Proceedings of PSE'94 (439-444) KINETICS BASED SIMULATION AND OPTIMISATION PACKAGE FOR INDUSTRIAL CATALYTIC REFORMERS 1 1* 2 Harendra Singh, M.O. Garg, A.K. Saxena, 22. 2 G. Das, H.B. Goyal and V.K. Kapoor 1. R&DCentre, Engineers India LUnited, Gurgaon, India. 2. Indian Institute of Petroleum, Dehradun, India. ABSTRACT A kinetics based simulation and optimisation package useful for design and optimal o;:>eration of semi-regenerative type catalytic reformers has been developed. The model is based on a.total of 22 lumps and 40 reaction schemes, superimposed by a comprehensive catalyst deactivation kinet~cs' and the prediction of residual ca:alyst liie. This model is linked in a sequential modular flowsheets sir.;ulator for predicting the performance of other reformer sub-systems - furnaces, cornpressor, debutaniser etc, besides product properties and unit economics. SQP algorithm has been provided to carry out both, off-line and on-line optimisation of the unit within the different plapt constraints. Four case studies have been presented to illustrate the various applications of this package. INTRODUCTION Ca talytic reforming has established itself as a major secondary processing facility in a modern refinery. Itcontributes significantly towards the overall refinery economics, particularly, in terms of higher octane barrel of gasoline and/or particularly pure aromatics (BTX) for petrochemicals feedstock. However, the process is highly energy intensive and in the present competitive environment, it becomes necessary to ensure optimal utilisation of the unit, while keeping the operating costs at the minimum. Towards this end, modelling. simulation and optimisation is the best cost effective solution to maintain the competitive edge in the industry. With the above in view, a simulation and optimisation package useful for design and optimal operation of semi-regenerative type catalytic reformers has been developed. This has been extensively tested with pilot plant and commercial data on both. Pt/ A1203 and Pt-Re/ A1203 catalysts. The package has been customised for IBM/PC 386 or equivalent and is provided with user-friendly interactive input/ output screens with graphical output. It can also be implemented on-line and interfaced sui ta bl y with advanced control system. This paper highlights the salient features of the package and presents four case studies which illustrate its application - from unit monitoring to unit optimisation and revamp/retrofit studies. Finally several future developments have been planned for the package. Use of Bellman I s theory of Dynamic Programming for establishing the optimal policy over the complete catalyst cycle is one such example. LUMPING SCHEME. AND REACTION NETWORK Lumping Scheme The feed to a catalytic reformer is typically straight run naphtha which usually contains several hundred chemical components. The modelling of reaction for such large number of individual components would be overwhelmingly combersome and perhaps not disirable, and thus it is appropriate to lump these components into a minimum set of kinetics lumps; the philosophy of lumping being largely governed by the similarity in reaction characteristics. One popular approach to lump hydrocarbon species is by hydrocarbon types, Le., paraffins, naphthenes and aromatics. Within each of these types, the reacti vi ties of di feerent species is strongly dependent on carbon number and, hence, in our model we have divided paraffins into five lumps from C5 to C9+. The CIO and higher components have been lumped along with the corresponding * Author to man all ca:rrespnjeoce moold be a:bresse:i... ~ r $. - 439- r ~Ar. r... K TkL 5""'" kvv""cj:..'c",",oji j"i" t;:.\.'u"""," O'VI PYD~ :>v-~ 1-;::""-~l4) jo'""'(j'-'i KoY~ I 30 r1~-,"j~ I \i<1 ~:J KO'[tCVv '1rv.J);t~~ "'\. ~~(J. E:.~.
