Refinery Planning and Scheduling: An Overview

Jens Bengtsson and Sigrid-Lise Nonås

Abstract In this chapter, we give an overview of recent literature on the planning

and scheduling of refinery activities. Planning of refinery activities ranges from
determining which crude oil types to acquire to which products that should be
produced and sold in the market. The scheduling ranges from scheduling of crude
oil unloading and blending to blending of components to finished products. This
overview treats three different categories of activities: planning and scheduling of
crude oil unloading and blending, production planning and process scheduling, and
product blending and recipe optimization. The focus will be on identifications of
problems, the models outlined for the specified problems, and the computational
difficulties introduced by the models. A final section discusses an agenda for future
research.

1 Introduction

Building a modern refinery is a huge investment that puts its owner in a position
where high fixed cost must be covered for a lengthy future. Because of this in-
vestment and fixed cost, efficient use of refinery resources is important both for
short-term and long-term profitability. In addition, refineries incorporate complex
equipment and produce complicated chemical reactions, posing difficult challenges
for determining the best use of the refinery capacity.

S.-L. Nonås ()

Department of Finance and Management Science, Norwegian School of Economics
and Business Administration (NHH), Helleveien 30, 5045 Bergen, Norway
e-mail: [email protected]
J. Bengtsson
Department of Accounting, Auditing and Law, Norwegian School of Economics
and Business Administration (NHH), Helleveien 30, 5045 Bergen, Norway
and
Department of Finance and Management Science, Norwegian School of Economics
and Business Administration (NHH), Helleveien 30, 5045 Bergen, Norway
e-mail: [email protected]

E. Bjørndal et al. (eds.), Energy, Natural Resources and Environmental Economics, 115
Energy Systems, DOI 10.1007/978-3-642-12067-1 8,
c Springer-Verlag Berlin Heidelberg 2010
116 J. Bengtsson and S.-L. Nonås

Apart from the complexity of the refinement and the challenges of determining
efficient processes within the refinery itself, other factors play key roles in the search
for profits within the industry. The markets for crude oil and petroleum products
have developed over the decades and have shown to be sensitive both to political
issues and disruptions and variations in demand and supply. The market for oil re-
lated products is well-developed and this also includes the freight market. The latter
is important since crude oils and products are dependent on transport to their in-
termediate and final destinations. Accordingly, profitability will also be affected by
transportation costs.
The supply chain of an integrated oil company stretches from the production
and purchasing of crude oil to customers buying petrochemical products or fuel for
heating or transport. There are many decisions that must be made along the supply
chain such as what crude oil mix to buy and sell, what components and products
should be produced, and whether they should be kept for internal use, stored, or
sold to external players.
To address the challenges in the oil and gas supply chain quantitative models and
mathematical programming techniques have been developed for several decades and
their use has significantly increased the ability to plan and control refinery activities
and to increase profits.
From a refinery management perspective there is a difference between planning
and scheduling. At the planning stage, the time horizon is typically several weeks or
months, and the decisions typically concern purchase of crude oils and the produc-
tion and sales of products. Since the markets associated with refinery operations are
volatile, the use of correct and updated information is important because this will
strongly affect the capability to identify market opportunities. The identification of
market opportunities is crucial for increased profitability. The capability of identi-
fying market opportunities will be dependent on a company’s ability (given current
condition: prices, production decisions, available crude, etc.) to determine its deci-
sion to buy, refine, and sell its products. To make such decisions, the company must
consider already booked and planned production, together with future prices.
Due to the complexity involved in different refinery operations throughout the
supply chain, the refinery scheduling problem is often separated into three different
sub-problems, see Fig. 1 below. The first sub-problem involves the crude oil unload-
ing from vessels or pipelines, its transfer to storage tanks and the charging schedule
for each crude oil mixture to the distillation units. The second sub-problem consists
of production unit scheduling, which includes both fractionation and reaction pro-
cesses. The third sub-problem is related to the blending, storage, and lifting of final
products.
Historically, refiners have built organizations based on the processes associated
with planning and scheduling. To drive operational efficiency, major refining com-
panies are now putting increased focus on managing supply chain activities as an
integrated process, closely connecting refinery planning and scheduling to improve
communication and total plant operation.
In this chapter, we will give an overview of the latest literature on refinery plan-
ning and scheduling and also make some suggestions for future research. A previous
Refinery Planning and Scheduling: An Overview 117

Crude oil intake Crude oil Component Blend Finished Lifting/shipping

pipepline/port storage tanks Charging tanks Refining processes inventories
headers product tanks points

Crude oil unloading Production unit Product blending and

and blending planning/scheduling recipe optimization

Fig. 1 Three sub-systems of an oil refinery

survey of literature on production planning and scheduling models for refinery op-
erations can be found in Pinto et al. (2000). The focus of the chapter will be on
the identifications of problems, the type of models outlined to solve the identified
problems, and the computational difficulty introduced by the models (it will not dis-
cuss how the computational difficulties are met). The overview is organized in three
different parts, which correspond to the sub-problems mentioned earlier. Literature
that focuses on modeling the whole refinery supply chain is also discussed. This can
be found under the production planning section.

