You are on page 1of 8

Computers and Chemical Engineering Vol. 4, pp. 2512.

58
0 Pewmoo Press Ltd.. 1980. Printed in Great Britain

OPTIMAL WATER ALLOCATION IN A


PETROLEUM REFINERY

N. TAKAMA,T. KURIYAMA, K. SHIROKO and T. UMEDA


ChiyodaChemicalEngineeringand ConstructionCo., Ltd., Yokhama,Japan

(Received12 Ocfober 1979)

Abstract-A method for solving the planning problem of optimal water allocation is presented. All
the alternative systems are combined into an integrated system by employing structure variables or
split ratios at the point where a water stream is split into more than two streams. The values of
structure variables are determined for given process conditions so as to minimize the total cost,
subject to constraints derived from material balances and interrelationships among water-using units
and wastewater-treating units. The optimization problem is solved by using the Complex method.
The method is illustrated by its application to the water allocation problem in a petroleum refinery.

Scope-In the last decade, a number of studies on wastewater reuse or optimal designs of
wastewater-treating systems have been presented. Though those studies have received much
attention, they have been carried out exclusively on wastewater-treating systems without paying
attention to water-using systems. However, the authors’ extensive survey on the present status of
water use in a petroleum refinery has shown that there is enough room to reduce a huge amount of
both fresh water and wastewater. The reduction can be accomplished by optimizing water allocation
in a total system consisting of water-using units and wastewater-treating units. In this paper, the
problem of maximizing water reuse is considered as a problem of optimizing water allocation in a
total system. Furthermore, the problem of determining a system structure is defined as a parameter-
optimization problem by employing structure variables. Due to the approach, the difficulties
associated with combinatorial problems are resolved. A planning problem of water allocation is
discussed here under the assumption that process conditions are determined beforehand as the result
of optimization in a design problem.

Conemsions and SIgnhIcanc+The method is applied here to a simple but practical problem of
optimizing water allocation in a petroleum refinery. Since the problem is a large dimensional and
nonlinear problem with stringent inequality constraints, it is not practical to solve it by the exclusive
use of mathematical programming methods. In this paper, the problem is transformed into a series of
problems without inequality constraints by employing penalty function. In the course of optimiza-
tion, a sequence of non-feasible points is generated. Therefore, difficulties associated with an initial
search for feasible points are omitted and a feasible point is yielded by modifying the penalty
function successively. The Complex method is, in general, inefficient in searching for an optimal
point because of the premature termination on a nearly flat surface. However, the premature
termination is avoided by reducing a system structure successively according to the intermediate
results of optimization.

INTRODUCTION system approaches. The detailed review on these studies


Water has been used in abundant quantities by chemical, was made by Mishra et al.[6]. Much information on the
petrochemical, petroleum refining and other process in- optimization studies on process units for wastewater
dustries. However, in recent years, the increased cost of treatment can be acquired from this survey. In addition,
wastewater treatment to meet environmental require- there have been significant developments of the methods
ments and the scarcity of less expensive industrial water for processing system synthesis, such as those of heat
have provided process industries with strong incentive to exchangers and separation systems. A method of utiliz-
minimize the amount of water consumption and waste- ing the system structure variables[7] is considered to be
water discharge. The major concern is to emphasize the useful to eliminate difficulties due to combinatorial prob-
importance of water reuse and a number of efforts have lems. The studies presented so far, however, only cover
been made towards achieving the goal of extensive water wastewater treating systems. The amount of wastewater
reuse in various process industries. was given beforehand and its reduction was not taken
There have been presented many ideas for wastewater into consideration. As far as the authors know, the
recovery and reuse in the petroleum refining industry [ l- optimal design problem including water reuse for the
51. These papers have exclusively described wastewater- total system consisting of water-using system and was-
treating systems for the realization of zero discharge. As tewater-treating system has not yet been solved.
for the optimal design methods for wastewater-treating The authors have carried out an extensive study on the
systems, several attempts have also been made by using present status of water use in a typical petroleum

