You are on page 1of 21

Maritime Policy & Management

The flagship journal of international shipping and port research

ISSN: 0308-8839 (Print) 1464-5254 (Online) Journal homepage: http://www.tandfonline.com/loi/tmpm20

Improving maritime inventory routing: application


to a Brazilian petroleum case

Gustavo Souto dos Santos Diz, Fabrício Oliveira & Silvio Hamacher

To cite this article: Gustavo Souto dos Santos Diz, Fabrício Oliveira & Silvio Hamacher (2016):
Improving maritime inventory routing: application to a Brazilian petroleum case, Maritime
Policy & Management, DOI: 10.1080/03088839.2016.1216622

To link to this article: http://dx.doi.org/10.1080/03088839.2016.1216622

Published online: 22 Aug 2016.

Submit your article to this journal

Article views: 24

View related articles

View Crossmark data

Full Terms & Conditions of access and use can be found at


http://www.tandfonline.com/action/journalInformation?journalCode=tmpm20

Download by: [University of Nottingham] Date: 02 September 2016, At: 11:22


MARITIME POLICY & MANAGEMENT, 2016
http://dx.doi.org/10.1080/03088839.2016.1216622

Improving maritime inventory routing: application to a Brazilian


petroleum case
a
Gustavo Souto dos Santos Diz , Fabrício Oliveirab and Silvio Hamacher b

a
Logistic Department, PETROBRAS, Rio de Janeiro, Brazil; bDepartment of Industrial Engineering, Pontific Catholic
University, Rio de Janeiro, Brazil

ABSTRACT KEYWORDS
The crude oil offloading and supply problem (COSP) is a type of opera- Maritime inventory routing;
tion maritime inventory routing (MIR) problem encountered by petro- mixed-integer linear
leum companies. In COSP, the company not only is responsible for the programming; fixed charge
network flow formulation;
ship scheduling to carry the crude oil from production sites to discharge
ship scheduling
ports but also must maintain inventory levels at both ports (production
and consumption) between safety operational bounds to avoid disrup-
tions in its crude oil production and/or refining processes. We show how
to improve significantly the decision-making process in a Brazilian pet-
roleum company using a mixed-integer linear programming (MILP)
model to represent COSP. Comparison tests with a current ship-schedul-
ing method adopted in the company indicated that the use of the MILP
model increased the transportation efficiency and reduced costs by 20%
on average. In addition to the quantitative gains, the use of a MILP
model to solve COSP has succeeded when encountering real-life events,
such as variation in production or consumption rates, berth unavailabil-
ity, and changes in the storage capacities at ports.

1. Introduction
Maritime transportation is one of the most important modals for global trading. UNCTAD (2013)
reports that approximately 80% of the total volume traded around the world are carried by sea. In
2012, 9.2 billion tons of goods were loaded at ports worldwide, and tanker trade (crude oil,
petroleum products, and gas) contributed nearly one-third of the total. Logistic problems invol-
ving maritime transportation have received increased focus from industry and academia over the
past few years (Fagerholt 2004; Cho and Perakis 2001; Christiansen, Fagerholt, and Ronen 2004;
Christiansen et al. 2013; Meng and Wang 2010). In particular, the petroleum industry and crude
oil transportation problems have receiving considerable attention (Bausch, Brown, and Ronen
1998; Perakis and Bremer 1992; Bremer and Perakis 1992; Furman et al. 2011; Rocha, Grossmann,
and De Aragão 2013).
In this paper, we focus on a maritime transportation problem known as maritime inventory
routing (MIR). In problems like MIR, the player responsible for carrying the products from the
supplier ports to the consumer ports is also in charge of managing the inventory level at the ports.
MIR problems feature the supply chain of many bulk product industries, such as grain, coal, iron
ore, and petrochemicals. In particular, the petroleum industry offers an interesting inventory-

CONTACT Silvio Hamacher hamacher@puc-rio.br Department of Industrial Engineering, Pontific Catholic University,
Rio de Janeiro, 22451-900, Brazil
© 2016 Informa UK Limited, trading as Taylor & Francis Group
2 G. S. D. S. DIZ ET AL.