2 Crude Oil Selection and Crude Oil Scheduling

The planning and scheduling of crude oil operations in a refinery is a critical task that
can save the refinery millions of dollars per year (Kelly and Mann 2003a, b). Crude
oils vary significantly in compositions, product yields, properties, and prices, and
their acquisition accounts for a large portion of the refineries’ cost. A key issue for
a refinery is, therefore, to identify and process optimal crude blends that maximize
profit margins (Li et al. 2007).
Typically, an oil refinery receives its crude oil through a pipeline that is linked
to a docking station, where oil tankers unload. The unloaded crude oil is stored at
the refinery in crude oil tanks. The crude oils are stored in these tanks, at least for
a minimum amount of time (to allow the separation of the brine), before the mix of
crude oils is transferred to charging tanks or directly blended and processed in the
distillation tower.
In general, two types of ships supply crude to a refinery: very large crude carri-
ers (VLCCs) that may carry multiple parcels of different crudes and small vessels
118 J. Bengtsson and S.-L. Nonås

carrying single crude. Due to its size the VLCCs often dock at a single buoy mooring
(SBM) station offshore whereas the small vessels berth at the jetties. The pipeline
connecting the SBM station with the crude tanks normally has a substantial holdup
(Li et al. 2007), while the holdup in the jetty pipeline is not that critical.
The shipping schedule for the crude oil tankers is determined by the procure-
ment and logistic department in conjunction with the crude oil purchase. Due to
lengthy waterborne transit times, this schedule is done a long time before the crude
oil tankers arrive at the refinery. In the case that the crude oil carrier transport differ-
ent crude oil parcels, the unloading sequence is also predetermined due to logistic
considerations. Before a VLCC can unload, it must first eject the crude that resides
in the SBM pipeline. This crude can, as suggested in Reddy et al. (2004b), be mod-
eled as an extra crude oil parcel from the carrier. A general assumption is that the
holdup of the jetty pipeline is negligible. In the literature, it is typically assumed that
the holdup of the SBM pipeline is negligible.
The crude oil scheduling is based on the current information on arrival of crude
oil vessels or carriers, crude composition in different tanks, and the optimal crude
feed to the crude distillation units (CDUs) from the production plan.

2.1 Selection of Crude Oils

The main goal of the crude selection is to find a feasible crude blend that maximizes
the profit of the planned production in the time horizon considered, taking into ac-
count the current storage of crude oil at the refinery and the crude tankers scheduled
to arrive at the refinery. The “wrong” crude mix can cost a refiner both in excess
feedstock expense and lost product revenue. To find the right crude mix the sched-
uler has to take into account both processing and economic considerations. After
the selection of crude oils, the crude procurement and logistics departments have to
secure the crudes and schedule them for delivery.
The monthly planning model, updated with more advanced crude blending sim-
ulations, is often used as a decision tool in order to determine the optimal crude
oil mix. Each refinery has built up its own history of how different crudes have
performed for their refinery in the past. This information is then used in the
crude blending simulation to achieve an optimal crude slate. Both Stommel and
Snell (2007) and Ishizuka et al. (2007) present good discussions and describe gen-
eral rules for how leading companies handle the selection and the scheduling of
crude oil.
In the literature, crude oil selection has been considered as part of the production
planning problem. The literature, however, neglects one important component.
The complex nonlinear blending relation of different crude oil mix is omitted in the
planning problem due to the increase in problem complexity. In the industry, these
relations are usually solved using in-house or more commercial simulation tools.
The general review of literature related to production planning is presented in a
later section.
Refinery Planning and Scheduling: An Overview 119

2.2 Scheduling of Crude Oils

The objective of crude oil scheduling is to minimize the operational cost while
meeting the target crude oil load to the CDU. For the scheduling one assumes a
fixed arrival schedule for vessels, knowledge of the quantity and quality of crude oil
at the vessels and the crude oil mix at the crude tanks, the minimum settling time
for the brine at the tanks, minimum and maximum level of crude in the tanks, min-
imum and maximum flow rate in the pipeline, and the target feeding-rate (quality
and quantity) of the crude oil blend for the CDU. It is also common to assume that
a tank cannot receive and feed crude at the same time.
Given these facts, the objective of the crude oil scheduling is then to determine
the unloading schedule for each vessel (this includes the timing, rates, and which
tank to transfer the oil parcels to); the transfer schedule of crude oil mix between
storage and charging tanks (if both are present in the refinery); the inventory levels
and crude oil concentration for all storage and charging tanks; and finally the charg-
ing schedule for each CDU (how much should be transferred to each CDU from
each charging tank).
To model the crude oil scheduling we need a large number of binary allocation
variables to consider the discrete scheduling decisions, such as selecting a tank to
receive a crude oil mix and nonlinear constraints to calculate the crude oil compo-
sition for the storage and charging tanks. So far, the models proposed for crude oil
scheduling have not considered nonlinear crude properties. The nonlinear property
constraints are approached by linear constraints that consider key crude compo-
nent concentrations or blending indexes that are linear on a volumetric or weighted
basis. Even if we assume linear crude properties, the crude oil scheduling results
in a complex mixed integer nonlinear (MINLP) model. The nonlinear terms are bi-
linear and originate from the mass balance and crude mix composition constraints
for the storage and charging tanks and the feed from these tanks. A nonlinear term
f.x; y; z/ is said to be bi-linear if it is linear with respect to each of its variables, i.e.,
f.x; y; z/ D x  y C y  z C z x.
Several authors have discussed different models and methods to solve this crude
oil blending and scheduling problem, taking into account different degrees of refin-
ery complexity. The scheduling problem can be modeled using either a continuous
time approach or a discrete time approach. The discrete time formulation tends to
rely on an excessive number of time periods to achieve the required accuracy, while
continuous time formulation results in complex models that require several assump-
tions or specialized algorithmic solution techniques. Recent trends in scheduling
models for chemical processes have however moved toward continuous time formu-
lations to avoid the high number of integer variables found in discrete time models.
The continuous time modeling is particularly suited to crude oil scheduling since
refinery activities can range from some minutes to several hours (Joly et al. 2002).
A review that compares discrete and continuous time approaches of scheduling for
chemical processes is provided by Floudas and Lin (2004).
One of the first models presented for the crude oil scheduling is a discrete mixed-
integer linear programming (MILP) presented by Lee et al. (1996). The authors
120 J. Bengtsson and S.-L. Nonås