CACE Vol. 4. No. 4-D


251
252 N. TAKAMA et al.

refinery[8]. As the result, it has been shown that there is discharge to the environment on the basis of the total
enough room to reduce a large amount of wastewater by system, the water reuse policy has to be determined so
maximizing water reuse and wastewater recovery. Fur- as to minimize the sum of the costs of fresh water and
ther increase in the efficiency of water use can be expec- wastewater treatment. The amount of water used in a
ted by the change of process conditions in the refinery. water-using subsystem is closely connected with the
In solving such large complex problems, it has been performance of the subsystem. Therefore, it is not ap-
found expedient to use a two-level approach. propriate to decrease the amount of water used in the
The two-level approach starts with decomposing the water-using subsystem only from the viewpoint of water
original large-scale and complex optimization problem saving. As described in the preceding section, the
into several smaller subproblems, and allocating func- amount of water used in the water-using subsystem is
tions of decision making for upper and lower levels, determined on the design level. The optimization prob-
respectively. On the lower level, each subproblem is lem considered here is to select the best system among
independently optimized for given condition at each step all alternative systems.
of iterations. Those conditions are determined on the Figure 2 shows the general structure of the system
upper level so as to optimize the preset objective func- under consideration, Among N subsystems, the sub-
tion combining the objective functions of subproblems. systems 1 and N are an imaginary input and an output
Those problem and subproblems correspond to the subsystems, respectively. The imaginary input subsystem
problems on the planning and the design levels, respec- is a source of fresh. water supplied to any subsystem.
tively. For the system associated with water use and The imaginary output subsystem is a final holding basin
wastewater treatment, the overall optimization problem which collects wastewater from all other subsystems and
can be devided into two problems. One is to determine discharges it to the environment. Each subsystem has a
the optimal allocation of water to water-using and was- mixing point and splitting point. A flow from any splitting
tewater-treating units. The other is to determine optimal point is directed to any other mixing point.
process conditions in those units for a given water al- The overall problem is defined as to determine both
location. The problem of optimizing water allocation design variables of subsystems and structure variables in
corresponds to the planning problem on the upper level such a way that a given objective function is optimized.
and the problem of optimizing process conditions in Mathematically the synthesis problem under con-
those units corresponds to the design problem on the sideration is defined as follows:
lower level. In this paper, the planning problem is solved
under the assumption that the process conditions are Minimize N
given beforehand. Isi, dij 3 fitui, vi, di,&)

STATRMENT OF PROBLEM subject to


In general, a system associated with water use in a
petroleum refinery consists of two subsystems as shown (2)
in Fig. 1; one where water is used for petroleum refining
and the other where wastewater is treated for reuse or

TREATED WATER
Ui =
c=I
Sij ’ Vj (3)

NON -TREATED WATER

WATER . WASTEWATER
FRESH WATER DISCHAKE
- USING . TREATING
SUBSYSTEM SUBSYSTEM

Fig. I. System for refinery water use and treatment.

0 WATER USING SUBSYSTEM

m WASTEWATER TREATING SUeSYSTEM

Fig. 2. General system structure for refinery water use and treatment.
Optimal water allocation in a petroleum refinery 253

OBJECTIVEFUNCTION
2 &j = 1 For the present problem of optimizing water al-
location, the objective function expressing the total
hi(Ui, Vi, di) 2 0 (9 annual cost is defined by
0 s 6, 5 1 (6) N-l N-l