routing environment for maritime transportation due to the need for transportation of huge
volumes combined with limited storage capacities at consumer and production sites.
This paper presents the case of a vertically integrated Brazilian petroleum company that needs
to carry crude oil from production sites to onshore consumer terminals. In addition, it is vital to
maintain the crude oil inventory level between certain operational bounds at production and
consumption points to avoid interruption of production processes because of a lack of storage
capacity (at production sites) and/or inventory (at consumption terminals). This problem,
referred to in this study as the crude oil offloading and supply problem (COSP), has the
characteristics of a MIR problem; it is not only computationally complex to solve but also
enormous in terms of quantity of assets and operational costs.
COSP is highly capital intensive. As an illustration, the company spends approximately US$ 1.2
million per day on fuel expenses. The average crude oil inventory level along the entire supply
chain is approximately 40 million barrels, which is responsible for the high inventory costs.
Additionally, the inventory managing activity is extremely important in COSP. The violation of
inventory physical limits at ports can represent crude oil production reductions or disruptions in
the refining process. A single day of interruption of average crude oil production may represent
more than US$ 5.0 million (using US$ 60.00 as the price of a crude oil barrel), which is more than
four times the daily fuel expense of the company’s entire fleet.
Considering the huge operational costs involved in COSP, the need for a decision support
system based on optimization techniques is mandatory to maintain an efficient operation. Thus,
the academic literature offers a wide range of theoretical mathematical formulations to support
decision in MIR problems. This type of problem has been studied in academia since the late
1990s. The first studies regarding MIR problems address the ammonia supply chain problem. To
address this problem, Christiansen (1999) proposed a mixed-integer linear programming (MILP)
formulation, whereas Flatberg et al. (1999) proposed a hybrid approach for the same problem
using a heuristic combined with a linear programming (LP) solver.
After these pioneer studies, researchers and industry noticed the considerable monetary
benefits that could be achieved by coordinating the planning activities of inventory management
and ship scheduling. Additionally, coordination could also improve the robustness and flexibility
of operations (Christiansen et al. 2013). This perception resulted in an increased focus on MIR
problems during the last decade.
Although researchers and industry recognized that optimization models could improve the
efficiency of MIR problems, they also realized how complex it is to solve this type of problem to
optimality for real-life instances (Furman et al. 2011; Agra, Andersson, et al., 2013; Papageorgiou
et al. 2014). However, Agra, Andersson, et al., 2013 claims that MIR problems are usually very
intensive in capital and a modest improvement in fleet scheduling and load quantities would
mean great economic benefits.
Over the last decade, researchers have been struggling to overcome problem complexity
barriers that prevent one to solve real-life instances to optimality. Several approaches have been
developed and implemented in the past few years. Heuristics and hybrid models are mostly used
to implement decision support systems in the industry due to its capacity on dealing with large
and complex problems (see Dauzère-Pérès et al. 2007; Furman et al. 2011; Fodstad et al. 2010).
However, the majority of approaches in the literature use MILP techniques to handle the problem
(see Christiansen 1999; Al-Khayyal and Hwang 2007; Christiansen et al. 2007; Furman et al. 2011;
Song and Furman 2013; Agra, Christiansen, et al., 2013; Papageorgiou et al. 2014).
Concerning exact methods such as MILP, researchers have been investigating decomposition
techniques and the development of reformulations to improve computational performance.
Christiansen (1999) proposed a decomposition of MIR problem into two subproblems: ship
routing and inventory management. Agra, Andersson, et al., 2013 have proposed a reformulation
called fixed charge network flow (FCNF) model and have implemented some important valid
inequalities that provide a tighter formulation in terms of linear relaxation, which would speedup
MARITIME POLICY & MANAGEMENT 3

Figure 1. Crude oil flow in the company’s supply chain.