consider one docking station, a set of storage tanks, a set of charging tanks and a
set of CDUs The objective is to find a schedule that meets the predetermined crude
slate for the CDU, while minimizing total operating cost (unloading cost, cost for
demurrage, tank inventory cost, and changeover cost for the CDUs). In addition,
the crude mix in the charging tanks should be within predefined quality measures
with regard to key component concentration. Their linear reformulation of the bi-
linear mass balance constraints is, however, not rigorous enough to ensure that the
crude mix composition for the storage and charging tanks is the same as the com-
position of the flow from the tanks. This inconsistency is denoted as “composition
discrepancy” by Li et al. (2002) and Reddy et al. (2004a). In general, composition
discrepancy may occur when individual component flows are used in a linear refor-
mulation for the bi-linear mass balance term for storage and charging tanks where
mass accumulates. Li et al. (2002) proposed a discrete mixed integer nonlinear pro-
gramming (MINLP) model that extended the model in Lee et al. (1996) by reducing
the number of discrete decision variables (they replace two bi-index binary variables
with one tri-index binary variable) and by including new features as multiple jetties
and allowing the possibility for two tanks to feed a crude distillation unit. The so-
lution approach outlined for the MINLP model may, however fail to find a feasible
solution even if one exists (Reddy et al. 2004a; Kelly and Mann 2003a, b).
Jia and Ierapetritou (2003, 2004) outline a continuous time MILP model for the
crude oil scheduling, using only linear terms for the crude mix operations as in Lee
et al. (1996). They assume no cost for crude or tanks changes, one tank feeding one
CDU at a time (and vice versa). The objective is to minimize demurrage, unloading
and inventory cost. As for Lee et al. (1996), the MILP model may suffer from com-
position discrepancies. The crude composition in the storage and charging tank may
not match its feed to the charging tanks and to the CDU.
Moro and Pinto (2004) propose a continuous time MINLP formulation for the
crude oil scheduling. The model considers one pipeline to unload the crude oil,
settling time in the crude tanks to separate brine from the oil, and at most, two
crude oil tanks feeding the CDU. They propose to measure the quality of the CDU
stream by limiting the concentration of the critical components in the feed to the
CDU. Their objective is to maximize the CDU feed rate while minimizing the crude
tank operating costs. Moro and Pinto (2004) also proposed a MILP approach of the
MINLP model where the bilinear crude mixing term is linerarized by discretizing
the amounts or types of crude oils present in the storage tanks. The MILP approach
suffers, however, from an increasing number of binary variables as the number of
discretization interval increases. Reddy et al. (2004a) developed, in parallel to Moro
and Pinto (2004), a continuous time MINLP formulation considering multi-parcel
vessels loading at one SBM, pipelines that transfer the crude parcels from the SBM
to the crude storage, and charging tanks that again feed the CDUs. Multiple storage,
tanks can feed one CDU (and vice versa). They also discuss how SBM parcels can
be created to take into account that before a vessel can unload, the crude in the SBM
pipeline (the SBM parcel) has to be transferred to a storage tank. The objective of the
scheduling model is to maximize gross profit (profit of products-cost of crude oils)
while minimizing the operating cost (CDU changeover cost, demurrage, penalty for
running under the crude safety stock).
Refinery Planning and Scheduling: An Overview 121

In addition to the features considered in the continuous time model the discrete-
time model outlined by Reddy et al. (2004b) accounts for multiple vessels unloading
to a set of jetties and transfer of crude oil between the storage tanks. They also
present a novel approach for dealing with parcel unloading, which uses fewer binary
variables than earlier work (Li et al. 2002; Lee et al. 1996). In addition, they allow
more than one unloading allocation in any time period, thus utilizing the entire time
period to the maximum extent. The objective for their scheduling model is to max-
imize the crude margin (total value of cuts from the crude distillation units minus
the cost of purchasing, transporting, and processing the crude) minus the operating
costs (changeover, demurrage, penalty for running under the crude safety stock).
The unloading cost and inventory cost is not considered since the amount of crude
is fixed for the scheduling horizon. The demand for crude oil for each CDU has,
however, to be satisfied. Their solution approach improves the solution approach
proposed by Li et al. (2002) but may still fail to find a feasible solution even if one
exists. Li et al. (2007) improved the MINLP formulation of Reddy et al. (2004a)
in two ways. First, constraints that disallow uncontrolled changes in the CDU feed
rates are inserted. Second, linear blending indexes, weight- and volume based, are
used to better approximate the nonlinear crude oil properties.
From this literature we see that the crude scheduling problem is approached with
both MILP and MINLP models and solved either with standard MILP and MINLP
solvers or with tailor-made solution approaches. A global optimization algorithm for
the crude scheduling problem would be preferable, but considering the large sizes
of practical problems and the need for quick solutions, that will require considerable
effort and is a great challenge for future research. The main difficulty is to deal with
the large number of integer variables in the model and the complexity of nonlinear
blending and crude oil mixing operations.
Note that disruptions such as crude arrival delay could make any given schedule
infeasible and necessitate rescheduling of operations. Adhitya et al. (2007) discuss
how to reschedule in order to make only minimal changes to the scheduled opera-
tions rather than undergo a total reschedule.