(i= 1, . . . . N,j=l,..., N)
x
C=h i=M+I Cui +x
i=M+l
C,j+ C,. (17)

where ui and vi are the input and output state vectors of The first and second terms on the r.h.s. are the annual
the ith subsystem, respectively. di is a design vector of return on investment and the operating cost for the
the ith subsystem and Sij is a split ratio from the jth wastewater treating system, respectively. The constant A
subsystem to the ith subsystem. Equation (2) expresses involved in the first term denotes a factor associated with
the behavior of the ith subsystem. Equations (3) and (4) the rate of return on investment. The last term in Eq. (17)
are material balance relationships at mixing and splitting is the cost for fresh water.
points, respectively. Inequality constraints (5) and (6) are
imposed to define an admissible region. The problem SYSTEM MODEL
defined by Eqs. (l)-(6) is reformulated by decomposing The system models expressing the behavior of the
into the planning and design problems along the line water-using and wastewater-treating subsystems are
described in the preceding section. described by the following Eqs. (18) and (19). The water
Planning problem: to optimize the allocation of water balance for the ith subsystem is represented by
for given design variables of subsystems.

Minimize N Qi = ,g &jQj (i = 2,. . . , N) (18)


{Sijldi*} z fi(% vit di*, &j) (7)
where Qi is the flow rate of effluent stream from the ith
subject to
subsystem. The amount of water used in each refining
Vi = gi(di*, Ui) process is determined beforehand, as described in the
(8)
preceding section. The flow rates of efauent stream from
Ui=$ &j’Vj
the subsystem, Qi’s (i = 2,. . . , M), are given constants in
Eq. (18), where A4 denotes the number of water-using
subsystems including the imaginary input subsystem. The
$ sij = l mass balance relationship of the kth pollutant in the
effluent stream from the ith pollutant in the effluent stream
hi(Xi,yi, di*) 2 0 (11) from the ith subsystem is given by
0 I &j I 1 (12)
Qi’~~=(l-r;*),~($.Qj.~;*)tPi (19)
(i=l,..., N, j=l,..., N)
(i = 2,. . . , N, k = 1,. . . , K)
where di*‘S are given as a solution of the design problem
on the lower level. where Pi” and ri“ respectively, are the rate of generation
Design problem: to optimize process conditions for the and the removal ratio of the kth pollutant at the ith
ith subproblem for a given water allocation. subsystem. In the planning problem, the values for these
parameters are assumed to be constant. In addition, since
pollutants are generated in water-using subsystems and
(13)
are removed in wastewater-treating subsystems, the fol-
lowing conditions are derived.
subject to
Pi“=O, Pik=O(i=M+l,..., N,k=l,..., K) (20)
vi = gi(di, ni) (14)
r,“=O, r/‘=O(i=l,..., M,-k=l,..., K). (21)
Ui=f &j*‘Vj (19
,=I
Moreover, the feed stream to each subsystem has to
hi(xi, yipdi) 2 0 (16) satisfy the quality acceptable for the purposes such as
petroleum refining and wastewater treatment. In parti-
for i = 1, . . . , N, where &*‘S are given as a solution of cular, the feed water to the imaginary output subsystem
the planning problem on the upper level. should satisfy the current national or local discharge
The present paper is confined to the planning problem regulations. In this connection, the following constraint
of optimizing water allocation under the assumption that associated with the kth pollutant is imposed on each
the process conditions of subsystems are given before- influent stream to the ith subsystem:
hand. The problem considered here is not so large in
scale as to require the linearization for the use of linear
programming method such as in the case of overall
refinery planning. However, the integration of many (22)
alternative systems makes the problem too complex to
(i = 2,. . . , N, k = 1,. . . , K)
carry out nonlinear simulation. Therefore, a nonlinear
objective function and linearized system models are where zik is the limitation for the concentration of the
employed here. kth pollutant in the stream influent to the ith subsystem.
254 N. TAKAMA et al.

