Modeling and Optimization of A Semiregenerative Catalytic Naphtha Reformer
Modeling and Optimization of A Semiregenerative Catalytic Naphtha Reformer
Modeling and Optimization of A Semiregenerative Catalytic Naphtha Reformer
and Catalvsis
Introduction
Catalytic naphtha reforming is practiced extensively in the there is ample potential for optimization of a catalytic naph-
petroleum-refining industry to convert gasoline boiling-range tha reformer. A proper selection of operating conditions
low-octane hydrocarbons to high-octane gasoline compounds within plant constraints is essential to maximize the prof-
for use as high-performance gasoline fuel. This is accom- itability of the reformer. Due to the catalyst deactivation with
plished by conversion of n-paraffins and naphthenes in naph- time, the process assumes a transient nature and its optimiza-
tha to isoparaffins and aromatics over bifunctional catalysts tion results in a time-optimal problem. This makes the opti-
such as Pt/AI,O, or Pt-Re/Al,O,. Recent environmental mization problem more challenging to solve. However, in the
legislation in the United States has banned the use of lead as absence of an analysis of this kind, the unit may remain un-
an additive for boosting antiknock properties of motor fuel. der suboptimal operating conditions, resulting in significant
Coupled with these stricter environmental regulations, there economic losses.
has been a consistent increase in the demand for higher fuel Use of mathematical models as a tool for either off-line or
efficiency standards of engines. This requires the use of higher on-line optimization analysis is growing rapidly in the refin-
compression ratios in engines, and therefore motor fuel with ing and petrochemical industries. A mathematical model re-
an even greater octane number. These considerations have quires various amounts of process knowledge and investment
continually forced the refiner toward producing higher-oc- of time and effort, depending upon the level of complexity
tane-number products from their catalytic naphtha reform- incorporated into these models. The advantage of utilizing
ers. This can be achieved by reforming the naphtha under rigorous mathematical models as compared to empirical ap-
more severe conditions, but this will also cause an increase in proaches is related to the fact that the prediction accuracy of
the rate of coke deposition, resulting in the reduction of cycle rigorous models can be significantly superior over a wide op-
lengths of the catalyst. Due to these trade-offs and others, erating range. Hence, detailed mathematical models are fre-
quently employed for optimization studies. In this work, a
Correspondence concerning this article should be addressed to J. B. R i g s
rigorous mathematical model of a semiregenerative catalytic
Current address of U.Taskar: Aspentech, Inc., Houston, TX. reformer, based on fundamental physicochemical concepts,
Separalor gas
Naphlha charge
from feed
preparation unil I-@-Healer Healer Healer
Reactor ellluenl-lo-feed
exchanger (I
Separalor Stabilizer
u
has been developed and utilized to conduct optimization the carbon number range C , to C,,,.The reformer-reactor
studies. The model parameters were estimated on the basis charge is combined with a recycle gas stream containing 60 to
of data obtained from an industrial unit. The modeling of the 90 mol % hydrogen. The total reactor charge is heated, at
complex chemical reactions occurring on the surface of the first by exchange with effluent from the last reactor, and then
bifunctional catalytic naphtha reforming catalyst during re- in the first charge heater. The inlet temperatures of the beds
forming was the most intricate part of the overall modeling vary between 750 to 790 K, and the reactors are operated at
effort. Appropriate kinetic modeling of these reactions was pressures of about 20 to 30 atm. The molar recycle ratio stated
imperative in achieving the desired prediction accuracy of the in terms of hydrogen to pure hydrocarbon feed varies from
model. A number of different approaches of varying levels of 4 1 to 8:l.
