 N7 N8 N9 N10 A6 A7 A8 A9 A10 Feed 2.27 16.52 27.06 9.37 0.31 0.10 12.26 17.56 9.88 0.51 0.00 0.77 2.74 0.65 0.00 0.00 Industrial Data 7.41 19.61 15.33 2.35 0.03 0.00 1.98 0.86 0.07 0.00 0.00 9.99 27.96 13.42 1.01 0.00 Model 9.45 16.95 12.75 1.82 0.04 0.00 0.87 0.88 0.12 0.00 0.00 9.88 30.75 15.87 0.58 0.03 Deviation -2.04 2.65 2.58 0.52 -0.01 0.00 1.11 -0.03 -0.05 0.00 0.00 0.11 -2.79 -2.45 0.43 -0.03 Acknowledgement The authors gratefully acknowledge the efforts and contributions of many individuals in PETROBRAS, mainly the Eng. José Carlos da Silva. The authors also thank FAPESP and FUSP for their financial support. References R. M. B. Alves and C. A. O. Nascimento, 2007, Analysis and Detection of Outliers and ystematic Erros in an Industrial Data Plant, Chemical Engineering Communication, 194, pp. 382-39 J. J. Ancheyta, M. E. Villafuerte, G. L. Díaz, A. E. González, 2001, Modeling and Simulation of Four Catalytic Reactors in Series for Naphtha Reforming, Energy Fuels, 15, 887-893 H. J. Krane, A. B. Groh, B. L. Shulman and J. H. Sinfelt, 1959, Reactions in Catalytic Reforming of Naphthas, Proceedings of the 5th World Petroleum Congress, 39-51 G. Padmavathi, K. K. Chaudhuri, 1997, Modeling and Simulation of Commercial Catalytic Naphtha Reformers, Can. J. Chem. Eng., 75, 930-937 M. P. Ramage, K. R. Graziani, P. H. Schipper, F. J. Krambeck, B. C. Choi, 1987, KINTPTR (Mobil’s Kinetic Reforming Model): A Review of Mobil’s Industrial Process Modelling Philosophy, Adv. Chem. Eng., 13, 193-266 J. C. Silva, 2002, Análise e Otimização de Processo Industrial de Reforma Catalítica de Nafta via Redes Neurais, MSc thesis, University of São Paulo R. B. Smith, 1959, Kinetic Analysis of Naphtha Reforming with Platinum Catalyst” Chem. Eng. Prog., 55, 6, 76-80