SOLUTION h5’ROD sequence of modified problems without the inequality


In order to solve the optimization problem, the simu- constraints by introducing a penalty and by reducing a
lation of the total system should be carried out by using system structure successively.
the system models described above. In the case of The penalty is defined by
lineal&d models, computation can be performed with
relative ease. By using matrix notation, Eqs. (20) and (21) (25)
are represented below.
where pik is zero if the inequality constraint defined by
Eq. (22) is satisfied, and unity otherwise. By adding the
penalty, the objective function becomes

F=Cta.p (26)
where C is the total annual cost defined by Eq. (19). a is
a weighting parameter successively increased from
iteration to iteration. It can be shown theoretically that,
as a +m, an unconstrained optimum of F will cor-

. respond to a constrained optimum of C[9].


For a given value of the parameter, a, the minimiza-
tion of the objective function is carried out by using the

r pIki Complex method. Since a sequence of non-feasible


points is generated in the course of optimization, an
initial search for a feasible point can be omitted. When a
(24 convergence is obtained by using the Complex method,
the reduction of a system structure is made on the basis
of the result. That is, streams with negligibly small flow
rates are automatically eliminated. Then, the value of
When the value of 8, is given, Eq. (23) can be easily parameter, a, is increased and the minimization of a
solved to obtain the values of dependent variables, Qi, by modified objective function is carried out for a reduced
using the Gauss elimination method. In Eq. (24), the system structure. By repeating these procedures, a
value of .r: is obtained for given values of &j, Qivrf and feasible solution is obtained. The computational pro-
Pi”. Those values have to satisfy the inequality con- cedure of the method is shown in Fig. 3. Since the
straints defined by Eq. (22). However, the inequality detailed description on the Complex method has already
constraints are stringent in the problem considered here. been made elsewhere[lO] by one of the authors, the
Therefore, it is practical to transform the problem into a further description is not made here.

Modify the pafo-

F
meter dll*l)

Reduwthe@em
structure
-.-.Oplimizatian by .-.- YeI
-.-.-
Complex method
piI

Fig. 3. Computational procedure for optimization.


Optimalwater allocationin a petroleum refinery 25s

lLLUSl’RATtVEEXAMPLE foul water, stripping unit, an oil separating unit, and a


The planning problem of optimizing water allocation in coagulating, sedimentating, and filtrating unit.
a petroleum refinery is considered here as a simple but Numerical computations are carried out by using the
practical example. The system consists of three water- result of the aforesaid study on water use in a petroleum
using and three wastewater-treating subsystems. The refinery. The total amount of water required by the
water-using subsystems include nearly all processes in water-using subsystems involves 85% of the total
the refinery. One subsystem consists of various proces- refinery water excluding cooling water and heating
ses in which steam is used. This subsystem requires steam. Since cooling water and heating steam are com-
strict quality of supplied water and discharges oily was- monly reused with or without a simple treatment, they
tewater with a slight contamination. Another subsystem are not considered here. In the application of the
consists of hydrodesulfurizing processes in which in- method, it is practical to rule out meaningless streams
jection water is used. This subsystem requires mild qual- such as a fresh water stream directed to a wastewater-
ity of supplied water and discharges oily wastewater with treating subsystem and a recycling stream around a
considerable hydrogen sulfide. The third subsystem is a single subsystem.
desalter included in a crude distillation unit. This sub- For the specified conditions presented in Table 1,
system requires mild quality of supplied water and dis- optimal values of split ratios are determined so as to
charges oily wastewater with a slight contamination. minimize the annual cost of the system. However, some
Three major pollutants such as hydrogen sulfide, oil, and of the ratios are dependent variables because the water
suspended solid are dealt with in the wastewater-treating requirement of each water-using subsystem is given
subsystem which consists of the following processes: a beforehand. The optimal result is shown in Table 2,

Table l(a). Specifiedconditionsfor optimization


-
SUBSYSTEM DATA (1)

water flow Pollutant generation Pollutant limitation


rate rate [ton/hrl [Ppnl
[ton/hr] (H2S) (Oil) (SS) (H2S) (Oil) (SS)

@Fresh water source l

0 Steam strippers 45.8 0.0179 0.0005 0.0012 0 0 0

@HDS high pressure 32.7 0.536 0.0033 0.0005 500 20 50


sections

@ Deealter 56.5 0.0013 0.0057 0.0020 20 120 50

Note 1) * : Not specified


2) Fresh water contains no pollutant.