the convergence of Branch-and-Cut algorithms. Apart from these studies, many authors have
proposed valid inequalities to improve their models (Persson and Göthe-Lundgren 2005; Al-
Khayyal and Hwang 2007; Grønhaug and Christiansen 2009; Engineer et al. 2012; Rocha,
Grossmann, and De Aragão 2013; Song and Furman 2013; Agra, Christiansen, and Delgado
2013; Papageorgiou et al. 2014). Although the recent efforts to improve modeling and solving
techniques of MIR problems, there are still space for improvements in this area (Papageorgiou et
al. 2014). A more comprehensive review about mathematical formulations for MIR problems is
presented in Agra, Andersson, et al., 2013 and Papageorgiou et al. (2014).
Despite the recent developments available in literature for dealing with MIR problems, there is
still an absence of works that addresses how these recent academic developments could be
implemented in the petroleum industry. This paper tries to bridge this gap by assessing the
benefits that could be obtained with the use of a mathematical modeling solution in place of the
current practice in a Brazilian petroleum company. This paper’s objective is to demonstrate how
to improve the ship-scheduling decision in a real-life MIR problem, such as COSP. Thus, we
present how the FCNF formulation can be used to solve a set of real-life instances of the COSP
problem. To check the robustness and effectiveness of the solution provided by the FCNF model,
we compare the results against the solution obtained by the scheduling team using an empirical
method currently in use at the company. We also compare the use of the FCNF formulation to
support efficient ship scheduling utilizing common and complicating real-life situations such as a
variation of parameters, including the production/consumption rate, the storage capacity, the
berth availability, and the minimum stock level. The comparison of results focuses on four main
aspects of the problem: planning horizon, inventory control at production sites, inventory control
at onshore terminals, and transportation costs.
The main contribution of this paper is to demonstrate a real improvement in the quality of
ship-scheduling decisions. We quantify the gains obtained with the use of an optimization-
based system to support ship scheduling through a comparison test against the use of the
empirical method adopted in the company. Additionally, we analyze certain qualitative benefits
of using an optimization-based tool such as wider visibility, better planning of activities related
to the crude oil supply chain and the response to a variation in parameters along a planning
4 G. S. D. S. DIZ ET AL.

horizon, such as production/consumption rate, storage capacities, and planned shutdowns in


the ports.
The remainder of this paper is organized as follows. Section 2 presents the Brazilian petroleum
company problem and the empirical method adopted by their scheduling team to support their
ship-scheduling decisions. The third section discusses the FCNF formulation and the operational
aspects related to it. Section 4 presents a reference instance used to test the FCNF model and the
comparison against the empirical method adopted by the company. Concluding remarks and
future research are presented at the end of the paper. We present the mathematical formulation at
Appendix 1.

2. Crude oil offloading and supply problem


In this section, we describe the company’s crude oil flow, the main aspects of COSP, and the ship-
scheduling decision-making process adopted in the company (named hereinafter as empirical
method) to stream all of the produced crude oil and supply the onshore terminal.

2.1. Problem description


The crude oil supply chain management involves great effort to manage inventory, transport, and
ensure crude oil quality. Figure 1 provides an overview of the crude oil flow, from production sites
to refining plants. After the production phase, a fleet of dedicated vessels carries most of the crude
oil to onshore terminals (a small part is pumped through pipelines directly to onshore terminals).
From the onshore terminals, the crude oil either flows to refineries through pipelines or is
exported to foreign customers. It is also necessary to import crude oil to complement refinery
supplies.
The COSP problem consists of managing the crude oil inventory level in the production sites,
transporting it by vessels to the onshore terminals and guaranteeing that the inventory level at
terminals remains within operational limits. We do not consider crude oil transportation through
pipelines in this paper.
To address maritime transportation, the company has a predefined, limited, and dedicated
time-chartered fleet. Vessels are from different classes and sizes, and each one has its own bunker
consumption rate and speed. Bunker consumption represents the largest portion of the transpor-
tation cost; however, port fees are also considerable.
Each production site and onshore terminal has its own physical and operational restrictions,
such as maximum deadweight allowed, depth and draft limits. Such restrictions are safety
requirements that cannot be violated under any circumstance. Production sites and onshore
terminals also have a limited tank farm capacity, which is sufficient to store the crude oil for a
few days until a vessel arrives to offload or supply it again. Each port (in the production sites or
onshore terminals) also has a production/consumption rate and minimum and maximum inven-
tory operational levels.
It is important to emphasize that the parameters described above are not always fixed along the
planning horizon. Berths at onshore terminals may be unavailable during a planning horizon
because of maintenance activity. Production and consumption rates may vary over the planning
horizon. In addition, storage capacity may change because of tank inspection periods. These are
just some of the real-life situations that can occur anytime during a planning horizon and can
illustrate the variation in parameters.