3 Production Planning and Scheduling

The refinery is built up of different processing units that transform a variety of input
streams into several output streams. The flow rate and product specification, e.g.,
octane number and sulphur content, of each output stream is determined by the flow
rate and product specification of the unit feed and the operating mode of the pro-
cessing unit. Nonlinearity arises from mixing the feed and, in the yield, from the
processing. The change of operating mode of a processing unit results in a period
with reduced capacity and disturbance in the yield (quality and quantity) of the pro-
cessing. Reduced capacity and disturbance in the yield also occur for a start up after
a period with maintenance. To correctly model the disturbance caused by a change
in operating mode is difficult, usually the issue is relaxed by incurring a setup cost
for each change in mode.
122 J. Bengtsson and S.-L. Nonås

For a general refinery, the planning specifies which crude or intermediate

products to purchase and which products to produce. The planning decisions are
taken based on forecast of future demand, and usually planning takes into account a
two or three months time horizon. The production plan is used in a rolling horizon
setting to take into account updated information regarding refinery and markets.
The decisions related to scheduling of refinery activities are generally performed
on the basis of shorter time horizons, i.e., days or weeks, to determine the specific
timings of the operations.

3.1 Production Planning

To support production planning decision making, refineries generally use com-

mercial packages based on linear one-periodic models that rely on constant yields
(PIMS from Aspen Tech and RPMS from Honeywell, Process Solutions). This has
motivated researchers to outline models that give a more accurate representation of
the refinery processes or activities. In this section, literature that takes into accounts
multiple periods and different degrees of nonlinearity in the mixing and process-
ing operations are presented. Papers that consider a supply chain perspective and
uncertainty in market data are also presented.
One of the first contributions to consider nonlinearity in the production planning
is that of Moro et al. (1998). Moro proposes a framework where every unit in the
refinery is represented as an entity and the complete refinery topology is defined by
connecting the unit streams. For the processing units nonlinearity can be considered
in the blending relations and in the process equations. A general MINLP model is
discussed for a diesel production planning problem, but this is only partly outlined
in the paper. Detailed blending and processing formulations are presented only for
the hydro-treating unit. They report that the refinery plan obtained from the MINLP
model improved the performance of the case company significantly compared to the
current operating decision that was based on experience and manual calculations.
The same planning model is discussed in Pinto and Moro (2000), here with results
from a new case study.
Neiro and Pinto (2005) formulate a MINLP model that extends the planning
model discussed in Moro et al. (1998) to account for multiple time periods and
uncertainty in market data. Uncertainty is considered in the product demand, the
product price and the cost of crude oil. The uncertainty is expressed in scenarios,
and the objective function includes weighted values of each scenario based on the
probability for each scenario to occur. For each time period, the main decisions are
which crude oil to select, how to operate the processing units, and how much of the
final products to hold in inventory. They show an exponential increase in solution
time with the number of time periods as well as with the number of scenarios. In
the work listed above, only sub-systems of the gas and oil supply chain have been
considered in a reasonable level of detail. Neiro and Pinto (2004) propose a frame-
work for modeling the whole petroleum supply chain, including crude oil suppliers,
distribution centers, and several complex conversion refineries interconnected by
Refinery Planning and Scheduling: An Overview 123