Table l(b). Specifiedconditionsfor optimization


SUBSYSTEM DATA (2)
water flow rate Removal ratio Pollutant limitation
Subsystem name Lton/hrl [%I LPpnl
(H2S) (Oil) 6s) (H2S) (Oil) (SS)

@ Foul water stripper l 99.9 0 0

@ oil separator * 0 95 20

@ Coagulating, sedi- t 90 90 97
mentatin, .5
Filtrating units

@ Final basin of l _ _ 2 2 5
wastewater

Note 1) l : Not specified


2) Fresh water contains no pollutant
256 N. TAKAMAet al.

Table l(c). Specified conditions for optimization

ECONOMIC DATA

Annual rate Of return [al 13.15

operating hour [hr/yearl 8,000

Subsystem cost InvestnlentCOst IS1 operatinb cat IS/hrl

0 16,800 x Q".7 1.0 XQ

0 4,800 x Qoe7 0

0 12,600 x Q"'7 0.0067 x Q

Utility cost

IS/ton1 0.30

Table 2. Comparison of optimal and conventional cases

optima1 case Conventional case

Objective function [103S/year1 707 1,550

Investment COst l103$1 676 1,060

Operating cost [103S/year1 371 1,087

Fresh water cost I103S/yearl 246 323

Pollutant in supplied [ppm] (H2S) (Oil) GS) (H2S) (Oil) ISS)


water to subsystem

0 0 0 0 0 0 0

0 390.0 10.0 25.0 0 0 0

0 0.0 0.0 0.0 0 0 0

v 2.0 1.9 0.9 0.4 0.4 0.6

@ (TON/HRI

P-l 32.7

Fig. 4. Optimal flow scheme.

together with the result of the conventional case. The l The fresh water requirement is decreased by about
flow schemes for the optimal and the conventional cases 24% that for the conventional case.
are shown in Figs. 4 and 5, respectively. The computing 0 Water reuse is practiced without any treatment so as
time to obtain an optimal solution is about 4Oseconds by to reduce the amount of fresh water and wastewater.
the IBM S370/158-U35 machine. 0 Water is treated by less expensive treatments rather
The application of the method has given great im- than expensive treatment so as to reduce the cost for
provements as shown below: wastewater treatment.
The result of example shows that it is possible to
l The annual cost is decreased by about 55% of that minimize the sum of costs for fresh water and for
for the conventional case. wastewater treatment by using the method.
Optimal water allocation in a petroleum refinery 251

Fig. 5. Conventional flow scheme.

Table 3. Process of convergence

cost Weight for NO. of


Iteration
IS/year1 Penalty streams

549 x 103 a 29

657 3.26 5 *(I 24

740 0.308 52 .a 23

822 0.224 53 *a 20

797 0.0442 54 *cx 17

906 0.0118 55 *LY 16

886 0.00003 56 .a 14

854 0.0247 57 *a 14

710 0.0 58 .a 12

707 0.0 59 *a 10

707 0.0 51O.a 10

707 0.0 511.a 10

Note 1) Penalty is defined by Equation(25).