C9+ species since the variation in selectivity of paraffins and naphthenes towards the aromatics diminishes with increase in carbon number. In case of naphthenes, reactivities of alkylcyclopentanes (ACP) is lower than those of alkylcyclohexanes (ACH), and hence naphthenes are further subdivided into ACP and ACH. Following the above, the reformer feed is th us characterised by 17 lumps which in addition to hydrogen and C1 to C4 components ma ke a total of 22 lumps. The reaction network for all the 22 lumps is shown in Figure 1. Reaction Network The break up of the above reaction network leads to 40 parallel reactions covering the fi ve major reforming mechanisms; these are, dehydrogenation of naphthenes to aromatics, isomerisation of paraffins and naphthenes, dehydrocyclization of paraffins, and hydrocracking and hydrogenolysis of paraffins..hougen Watson type to describe rate the expressions have reaction kinetics been of all used the a hove reactions. The rate constants in these expressions have been estimated based on pilot plant data as well as those reported in the li tera ture, for both, mono-metallic (Pt/A1203) and bi-metallic (Pt-Re/A1203) catal ysts. The equilibrium constants are expressed as a function of temperature and are based on the thermodynamic properties of the lumps considered. CATALYST DEACTIVATION In a semi-regenerative type catalytic reformer, the activity of the catalyst and its selectivity for the desired product decreases progressively over a period of time due to deposition of coke on catalyst surface. Beltramini (1991) i-:as proposed a mechanistically based dual site deactivation model, describing the decay in both metal and acidic activities. The ma thema tical framework provided by Beltramini has been used in our work; however, it has been suitably modified to cover the following aspects: i) effect of feed composition and temperature on acidic and metallic site fouling; ii) influence of hydrogen partial pressure on acidic and metallic site cleaning;..., a:'j effect of coking on coking reactions; iv) effect of variation in acidic and metallic acti vities in different s. The deactivation model is tuned to a particular unit by evaluating the constants in the model based on actual commercial data. REFORMER FLOW-SHEET SIMULATOR In addition to the model, separate models have been developed for other reformer subsystems; furna.ces, feed/ effluent heat exchangers, flash drum, recycle gas. compressor and the debutaniser. These models have been embedded in a sequential modular flowsheet simulator with the recycle loop as the tear stream. The package, after convergence, computes the products yields and properties as well as an economic summary for the unit. Among other things, the economic summary also accounts for the effect of residual catalyst cycle length. A simplified information flow diagram of the simulation package is given in Figure 2. The calculation of residual catalyst cycle length is such as, base~ space on various severity parameters, velocity, feed quality, hydrogen purity and hydrogen partial pressure, integrated average bed temperature etc. A quantitative estimation of these parameters on residual catalyst cycle length has been derived based on pilot plant studies. REFORMER OPTIMISATION The reformer simulation package described above is a valuable tool for optimising the operation of a unit with respect to the severity variables. However, such a study would have to be carried out within the present plant and operating constraints. For this purpose, the repetitive use of the model to generate several possible alternatives can become tedious and time consuming and calls for development of an automated procedure. With the above in view, the simulation package has been suitably interfaced with a general purpose optimisation algorithm. Sequential Reduced Quadratic Programming (SRQP) algorithm of Biegler and Cuthrell (1985) has been used for this purpose. This algorithm is quite robust and has more or less established itself as an industrial standard. The optimisation of the reformer unit would typically be carried out by maximising the daily profit; however, the user can also choose other objectives, such as, reformate yield for maximisation. In all of these, the corresponding effect on catalyst residual cycle length will also be evaluated and it is thus, suggested that the objective function should be appropriately weighted for the residual catalyst cycle length. In an extreme case, it is possible to maximise the catalyst cycle length alone. The maximisation of the objective would always be carried out within the operating and plant hardware limits, such as those of the recycle gas compressor, furnaces, debutaniser etc. APPLICATION OF THE PACKAGE The simulation and optimisation package as described above can be used, both, off-line and on-line. For off-line applications, the package has been customised for IBM/PC-AT 386 or compatibles and has been provided with a user-friendly interactive input/output screens with graphical displays. Typical off-ine applications of this. package are as -440-