intermediate and end product streams. This study outlines a large scale one periodic
MINLP planning model for the system addressing crude oil purchasing, production
units processing, inventory management, logistics, and end product sales decisions.
Neiro and Pinto (2004) consider nonlinear blending for the different processing units
and storage tanks, and nonlinear operating conditions in accordance to the yield
from the processing units. They consider a supply chain with four refineries con-
nected with pipelines and storage tanks, each with different capacity and topology.
Refinery planning and the refinery scheduling are generally performed sequen-
tially, mainly due to the complexity of the refinery sub-problems. When determining
the refinery production plan, many of the scheduling constraints are uncertain. To
obtain a feasible schedule which utilizes resources in, or close to, an optimal fashion,
it is important that the company determine a plan which makes this possible. If the
planning and scheduling is done sequentially, there is no guarantee that the pro-
duction plan can give an operable schedule. Kuo and Chang (2008) have addressed
this issue and present a MILP planning model that addresses stream allocations and
processing run modes for several scheduling intervals. By considering the whole
refinery supply chain and splitting the planning period into several sub-intervals,
Kuo and Chang, are better able to match the planning and scheduling decisions and
improve the performance of the supply chain scheduling activities.
Environmental regulations and the risks of climate change pressure refineries to
minimize their greenhouse gas emissions. Refineries also face more stringent prod-
uct specifications on oil products which typically increase their energy consumption
and CO2 emissions. Szklo and Schaeffer (2007) address this problem, also with a
specific focus on the Brazilian refining sector. Holmgren and Sternhufvud (2008)
discuss different possibilities for reduction of CO2 emissions for petroleum refiner-
ies in Sweden. More analytical approaches to this problem have also been addressed.
Babusiaux (2003), Nejad (2007), Pierru (2007), and Babusiaux and Pierru (2007)
have proposed different methods for allocating the CO2 emissions among the dif-
ferent refinery products produced.
Elkamel et al. (2008) propose a MILP for the production planning of refinery
processes. They consider how to find suitable CO2 mitigation options for the pro-
cessing units that meet both a CO2 emission target and the final product demand
while maximizing profit.
Zhang and Hua (2007) propose a MILP model for a multi-period planning model
that considers the integration of the processing system and the utility system for the
refinery industry. The objective here is to determine an optimal material and energy
flow in order to maximize the overall profit.
Uncertainty is present in different forms in the different sub-system in the
refinery. Some of the latest work that considers uncertainty in the planning and
scheduling of refinery activities is by Neiro and Pinto (2005, 2006), Pongsakdi
et al. (2006), and Zimberg and Testuri (2006). Pongsakdi et al. (2006) address the
planning of crude oil purchasing and its processing based on the network structure
proposed by Pinto et al. (2000). Uncertainty is considered both in product prices and
product demands. The stochastic problem is modeled as a linear two stage stochastic
model with recourse and is solved using a special implementation of the average
124 J. Bengtsson and S.-L. Nonås

sampling algorithm introduced by Aseeri and Bagajewicz (2004). Test results show
that in comparison to the stochastic solution the deterministic solution has a lower
expected gross refinery margin and a larger risk. The profit is maximized taking into
account product revenues, crude oil costs, inventory costs, and cost of unsatisfied
demand.
Zimberg and Testuri (2006) consider the crude oil procurement and processing
for a case company that has a specific focus on the bunker fuel oil production.
Generally, the gas oil, diesel, and fuel oil products are more dependent on the right
crude oil mix than the gasoline products. A simplified two stage stochastic process
is considered. The first stage decision, purchasing of crude oil, is taken two months
before the second stage decisions, the processing of the crude oil and blending of
intermediate products. A case study that compares the deterministic (mean values)
and stochastic solution of the problem is presented. The risk is not considered, only
the expected refinery margin. Neiro and Pinto (2005, 2006), as previously discussed,
take into consideration uncertainty in product demand, product prices and crude oil
cost. The uncertainty is modeled in discrete scenarios and weighted according to the
probability of occurrence.
Pitty et al. (2008) and Koo et al. (2008) develop a decision support for an inte-
grated supply chain, which in this case means that it handles activities such as crude
oil supply and transportation in addition to intra-refinery activities. The decision
support is based on a simulation-optimization framework and takes stochastic vari-
ations in transportation times, yields and prices into account. Pitty et al. present the
complete dynamic model whereas Koo et al. use the decision support to optimize
the design and operation in three different supply chain related cases.
Current research has to a large degree been focused on modeling and analyzing
different types of refinery planning problems and has used commercial solvers as
GAMS (OSL, DICOPT, CONOPT) with the different solution approaches they offer
to solve the outlined problems. Nonlinearity is considered in some degree and for
some problems multiple periods is proposed. Also, a supply chain view has been
considered. Some recent papers consider decomposition strategies to solve the large
complex nonlinear planning problem. Future research should consider methods for
solving the complex MINLP problems more efficiently and focus on more advanced
methods that consider uncertainty in market data.

3.2 Production Scheduling

In the literature, MILP models are generally outlined for production scheduling
problems. The models focus on part of the refinery activities and incorporate dif-
ferent degree of details regarding the blending and processing in the processing
units. The models are generally solved using standard commercial solvers as GAMS
(CPLEX, OSL, DICOPT, CONOPT). Pinto et al. (2000) propose a discrete MILP
model for the production and distribution scheduling of diesel. The scheduling
Refinery Planning and Scheduling: An Overview 125