2) (I = 1.0 x 1016

DISCUS!SlON results obtained for the previous values of a. Initial


Since the constraints associated with the problem points for the next iteration are generated arbitrarily on
considered here are quite stringent, the problem is trans- the basis of the reduced structure. The search for an
formed into a sequence of unconstrained problems by optimal point is carried out by using the Complex
introducing a penalty function. As shown in Table 3, the method. The Complex method is, in general, inefficient
method generates a non-feasible sequence of search because the simplex formed with search points is con-
points in the course of optimization. Due to the non- tracted for the search on a nearly flat surface. However,
feasible approach, difficulties associated with an initial the successive reduction of system structure prevents
search for feasible points are avoided. The penalty the search by using the Complex method from premature
prevents the search points from getting too far from the termination.
feasible region. Furthermore, the method yields a feasi- Complex and huge-dimensioned problems such as the
ble solution by increasing successively the weighting problem described here have generally a multi-modal
parameter of the penalty denoted by a in Eq. (26). While characteristic. For such cases, direct search methods
the value of parameter, a, is relatively small, the solution including the Complex method cannot always obtain a
is sought in favor of minimizing the cost rather than globally optimal solution. The global optimal point can be
minimizing the penalty. On the contrary while a is sought by repeating a number of computations starting
relatively large, the solution is sought so as to minimize from different initial points.
the penalty and finally to obtain a feasible solution.
Moreover, the successive use of the penalty is com- Acknowledgement-The authors would like to acknowledge
bined with the successive reduction of a system struc- Chiyoda Chemical Engineering and Construction Co., Ltd. for the
ture. This reduction is accomplished by the automatic support of the present study and the permission to publish this
elimination of negligible streams so as to reflects the paper.
258 N. et al.
TAKAMA

NOMENCLATURE REFERENCES
c annual cost I. B. A. Carnes, D. L. Ford & S. 0. Brady, Treatment of
Coi investment cost for the ith subsystem refinery wastewaters for reuse. Nut. Con!. Complete Water
C,, annual operating cost for the ith subsystem Reuse, Washington DC. (1973).
C,. fresh water cost 2. V. Skylov & R. A. Stenzel, Reuse of wastewaters-pos-
di design vector in the ith subsystem sibihties and problems. Proceedings of the Workshop
F modified objective function including the penalty (AZChE), Vol. 7 (1974).
i objective function for the ith subsystem 3. R. W. Hospondarec & S. J. Thomson, Oil steam system for
number of pollutants wastewater reuse. Proceedings of the Workshop (AZChE), Vol.
M number of water-using subsystems including the imaginary 7 (1974).
input subsystem 4. M. Sane, U. S. Atkins & Partners, Industrial water
N number of subsystems in the total system management. ht. Chern. Engng Symp. Series No. 52 (1977).
P penalty associated with constraints defining the acceptable 5. D. Anderson, Practical aspect of industrial water reuse. Znt.
quality of influent streams Chem. Engng Symp. Series No. 52 (1977).
Pi” generation rate of the kth pollutant in the ith subsystem 6. P. N. Mishra, L. T. Fan & L. E. Erickson, Application of
Qi flow rate of effluent stream from the ith subsystem mathematical optimization techniques in computer aided
r,’ removal ratio of the kth pollutant in the ith subsystem design of wastewater treatment systems. Water-1974 (II),
u, input state vector in the ith subsystem AZChE Symp. Series 71, 145(1975).
vi output state vector in the ith subsystem 7. T. Umeda, A. Hirai & A. Ichikawa, Synthesis of optimal
x; concentration of the kth pollutant in the effluent stream processing system by an integrated approach. Chem. Engng
from the ith subsystem Sci. 27, (1972).
Zik concentration limitation for the kth pollutant in the influent 8. T. Kuriyama, N. Takama, K. Shiroko & T. Umeda, Optimiz-
stream to the ith subsystem ing water reuse in a petroleum refinery. PAChEC, Denver
(1977).
Greek symbols 9. D. Davies & W. H. Swarm, Review of constrained optimiza-
a weighting parameter associated with the penalty tion. Optimization (Ed. R. Fletcher), Academic Press, Lon-
Sir split ratio of the stream from the jth subsystem to the ith don (1978).
subsystem IO. T. Umeda, Optimal design of an absorber-stripper system.
A factor associated with an annual fixed cost Ind., Engng Chem. Proc. Des. L&w. 8 (1%9).

You might also like