follows: i) Evaluation of alternative feedstocks It is quite obvious, that such a study with a stand alone 'simulation package wou1:d have taken several runs and even then it would not have been possible to ensure an optimum. ii) Prediction of optimum cycle lengths Revamp Study Hi) Revamp studies scheme) iv) Sensitivity analysis ( with no change in flow The package can also be implemented on-line. In this mode, it is required to provide suitable interfaces for data retrieval, data reconciliation etc., and above all for deposition of the optimum values to the unit through currently available multi variable predictive control algorithms, such as, DMC, IDCOM-M etc. EXAMPLES STUDIES The above package has been extensively tested commercially for a wide variety,of off-line studies. Herein, we summarise four (4) examples which demonstrate the. capabilities of the package. Pilot Plant Sllnulation To begin with, the model predictions were validated against carefully planned pilot plant runs; the purpose was to check the sensitivity of the model predictions with respect to severity variables. From the results presented in Table I, it is seen that the model predictions a"'e in good agreement wi th the pilot plant data. Commercial Unit Simulation A commercial semi-regenerative type reformer in a refinery was simulated. This unit using bi-metallic catalyst produces reform ate which is subsequently solvent extracted to recover benzene and toluene. Both, operating and yield data was collected at start of run (SOR) a:jd later on at 100 and 140 days into the operation. Table 2 compares the plant data with simulated values. The close match of the simulation results establishes the applicability of kinetic rate constants for commercial operation. Moreover, the simulated results for stages I a:jd II illustrate the accuracy of the catalyst deactivation kinetics in our model. Reformer Optimisation A detailed study was undertaken for optimising the operating conditions of a commercial reforming unit. The purpose was to maximise the profit at the present operating levels of fresh feed and recycle rates given the constraints of temperature ranges imposed by charge heater and inter furnaces. The results of this study are summarised in Table 3. It may be mentioned that a relatively low weightage was given to the catalyst cycle length in the objective function, due to which, the last inlet temperature reached the upper bound. The study indicated a net increase in profit of around 27% above the base case. Finally, the package was used to evaluate al terna ti ve strategies for enhancing the throughput of an existing SR type reformer. Thisreformer operates with mono-metallic catalyst. The possible altern ati ves considered are as follows: i) Increase of throughput in the first ; H) Divt:rt additional feed to the tail. The results of the above study are summarised in Table 4. As per Case 2 it was observed that the required first inter heater duty was higher than the design duty and also there was significant decrease in the catalyst cycle length without associated improvement in the yields. However, in Case 3 where additional feed was directed to the last, hydrogen purity improved and the decrease in the cycle length relative to base case was found to be less than Case 2. Even this decrease can be ov~rcome to a large extent by increasing the H2/HC 'ratio as illustrated by Case 4. It is interesting to note that at the present throughput level, di verting 10% of th e feed to the last resul ts in the increase of cycle length by about 13% at similar yield levels (Case 5). The above study was specifically undertaken to optimally utilize the tail which otherwise appeared to be under-utilised. This example provides a good illustration of the use of this package for