system considered a set of crude distillation units that produce specified intermedi-
ate (or end) products for storage in a set of intermediate tanks before they are mixed
and sent through pipelines to supply the consumer market where there is a known
demand. The proposed MILP model considers linear blending relations. Results are
reported for a case company considering market demand for three different types of
diesel oil. A scheduling horizon of one day considering one hour time intervals is
addressed.
Joly et al. (2002) outline both a discrete MINLP model and a discrete MILP
model for the scheduling problem that considers decisions related to mixing, storage
and distribution. The system configuration includes 1 deasphalting unit, 1 cracking
unit, 2 diluents storage tanks, 15 storage tanks for final products, and 2 pipelines.
The nonlinear terms in the MINLP model refer to the calculation of viscosity in
the oil mix and are linearized in the MILP model. The models produced similar
results in terms of solution quality, while the solution time for the MILP model was
a bit longer. Moro and Pinto (2004) and Joly et al. (2002) also discuss a scheduling
problem that addresses how to make use of the given raw materials, processing units,
and storage in order to meet the LPG product deliveries. They did not investigate
formulations related to product quality.
Goethe-Lundgren et al. (2002) outline a discrete MILP model for a refinery
production scheduling problem considering one distillation unit and two hydrotreat-
ment units that can each operate in 5–10 modes. The model considers how to run
the processing units in an optimal manner to minimize production and inventory
costs. The model can also be used to evaluate the feasibility and the production cost
for given product and crude oil shipping plans. To make the base schedule robust,
Goethe-Lundgren et al. (2002) implemented new constraints in the model that as-
sure that enough end products are available if a tanker should arrive one day too
early, and to assure enough storage capacity if a tanker is delayed one time period.
They also report how the flexibility decreased (the cost increased) when a more
robust schedule was offered.
Jia and Ierapetritou (2004) propose continuous time MILP formulations for spe-
cific crude oil, production and product scheduling problems. A lube-oil refinery is
considered for the processing units parts. The study proposes a continuous time
MILP formulation for the scheduling problem that takes into account material bal-
ance in tanks and production units; capacity, allocation, and sequence constraints;
and demand constraints.
Due to the complexity of the problem, the current work proposed for the pro-
duction scheduling relaxes most of the nonlinear relations, and only simple refinery
systems or sub-systems of the refinery topology are considered. Future work in this
area should focus on formulating models and finding corresponding solution ap-
proaches that enable companies to solve nonlinear production scheduling models of
real-world sizes in a reasonable time. In addition, new work that studies how the
daily scheduling decisions might best be incorporated in the production planning
and how uncertainty in the market data could be modeled in the scheduling problem
would also benefit the gas and oil industry.
126 J. Bengtsson and S.-L. Nonås

4 Product Blending and Recipe Optimization

The product blending problem concerns the best way to mix different components
and additives in order to meet quality and demand requirements of the final products.
Several final products can be produced out of a number of combinations of compo-
nents, but some properties of the final product do not show linear relationships, e.g.,
pour point of diesel. These relationships put requirements on the optimization model
so that the nonlinearity must be handled in some way.
The product blending problem is usually split into a long-term problem and a
short-term problem. In the case of long-term situation, the problem is basically
to determine optimal recipes that maximize profit given quality specifications and
quantity requirements. In the short-term situation, detailed operational and temporal
constraints come into play and the basic issue becomes that of scheduling.
Glismann and Gruhn (2001) claim that short-term scheduling of blending pro-
cesses is more complicated than scheduling of most other processes because of
the option of blending a product in many different ways. In the scheduling of
blending, consideration of recipe optimization and short-term scheduling within
an integrated approach becomes necessary. To address this Glismann and Gruhn
develop an integrated approach to coordinate nonlinear recipe optimization and
short-term scheduling of blending processes. They propose a two-level optimization
approach that sets up a large scale NLP model to determine product quantities and
recipes. Given the result from the NLP model, a MILP model based on a resource-
task-network is used for the scheduling problem to optimize resource and temporal
aspects. Both models are discrete time models. Whereas the NLP model maxi-
mizes profit, the MILP model minimizes deviations from given tank volume goals.
Glismann and Gruhn also present alternative strategies to handle situations where a
given goal cannot be met. They advocate integrating the planning and scheduling by
using an iterative approach so that if the goal at the scheduling level cannot be met
(due to bottlenecks), then the new information will be brought back to the planning
level and the modified NLP problem solved again. After this, the MILP problem
would be reconsidered until a feasible or satisfying solution is found.
Jia and Ierapetritou (2004) consider scheduling of gasoline blending and distri-
bution as one of three sub-problems. The other two sub-problems consider crude oil
and processing units. They assume that the recipe of each product is fixed, in order
to keep the model linear. The problem is modeled as a MILP-problem in contin-
uous time and based on the state-task-network (STN) representation introduced by
Kondili et al. (1993). In Jia and Ierapetriou the objective function is formulated such
that certain flows are as close as possible to a goal flow, but they also mention that
other objective functions can be used.
Mendez et al. (2006) point out that there are a number of mathematical pro-
gramming approaches available to the short-term blending and scheduling problem.
But in order to reduce problem difficulty, most of them rely on special assump-
tions that generally make the solution inefficient or unrealistic for real world cases.
Refinery Planning and Scheduling: An Overview 127

