Chinese Journal of Chemical Engineering, 16( I) 67-70 (2008)

An Optimization Model for the Production Planning of Overall

Refinery*

GAO Zhen (&&)I-**, TANG Lixin (@?$%)*,JIN Hui (&%)I3* and XU Nannan (q$&#)'
The Logistics Institute, Northeastern University, Shenyang 1 10004, China
* Liaoning Institute of Technology, Jinzhou 121001, China
Abstract This article addresses a production planning optimization problem of overall refinery. The authors for-
mulated the optimization problem as mixed integer linear programming. The model considers the main factors for
optimizing the production plan of overall refinery related to the use of run-modes of processing units. The aim of
this planning is to decide which run-mode to use in each processing unit in each period of a given horizon, to satisfy
the demand, such as the total cost of production and inventory is minimized. The resulting model can be regarded as
a generalized lot-sizing problem where a run-mode can produce and consume more than one product. The resulting
optimization problem is large-sized and NP-hard. The authors have proposed a column generation-based algorithm
called branch-and-price (BP) for solving the interested optimization problem. The model and implementation of the
algorithm are described in detail in this article. The computational results verify the effectiveness of the proposed
model and the solution method.
Keywords refinery planning, lot-sizing, optimization, column generation

1 INTRODUCTION terms of time periods. Recent studies for optimization

The main aim of production planning is to decide
what to produce, how much to produce and when to
produce for a given plan horizon in a company. The
production plan includes yearly plan, seasonal plan
and monthly plan in terms of time horizon. The objec-
tive of production planning in a refinery is to generate
as many valuable products as possible, such as gaso-
line, jet fuel, diesel, and so on, and at the same time
satisfying market demand and other constraints.
Oil refining is one of the most complex chemical
industries, which involves many different and com-
plicated processes with various possible connections.
It is well known that the oil refining is a typical con-
tinuous production process. Maybe a typical refinery
includes tens of units. Therefore the optimization of
the production planning of the overall refinery is con-
sidered as one of the most difficult and challenging
tasks, which is also often formidable, even impossible
[l]. Nevertheless, the production plan optimization is
an important profit growth point thus it also becomes
a burning hot topic in both industry and academia.
Various optimization models have been developed for
individual units with specific technological character-
istics. However, the optimization of the production
units does not achieve the global economic optimiza-
tion of the plant. Usually the objectives of the indi-
vidual units are conflicting and many times infeasible
thus many production paths are restricted or disabled.
The production planning optimization for refinery-wide
has been addressed by using linear programming in
the past decades. Although the linear programming
models are not good enough to consider the discrete
features of the planning problem, such as the dynamic
feature of demand, uneven features of the supplement
of crude oil and the production of processing units in
of production planning have been toward the devel-
opment of nonlinear programming and mixed integer
linear programming models 12, 31. The main study of
this article is to propose a production planning model
and algorithm for refinery-wide optimization.
2 PROBLEM STATEMENT
The problem studied by the authors is at Jinxi
Refinery of Liaoning Province, China. Jinxi Refinery
is a large-scale fuel oil-lubricant oil type oil refining
plant. It has over 50 production units including two
distillation units (AVU), two catalytic cracking units
(FCC), one continuous catalytic reforming unit (CCR),
one delayed coking unit, two hydrotreating and hy-
drofining units and some auxiliary units. The process-
ing ability of crude oil is over six million tons per year.
A simplified technology flow chart of the refinery is
as follows:
A run-mode indicated in Fig. 1 may be explained
as follows: the crude oil is processed by the atmos-
pheric and vacuum distillation unit (AVU) producing
gas, liquid petroleum gas (LPG), light naphtha, vac-
uum gas oil (VGO), and vacuum residue (VR). The
naphtha is processed by CCR producing reforming
gasoline. The atmospheric and vacuum distillation
diesel and FCC diesel are fed to the Hydrotreating unit
producing refining diesel. The atmospheric and vac-
uum distillation VGO, Delayed coking VGO and VR
are fed to the FCC producing gasoline. VR is fed to
the Delayed coking unit producing gasoline, diesel
and VGO. These are the most typical run-mode for the
main units. If different run-modes are used for these
processing units, changes will take place in the yield
levels and structures of the products in these units.