carrying out such revamp studies. FUTURE DEVELOPMENT Improvementsare being made in the basic kinetic'. schemes as more understanding is gained about..the reforming reactions. The inclusion of multiobjective function - profit, yield and catalyst cycle length is a classical case of pareto solutions, and thus, use of the developments in optimization theory in management sciences, are being' explored. The slow deactivation of the catalyst over a cycle presents an interesting case of a multistage decision process (Roberts, 1964), wherein the unit can be optimized at each stage of the cycle. The dynamic programming principles of Bellman (1957) can thus be conveniently applied to optimize the entire reformer operation, from startup to shutdown and subsequent regeneration; this would intrinsically lead to optimal utilization of catalyst activity. Lastly, the steady state m.odel can be extended -441-
to include variation with time, Le. a dynamic model of reforming unit. This model can then be used for several, important applications, such as startup and shutdown, control system synthe:;;is and design, operability studies, design of advanced control".systems and hazard anal y :;;is. FEED INFORMATION OPERATING DATA ECONOMIC DATA. ACKNOWLEDGEMENTS The kind permission granted by Director (lip, Dehradun) and the management of Enginees India Limited to publish this paper is gn.tefully acknowledged. LITERATURE CITED [ 1] Beltramini, J.N., Wessel, T.J., Datta, R., AIChE J., 37 (6), 845 (1991). [2] Biegler, L.T. and Cuthrell, J.E., COr:1p. & Chern. Eng., 9(3), 257 (1985). [3 ] Bellman, R., Dynamic Programwing, Princeton University Press, New Jersey (1957). [ 4] Roberts, S. M., Dynamic Programming in Chemical Engineering and Process Control, Academic Press, New York (1964). PRODUCT QUALITY CALCULATIONS ~NH9+~A9+ - P9+ ~ 1 ~ NP9+ RECYCLE GAS COMPRESSOR HP CALCULATION ~NH8~A8 p~ 1~ NPB ~NH7~A7 Py NH6-.... A6 FEED/EFFLUENT EXCHANGER CALCULA TIONS RESIDUAL RUN LENGTH PREDICTIONS P6~ 1 ~ ~NP6 i.i....j.....; po = Paraffins NP = Cyclo Pentanes.'. = Aromatics NH = Cyclo Hexanes :'~umber indicates the carbon number FIGURE 1: REACTION NETWORK. FIGURE 2: CRU MODEL FLOW DIAGRAM -442-
TABLE 1: COMPARISON OF MODEL PREDICTIONS WITH PILOT PLANT DATA Case Base Case 1 Case 2 Pressure, Kg/cm2 abs 28.5 26.0 23.0 Temperature, C 485 482 480,Ex- yields (Wt% on feed) Experimental Predicted Experimental Predicted Experimental Predicted H2 0.92 0.88 0.95 0.85 0.94 0.87 C5+ 77.74 78.28 79.10 79.20 80.90 80.62 C6A 2.09 1.'94 1. 92 1.82' 1. 81 1. 73 C7A 12.00 11.61 11. 90 11.40 11.72 11.28 C8A 21.33 21.26 20.70 20.63 20.60 20.37 C9A 4.27 4.27 4.30 4.21 4.36 4.18 TABLE 2: SIMULATION RESULTS FOR A COMMERCIAL SR REFORMER SOR STAGE I STAGEII ~---------------------------------------------------------------------------------------- Time, hrs 0 2376 3312 Plant Model Prediction Plant Model Prediction Plant Reactor Temperature drops; C Model Prediction Reactor it 1 78 78.05 76 77.54 78 76.59 Reactor # 2 34 34.22 31 30.80 35 34.98 Reactor # 3 6 6.31 3 5.48 6 2.88 Total Del T 118 118.58 110 113.82 119 11445 Ex- yields; Wt% on feed Hydrogen 1. 78 1.62 1.64 1.38 1.93 1.42 C5+ 88.63 88.43 82.84 81. 76 83.86 85.02 Benzene 24.54 24.82 26.28 25.93 31. 08 29.94 Toluene 17.87 17.06 12.66 11. 65 13.75 11.75 ------------------------------------------------------------------------------------------------- -443-
TABLE 3: RESULTS OF OPTIMISATION STUDY Variable Base Minimum Limit Maximum Limit Optimum Feed rate, MT/D 900.0 900.0 900.0 900.0 Recycle rate, MT/D 521.0 521.0 521.0 521.0 Reactor inlet temperature 's; C #1 498.0 495.0. 510.0 498.31 #Z 498.0 495.0 510.0 500.14 #3 498.0 495.0 510.0 510.00 Separator Pressure, 16.0 16.0 16.5 16.0 Kg/cmZ abs N etprofi t Base :;" Ba.se:'-+ Z7%. TABLE 4: STUDIES FOR THROUGHPUT INCREASE IN A COMMERCIAL REFORMER Case Ex- yields u -------------.------------- HZ C5+ Benzene Toluene Interheater Duties, Gcal/hr I II Inter Inter heater heater Hydrogen Purity #1 Normal feed 0.44 76.08 40.77 5.58 2.789 0.019 58.00 1.0 to first (HZ/HC=5.5) #Z 10% additional 0.46 76.99 40.76 5.47 3.039 0.040 59.52 0.81 feed to first Relative Cycle length #3 10% additional 0.50 78.49 40.71 5.37 2.687 0.820 62.20 0.88 feed to last (HZ/HC=5.5) #4 10% additional 0.49 78.28 40.71 5.34 2.705 0.780 62.19 0.98 feed to last (HZ/HC=5.9) #5 90% of normal 0.48 77.62 40.72 5.47 Z.440 0.794 60.89 1.13 feed to first and 10% additional feed to last -444-