Some of the common simplifying assumptions are (a) fixed recipes are predefined,
(b) components and production flow rates are known and constant, and (c) all prod-
uct properties are assumed to be linear.
Mendez et al. develop a novel iterative MILP formulation for the simultaneous
optimization of blending and scheduling and formulate the problem both in discrete
and continuous time. The components flow from the processing unit is stored in
component tanks before the components are blended in blend headers and sent to
product tanks. The resulting nonlinear MINLP blending and shipment problem is
modeled as a successive MILP problem. The objective function maximizes profit
and is based on the assumption that the cost of components can be observed or
determined.
Mendez et al. also highlight the fact that the multi-period product blending and
product shipping problem is a complex and highly constrained problem where
feasible solutions may be difficult to find. To increase the speed of the solution
procedure, preferred product recipes could be included in the problem to help find
a feasible solution more quickly. To avoid infeasible solutions, Mendez et al. pro-
pose to include penalty for deviation from preferred product recipe and penalties
for deviations from specified product qualities. They also propose to allow pur-
chase of components at a higher cost from a third-party to relax minimum inventory
constraints.
Future work might focus on how to determine the component value and the
preferred product receipt, in order to optimize the combined performance of the
short-term blending and product shipments and how to coordinate the scheduling
decisions with long-term planning decisions.
In the literature, the values of the components are commonly presented as a
known value, as in Mendez et al. (2006). Often, however, the refinery’s value of
a certain component is unknown due one of two reasons; the component’s value
cannot be found from an external market or the value is not appropriate since lead
times makes this option infeasible. A variety of methods, based on marginal val-
ues of components and properties and product values, can be used to determine the
value of the components, and special attention should be made to the value that is
used because different component values can give different optimal blends.
We know that the short-term blending decision is affected by two facts: (a) flexi-
bility in the short-term is restricted, making it sometimes hard to stick to the recipe,
which is considered optimal in the longer term, typically given by the planning
model and (b) the relative value of blending components at the blending point in
time might be different from the value determined in the long-term optimization.
Thus, in the short-term, another recipe may be more profitable than the long-term
optimal recipe, and the deviations from the long-term recipe may indicate that other
blending recipes should be used. In the ideal world, it would be possible to observe
the values of components, and in order to more closely approximate the values
of these components, there must be integration between short-term and long-term
decision.
128 J. Bengtsson and S.-L. Nonås

5 Discussion and Further Research

This chapter has analyzed papers that consider planning and scheduling problems in
refineries. The papers consider planning and scheduling problems mainly for refin-
ery sub-systems and for refinery supply chains, in different forms and with different
degrees of detail. It has been common to use commercial MILP and MINLP solvers
to address the refinery model proposed. In addition, specialized algorithms have
been proposed to solve specific industry sized problems in a reasonable time. Un-
fortunately, no general solution technique has so far been outlined that can solve
real world problems in a satisfactory manner.
Due to the complexity of the refinery planning and scheduling problem, the
works proposed to this date relax most of the nonlinear relations. Future work should
focus on formulating correct NLP that take into account all aspects of the refinery
sub-systems and on developing solution techniques that enable companies to solve
nonlinear scheduling models of real-world sizes in a reasonable time.
Better coordination of refinery activities can increase throughput, reduce qual-
ity give away and demurrage, and improve the refinery’s ability to evaluate special
opportunities that may arise. Future research should consider how to coordinate the
scheduling decisions with the long-term planning decisions.
Beyond developing solution techniques and more advanced models, there is cur-
rently an increased focus on environmental impact from activities associated with
refining of crude oil. Tougher environmental regulations on oil products set by au-
thorities and an increased focus on reduction of CO2 emissions put new constraints
and requirements on decision making and adds more complexity to an already com-
plex situation. Given contemporary consensus about the environment, more research
focusing on environmental impact is needed.

References

Adhitya, A., Srinivasan, R., & Karimi, I. A. (2007). Heuristic rescheduling of crude oil operations
to manage abnormal supply chain events. AIChE Journal, 53, 397–422.
Aseeri, A., & Bagajewicz, M. J. (2004). New measures and procedures to manage financial risk
with applications to the planning of gas commercialization in Asia. Computers and Chemical
Engineering, 28, 2791–2821.
Babusiaux, D. (2003). Allocation of the CO2 and pollutant emissions of a refinery to petroleum
finished products. Oil & Gas Science and Technology-Revue del Institut Francais du Petrole,
58, 685–692.
Babusiaux, D., & Pierru, A. (2007). Modeling and allocation of CO2 emissions in a multiproduct
industry: the case of oil refining. Applied Energy, 84, 828–841.
Elkamel, A., Ba-Shammakh, M., Douglas, P., & Croiset, E. (2008). An optimization approach for
integrating planning and CO2 emission reduction in the petroleum refining industry. Industrial
& Engineering Chemistry Research, 47, 760–776.
Floudas, C. A., & Lin, X. X. (2004). Continuous-time versus discrete-time approaches for schedul-
ing of chemical processes: a review. Computers and Chemical Engineering, 28, 2109–2129.
Refinery Planning and Scheduling: An Overview 129