sophistication have been developed in the past to model the The major reactions in the first reactor, such as dehydro-
reforming chemistIy (e.g., Smith, 1959; Krane et al., 1960; genation of naphthenes, are endothermic and very fast, caus-
Kmak, 1972; Marin et al., 1983; Ramage et al., 1987). In ad- ing a very sharp temperature drop in the first reactor. For
dition, a variety of modeling schemes have been attempted to this reason, catalytic reformers are designed with multiple re-
represent the deactivation phenomena on the surface of the actors and with heaters between the reactors to maintain re-
reforming catalyst (DePauw and Froment, 1974; Beltramini action temperature at operable levels. As the total reactor
et al., 1991). The kinetic scheme employed in this modeling charge passes through the sequence of heating and reacting,
work was largely based on previously published studies. the reactions become less and less endothermic and the tem-
perature differential across the reactors decreases. The efflu-
ent from the last reactor, at temperatures from 750 to 790 K,
is cooled to 315 to 320 K, partly by heat exchange with the
Catalytic Naphtha Reforming-Process and reactor charge. The stream then enters the product separator
Chemistry where flash separation of hydrogen and some of the light hy-
The process flow diagram of the reformer modeled in this drocarbons (primarily methane and ethane) takes place. The
work is shown in Figure 1. At the core of the reforming flashed vapor, containing 60 to 90 mol % hydrogen, passes to
process are three or four fixed-bed adiabatically operated re- a compressor and then circulates to join the naphtha charge.
actors in series that conduct the solid catalyzed vapor-phase Excess hydrogen from the separator is sent to other hydro-
reforming reactions. This is a semiregenerative type of unit, gen-consuming units in the refinery. The separator liquid,
that is, the catalyst is regenerated periodically to compensate comprised mostly of the desired reformate product but also
for the loss in activity of the catalyst due to coke deposition. containing light gases, is pumped to the reformate stabilizer.
The naphtha used as a catalytic reformer feedstock usually Reformate off the bottom of the stabilizer is sent to storage
contains a mixture of paraffins, naphthenes, and aromatics in for gasoline blending.
coking kinetics on the surface of the catalyst. A deactivation Figure 2. Reaction patch for C8.
-98
77-
-92
-80
710
02 04
Fracbondc.wVa
06
WO@
I
08 1
'7
75
750 755 760 765 770
I
775 780 785
788
790
Thrd bed miet temperature. K
Figure 4. Temperature profile through reactor system.
I
year) was noted. The operating conditions prevalent at the 4
lli Hydrogen
base case (Table 3) were maintained over the entire period. 10:
As expected, the tendency for coke formation increases in II
going from the first to the last bed due to higher average 9:
temperatures in each successive bed and the accumulation of 8-
coke-forming components.
7-
The sensitivity of the model output variables, such as oc-
tane number and reformate yield, to variations in process
variables was studied over a wide operating range. A sensitiv-
ity analysis of this type provides an estimate of the gain of
the process with respect to each of the process variables. Due
750 755 7M, 765 770 775 780 785 790
to the lack of available operating data in different operating mlrd bed mlet temperature. K
ranges, it was important to at least ensure that the process
gains predicted by the model were in the proper direction. As Figure 6. Sensitivity of products to third-bed inlet
an example, Figure 6 indicates the model-predicted varia- temperature.
tions for the research octane number and the volumetric re-
formate yield, with changes in the third-bed inlet tempera-
ture at the start-of-cycle conditions. As is usually observed, maintaining the operating conditions that would achieve a
the octane number of the reformate rises with inlet bed tem- proper balance between severity and catalyst activity.
perature, but at a loss of the reformate yield. Figure 7 plots
the octane number over the entire run length of the catalyst,
but at different third-bed inlet temperatures with other con- Formulation of the Optimization Problem
ditions being maintained constant. At higher inlet bed tem- The objective function was equated to the profitability of
peratures, the octane number is higher in the beginning, but the unit by considering the prices of products and costs of
drops more rapidly later during the run. This is the result of utilities (Table 6). However, only the terms that could be in-
significantly higher rates of coke formation expected at ele- fluenced by varying the operating variables were included.
vated temperatures. These results indicate the importance of The most significant product was the reformate, whose price
was related to the quality, that is, the octane number. Other
by-products in the process, such as hydrogen and cracked light
gases, were also considered. The fuel cost and compression
cost in the recycle gas compressor were accounted for in the
objective function. The total objective-function value was
evaluated by integrating the instantaneous value of the objec-
tive function from start of cycle till shutdown. The CPU time
on a DEC-Alfa 150-MHz PC, required for computation of
the total objective function over a one-year cycle length, was
15-20 min, depending on the operating conditions employed.