Received 2007-05- 10, accepted 2007- 10-19.

* Supported by the National Natural Science Foundation for Distinguished Young Scholars of China (No.70425003), the Na-
tional High Technology Research and Development Program of China (No.2006AA04Z174), the Natural Science Foundation
of Liaoning Province (No.20061006) and the Enterprise Post-Doctorial Foundation of Liaoning Province.
** To whom correspondence should be addressed. E-mail: gzneu@ 126.com
68 Chin. J. Chem. Eng., Vol. 16, No. 1, February 2008
- naphtha
jet fuel
A
diesel
crudeoil v -
t
U
VGO
VR
The production planning problem at Jinxi Refin-
ery comprises of the optimal yearly production plan
consisting of yield level of each product in each
month, given customer demand, capacities of units,
and initial inventories of crude oil, components and
products. The optimization of production planning is
equivalent to selecting the best operating routes as
well as run-modes for all the processing units. A
run-mode of operation for a processing unit is speci-
fied by the combination of products consumed and
produced in the process, and by the yield levels of the
products.
At Jinxi Refinery, the main factors considered for
the planning problem are fulfillment of the demand,
prices of products, inventories of products, and crude
oil processing capacity. Currently, the planning is car-
ried out manually with the help of spreadsheet models
and is time consuming. No optimization tools are used
for the planning.
For optimizing the planning, a reference was
made to the model in the literature [l], attempts have
been made to update it for production plan problems.
The length of the time period has been extended to
one month to suit the production plan horizon and to
add a capacity constraint to the problem. The authors
have also redefined the cost for the use of a run-mode
of a processing unit to make it include costs of utilities
such as water, electricity, steam and wind power. The
model for the problem is a mixed integer linear pro-
gramming (MILP) which is a generalized lot-sizing
optimization problem where according to a run-mode,
more than one product can be produced and consumed.
The problem formulation is as follows:
s.t.
c Yymt =1
reforming
*
t
-
gasoline e
- * 4 various
A diesel $ products
--+
H, treating 5
00
t
.*
-0
* B
* B
The objective function ( 1 ) expresses minimizing
the total cost consisting of inventory and production
costs. Constraints (2) express production balance and
constraints (3) express consumption balance. Con-
straints (4)express supplement-demand balance. Con-
straints ( 5 ) explain for each processing unit in each
period one and only one run-mode can be taken. Con-
straints (6) help in capacity restriction. Constraints (7)
and (8) explain that backorder is not to be allowed.
The model can be regarded as a generalized lot-sizing
problem in which a run-mode leads to many products
being produced. The lot-sizing problems have been
extensively studied in literature and have been proved
to be NP-hard [4-61. The authors propose a
branch-and-price algorithm for solving the generalized
lot-sizing problem. Branch-and-price algorithms are
the generalization of branch-and-bound algorithms
that solve LP relaxations using column generation.
Branching occurs when no column enters the basis
and the LP solution does not satisfy the integrality
conditions. Branch-and-price algorithms have been
used for solving many difficult integer and combina-
torial programming problems [7- 101.
3 ALGORITHM
The algorithm proposed in this article mainly in-
cludes three parts. The first part is the original prob-
lems set-partitioning reformulation, and the second
and third parts are a column generation procedure and
a branch and bound procedure respectively, which
help to solve the resulting set partitioning problem.
3.1 Set-partitioning reformulation
Given a set of production schedules for each unit
q, the single-level multi-item capacitated lot-sizing
problem can be formulated as the following set parti-
tioning problem:
 Chin. J. Chem. Eng., Vol. 16, No. 1, February 2008 69

(9) If the reduced costs & 2 0 , for all q E Q , hold,

then the main problem (MP) is solved optimally, oth-
erwise the MP is not solved, and the column with the
Subject to least /?GO is to be added to the restricted main prob-
Co@=l' vq lem (RMP) to reoptimize.