Glismann, K., & Gruhn, G. (2001). Short-term scheduling and recipe optimization of blending
processes. Computers and Chemical Engineering, 25, 627–634.
Goethe-Lundgren, M., Lundgren, J. T., & Persson, J. A. (2002). An optimization model for refinery
production scheduling. International Journal Production Economics, 78, 255–270.
Holmgren, K., & Sternhufvud, C. (2008). CO2 -emission reduction costs for petroleum refineries
in Sweden. Journal of Cleaner Production, 16, 385–394.
Ishizuka, I. T., Ueki, O., & Okamo, U. (2007). Improve refiner margins with integrated planning
and scheduling systems. Hydrocarbon Processing, 86, 75–79.
Joly, M., Moro, L. F. L., & Pinto, J. M. (2002). Planning and scheduling for petroleum refineries
using mathematical programming. Brazilian Journal of Chemical Engineering, 19, 207–228.
Jia, Z., & Ierapetritou, M. (2003). Mixed-integer linear programming model for gasoline blending
and distribution scheduling. Industrial & Engineering Chemistry Research, 42, 825–835.
Jia, Z., & Ierapetritou, M. (2004). Efficient short-term scheduling of refinery operations based on
continuous time formulation. Computers and Chemical Engineering, 28, 1001–1019.
Kelly, J. D., & Mann, J. L. (2003a). Crude oil blend scheduling optimization: an application with
multi-million dollar benefits - Part 1. Hydrocarbon Processing, 82, 47–53.
Kelly, J. D., & Mann, J. L. (2003b). Crude oil blend scheduling optimization: an application with
multi-million dollar benefits - Part 2. Hydrocarbon Processing, 82, 72–79.
Kondili, E., Pantelides, C. C., & Sargent, R. W. H. (1993). A general algorithm for scheduling
batch operations. Computers and Chemical Engineering, 17, 211–227.
Koo, L. K., Adhitya, A., Srinivasan, R., & Karimi, I. A. (2008). Decision support for integrated
refinery supply chains – part 2. Design and operation. Computers and Chemical Engineering,
32, 2787–2800.
Kuo, T. -H., & Chang, C. -T. (2008). Application of a mathematic programming model for in-
tegrated planning and scheduling of petroleum supply networks. Industrial & Engineering
Chemistry Research, 47, 1935–1954.
Lee, H., Pinto, J. M., Grossmann, I. E., & Park, S. (1996). Mixed-integer linear programming
model for refinery short-term scheduling of crude oil unloading with inventory management.
Industrial & Engineering Chemistry Research, 35, 1630–1641.
Li, W. K., Hui, C. W., Hua, B., & Tong, Z. (2002). Scheduling crude oil unloading, storage, and
processing. Industrial & Engineering Chemistry Research, 41, 6723–6734.
Li, J., Li, W., Karimi, I. A., & Srinivasan, R. (2007). Improving the robustness and efficiency of
crude scheduling algorithms. AIChE Journal, 53, 2659–2680.
Mendez, C. A., Grossmann, I. E., Harjunkoski, I., & Kaboré, P. (2006). A simulation optimiza-
tion approach for off-line blending and scheduling of oil-refinery operations. Computers and
Chemical Engineering, 30, 614–634.
Moro, L. F. L., & Pinto, J. M. (2004). Mixed-integer programming approach for short-term crude
oil scheduling. Industrial & Engineering Chemistry Research, 43, 85–94.
Moro, L. F. L., Zanin, A. C., & Pinto, J. M. (1998). A planning model for refinery diesel production.
Computers and Chemical Engineering, 22, 1039–1042.
Nejad, A. T. (2007). Allocation of CO2 emissions in joint product industries via linear program-
ming: a refinery example. Oil & Gas Science and Technology-Revue del Institut Francais du
Petrole, 62, 653–662.
Neiro, S., & Pinto, J. M. (2004). A general modeling framework for the operational planning of
petroleum supply chains. Computers and Chemical Engineering, 28, 871–896.
Neiro, S. M. S., & Pinto, J. M. (2005). Multiperiod optimization for production planning of
petroleum refineries. Chemical Engineering Communication, 192, 62–88.
Neiro, S. M. S., & Pinto, J. M. (2006). Langrangean decomposition applied to multiperiod planning
of petroleum refineries under uncertainty. Latin American Applied Research, 36, 213–220.
Pierru, A. (2007). Allocating the CO2 emissions of an oil refinery with Aumann-Shapley prices.
Energy Economics, 29, 563–577.
Pinto, J. M., & Moro, L. F. L. (2000). Planning model for petroleum refineries. Brazilian Journal
of Chemical Engineering, 17, 575–586.
130 J. Bengtsson and S.-L. Nonås

Pinto, J. M., Joly, M., & Moro, L. F. L. (2000). Planning and scheduling models for refinery oper-
ations. Computers and Chemical Engineering, 24, 2259–2276.
Pitty, S. S., Li, W., Adhitya, A., Srinivasan, R., & Karimi, I. A. (2008). Decision support for
integrated refinery supply chains – part 1. Dynamic simulation. Computers and Chemical En-
gineering, 32, 2767–2786.
Pongsakdi, A., Rangsunvigit, P., & Siemanond, K. (2006). Financial risk management in the plan-
ning of refinery operations. International Journal of Production Economics, 103, 64–86.
Reddy, P. C. P., Karimi, I. A., & Srinivasan, R. (2004a). A novel solution approach for optimizing
scheduling crude oil operations. AIChE Journal, 50, 1177–1197.
Reddy, P. C. P., Karimi, I. A., & Srinivasan, R. (2004b). A new continuous time formulation for
scheduling crude oil operations. Chemical Engineering Science, 59, 1325–1341.
Szklo, A., & Schaeffer, R. (2007). Fuel specification, energy consumption and CO2 emission in oil
refineries. Energy, 32, 1075–1092.
Stommel, J., & Snell, B. (2007). Consider better practices for refining operation. Hydrocarbon
Processing, 86, 105–109.
Zhang, B. J., & Hua, B. (2007). Effective MILP model for oil refinery-wide production planning
and better energy utilization. Journal of Cleaner Production, 15, 439–448.
Zimberg, B., & Testuri, C. E. (2006). Stochastic analysis of crude oil procurement and process-
ing under uncertain demand for bunker fuel oil. International Transactions in Operational
Research, 13, 387–402.