The objective in the optimization analysis was to find the
operating policies within operational constraints such that the
5
formulated objective function would be maximized. The deci-
n
sion variables for each optimization run were the four reactor
0 0.2 0.4 66 68 1 inlet temperatures and the total recycle ratio. Some of the
F r w W cawyst
commonly encountered constraints in the actual operation of
Figure 5. Profile of main component types through the catalytic naphtha reformer were enforced. The upper
reactor system. bounds of 790 K were considered on the reactor inlet tem-
I
*Heavy means 10% increase in aromatics and 10% decrease in paraffins
in feed.
**Light means 10% decrease in aromatics and 10% increase in paraffins
in feed.
'Nominal operation involves using same inlet bed temperatures (767 K)
for all the beds and a recycle ratio of 8.73 over the entire period of
Figure 8. Optimization procedure. operation.
75-1 c75
0 1 o 0 o 2 o o o 3 o 0 O 4 o o o x w x ) 8 o o o 7 o 0 o 8 o o o
Figure 11. Coke content profiles for time-optimal model Tme.hours
of operation.
Figure 12. Product profile for fixed-octane mode.
Sensitivity Analysis
Fixed-Octane Mode The optimization cases studied here clearly illustrate the
The base case for the fixed-octane mode involved setting potential for improving the profitability of the unit by con-
the inlet bed temperatures at the same value, but allowing ducting an analysis of this sort. However, the results are sub-
that value to vary with time along with the recycle ratio in ject to the accuracy of the model coefficients, particularly the
order to maintain the reformate near 95 octane for the length rate parameters of the reforming and coking reactions. A
of the run. The time-optimal case was solved by using the model-parameter sensitivity study was conducted to assess the
control vector parameterization procedure for the reactor impact of errors in some of the main parameters on the ob-
temperature and the recycle ratio such that the variation be- jective-function value. A 10% relative error was introduced
tween the product octane and the target (95 octane) was min- in each parameter separately. The decision variable values
imized. It was found that the time-optimal case was able to found after the optimization analysis on this model using the
maintain the specified octane only until the time equaled time-invariant mode were substituted back into the original
6,500 hours instead of the full 8,000-hour run length. model (i.e., the model without any errors) and the objective-
The optimal results for the fixed-octane mode were ob- function value was reevaluated. As expected, this objective
tained by setting each of the four inlet bed temperatures in- function value shows some deterioration from the original
dependently. The optimizer searched for the maximum profit objective function value, which was found by performing the
conditions, while the 95 octane constraint was considered by optimization analysis on the error-free model. The results of
adding a penalty function in terms of the reformate octane the sensitivity analysis are presented in Table 8. The uncer-
deviation to the economic objective function. tainties associated with the activation energies of the main
Figure 12 shows the optimal results for the fixed-octane reforming reactions showed a lower level of impact on the
mode. Note that there is some variation in the octane num- value of the objective function. The uncertainties associated
ber of the reformate, but it generally stays close to a 95 oc- with deactivation parameters had the largest effect on the
tane. Note that after 3,000 hours of operation, the reformate results. However, in all cases the magnitude of the deteriora-
yield decreased slowly from 82.5% to 80% in order to main- tion in the objective function indicated that the process prof-
tain the specified octane number. The optimal results showed itability was not unduly sensitive to uncertainty in the model
only a 2.5% improvement in the economic objective function parameters.
Greek letters
y =specific gravity of a true boiling-point curve function
Conclusion y,, = stoichiometric coefficient of species i in reaction j
AH,. =heat of reaction of ith reaction. kcal/kmol
A rigorous mathematical model of a semiregenerative cat-
alytic naphtha reformer employing a detailed kinetic scheme
was developed. The model was benchmarked with the indus-
.XI