3.3 Branching rule

r ! 1
Oq E {0,11, Vq,j
where 0, is a binary variable, equal to one when
schedule j is selected for unit q, and zero otherwise;
6, is the cost associated with schedulej for unit q and
pqlr is the capacity requirement in period t for that
As the LP solution of the root node is not neces-
(12) sarily an integer, a branch-and-bound procedure has
been used to get integer solutions. Note that the opti-
mal solution B to LP relaxation is infeasible, if and
only if C p&" = a is fractional for some q and t.
j

same schedule. Let xPlt, zPlr,yyml,,I,,,, ( t = 1, .-, de- Thus the authors propose a branching rule as follows.
note the production quantity, consumption quantity, On the left branch let pd, BQ SLaJ; on the right
run-mode variable and inventory associated with j
schedule j for unit q in period t, respectively. Then branch let: C p d p d 3 ral .
bql = C( C c:mlyymlt+ c;rlpl, and P q f =
tsT m M 1 9 bP,f. for j
The branching rule can be enforced using the
t = 1, --,T. For the proof of the equivalent of the two following procedures:
problems please refer to Ref. [ I l l . Note that the re- (1) Deleting old column procedure
sulting set partitioning problem has many more vari- 0 In the left branch, delete columns for unit q
ables than the original problem; therefore a column production level xPjl is greater than La].
generation-based procedure is proposed for the problem. 0 In the right branch, delete columns for unit q
production level xpjris lesser than ral .
(2) Generating new column procedure
3.2 Pricing problem solution 0 In the left branch, add the upper bound con-
straint: xpjl< La1 .
For problem SP, the reduced costs are pd = bd - 0 In the right branch, add the lower bound con-
T
straint: x P j , 3ral.
C q p , , -aq , for all q, j . Where cf, q E Q and 4,
,=I 4 COMPUTATIONAL RESULTS
LE T, are dual variables associated with the constraints
(10) and ( l l ) , respectively. To check the optimality, The authors have performed above branch-and-price
the pricing problem which is a single-item uncapacitated algorithm in a Pentium IV personal computer using
lot-sizing problem (ULS) is constructed and solved. VC++ 6.0 programming language. For verifying the
(ULS) minP (13) effectiveness of the proposed algorithm, they have
Subject to computed the results for several groups of practice
data coming from the Jinxi Refinery and compared
It-, + X, - I , = d, , Vt (14) them with the practice results from the refinery. The
results are shown in Table 1. The meaning of the
xt < M y , , vt (15) column Problem I D is P/units/products/periods/l,. For
identifying the solution quality of the algorithm, the
I , , IT = 0 (16) authors have compared the results computed by the
x , , I , 3 0 , vt (17) algorithm with the results of several heuristics for

Table 1 Computational result for practice problems

Problem ID LP value"'X lK6/RMB Yuan IPvalueax10 %MB Yuan Columns generated Nodes Dual gap/% CPU time/s
P/8/30/ 12/50 1102.28 1104.00 2475 34 0.18 40
P/8/40/12/50 1314.19 1318.00 3863 38 0.29 45
P/8/50/12/50 1323.52 1326.00 4203 13 0.23 29
P/8/30/12/60 1330.01 1334.00 2892 17 0.23 20
P/8/40/12/60 1400.82 1408.00 5645 49 0.6 114
P/8/50/12/60 I A 3 7 45 1428.00 1409 78 0.57 441
0 Optimal linear programming objective function value.
0 Optimal integer programming objective function value.
70 Chin. J. Chem. Eng., Vol. 16, No. 1, February 2008

Table 2 Comparing the solution quality of BP with other heuristics

Problema LP value@ IP value@ BP Lambrect Venderveken Dixon Silver Lagrangian Relaxation
TVWl 7996.67 8430.00 8670.00 8970.00 8710.00 8710.00
mw2 7722.21 7910.00 7980.00 8800.00 7930.00 7920.00
TVW3 7534.17 7610.00 7610.00 7970.00 8970.00 7660.00
TVW4 7446.17 7520.00 7520.00 8000.00 8000.00 7520.00
0 The problems TVW1-TVW4 are taken directly from Ref. [61 for the 8x8 (itemsxperiods) problem with different capacities.
0 Optimal linear programming objective function value.
@ Optimal integer programming objective function value.

benchmark problems from literature. The results are set of perid$

shown in Table 2.
5 CONCLUSIONS
The production planning optimization for refinery-
wide is considered as one of the most difficult and
challenging tasks. This article has formulated a gener-
alized lot-sizing problem model for the problem for a
typical refinery and proposed a branch-and-price algo-
rithm for solving the optimization problem. The com-
putational results show that the proposed model and
algorithm are effective. It is promising for improving
the production plan of the refinery. The future research
directions are expected in continuous time modeling
for production plan optimization in refineries.
NOMENCLATURE
production level of productp in processing unit 4 using run-mode m
the cost associated with schedulej for unit q
consumption level of product p in processing unit q using
run-mode m
unit inventory cost of product p
unit price of product p
production cost of processing unit q under run-mode m in period t
demand of product p in the period t
inventory level of product p at the end of period t
unit capacity consumption of product p
set of run-modes
index, run-modes
set of products
index, products
the capacity requirement associated with schedule j in period f for
unit q
set of units
index, units
index, periods
yield level of product p in period t
equal to 1 if run-mode m is used for unit q in period t, otherwise
equal to 0
consumption level of product p in period t
dual variable to index q
binary variable for selection of run-modes of unit q
dual variable to index f
REFERENCES
1 Lundgren, M.G., Lundgren, J.T., Persson, J.A., An optimization
model for refinery production scheduling, Int. J . Prod. Econ., 78,
255-270 (2002).
2 Pinto, J.M., Joly, M., Moro, L.F.L., Planning and scheduling mod-
els for refinery operations, Comput. Chem. Ens., 24, 2259-2276
(2000).
3 Zhang, N.,Zhu, X.X.,A novel modelling and decomposition strat-
egy for overall refinery optimization, Comput. Chem. Eng., 24,
1543-1548 (2000).
4 Larnbrecht, M.R., Vanderveken, H., Heuristics procedures for the
single operation, multi-item loading problem, AllE Trans., 11,
319-326 (1979).
5 Dixon, P.S., Silver, E.A.. A heuristics solution procedure for
multi-item, single-level, limited capacity, lot-sizing problem, J.
Oper. Manag., 2, 23-39 (1981).
6 Thizy, J.M., van Wassenhove, L.N., Lagrangian relaxation for the
multi-item capacitated lot-sizing problem, a heuristics approach,
AIlE Trans., 17, 308-313 (1985).
7 Savelsbergh, M.W.P., A branch-and-price algorithm for the gener-
alized assignment problem, Oper. Res., 45,831-841 (1997).
8 Desrochers, M., Desrosiers, J., Soloman. M., A new optimization
algorithm for the vehicle routing problem with time windows, Oper.
Res., 40,342-354 (1992).
9 Vance, P.H., Branch-and-price algorithms for the one-dimensional
cutting stock problem, Comput. Opt. Appl., 9,211-228 (1998).
10 Cattrysse, D., Maes, J., van Wassenhove, L.N., Set partitioning and
column generation heuristic for capdcitated dynamic lot-sizing, Eur.
J . Oper. Res., 46.38-47 (1990).
I1 Nemhauser, G.L., Wolsey, L.A., Integer and Combinatorial Optimi-
zation, Wiley, Chichester, UK (1988).