2.2. Empirical method for ship scheduling


To address the complexity of COSP operations, the ship-scheduling activity plays an important
role. Ship scheduling is a daily short-term process responsible for defining which ship will carry
MARITIME POLICY & MANAGEMENT 5

the crude oil from a production point to an onshore terminal. Usually, the schedulers determine
the voyages for the vessels that will become available in the next day. A vessel is considered as
being available as soon as it has fully unloaded its cargo at an onshore terminal, which marks the
end of its current voyage and makes it available to start a new one. Once a vessel becomes
available, the team includes it in a list of available vessels, where all of them are free to receive a
new scheduling. The voyage includes one offloading operation in a maximum of two production
sites (visited sequentially) and one discharge operation at an onshore terminal
After having a list of available vessels (ready to start a new voyage), the scheduling team must
turn their attention to inventory levels. Their priority is to avoid interruptions in the crude oil
production due to lack of storage capacity at production sites. According to the scheduling team,
the ideal time for an offloading operation is before the crude oil stockpile attains the maximum
storage capacity in a production site; or when it attains the amount equal to the typical cargo size
of a production site. As they do not rely on any computational intelligence to make their decision,
they limit their actions to those production sites where the offloading operations must occur in
the next 7 days.
As preventing inventory level from attaining the maximum storage capacity in a production
site is what most influences the everyday ship-scheduling plan, they often overlook the inventory
level at onshore terminals when defining the offloading time. Inventory level at onshore terminals
is usually taken in consideration only when the offloading operation is already being performed.
When it is still possible, it influences the destination of the vessel.
We may summarize the process in the following steps that are repeated in a daily basis:

(1) Find a list of available vessels;


(2) Define new voyages for these vessels taken in consideration the inventory level at produc-
tions sites for a planning horizon of approximately 7 days ahead;
(3) Check inventory levels at onshore terminals to determine the destination of the vessel; and
(4) Revise the voyages already scheduled. If it is still possible, change destination port to avoid
inventory levels at onshore terminals to violate minimum operational bound.

An important observation is that this simplified approach is essential to maintain tractability of


the problem for the scheduling team. Following this process, the entire problem is reduced to a
smaller one that considers the following: part of the fleet (those that will be available in the next
day), inventory control at certain production points (those with an increased inventory level), and
a short planning horizon (approximately 7 days ahead). The scheduling team considers the
inventory level at onshore terminals only after the offloading operation has been decided. It
serves to determine the destination of the crude oil cargo. This strategy is followed every day,
always revising the scheduling plan made in the previous day (if necessary) and adding new
voyages to the plan for the vessels in the list of available vessels.
The scheduling team developed the empirical method based on their own experience. This
method does not consider costs in its priorities. Because of the complexity of COSP and the
magnitude of the monetary losses related to the interruption of processes, the main idea behind
the empirical method is to maintain the production and refining processes without any undesir-
able disruption. Fortunately, the academic literature offers plenty of mathematical formulations
that address MIR problems. Next, we briefly describe the characteristics of some mathematical
formulations available in the academic literature and operational aspects that directly influence
the ship-scheduling decision-making process when solving a COSP.

3. MIR formulations and operational aspects


One important issue regarding MIR problem formulation is how the time dimension is modeled.
According to Papageorgiou et al. (2014), mathematical formulations for MIR problems may be
6 G. S. D. S. DIZ ET AL.

classified as continuous-time or discrete-time models. According to Christiansen et al. (2013),


Agra, Andersson, et al., 2013, and Papageorgiou et al. (2014), continuous-time models are more
suitable when consumption and/or production rates are fixed, and discrete-time models are used
to overcome complicating factors associated with the oscillation of consumption and production
rates along a planning horizon. Because variations in parameters such as production/consumption
rates are important aspects of COSP, the formulation chosen to model COSP should be able to
represent this issue.
In discrete-time models, operation, waiting, and sailing times must be an integer number of
uniform time units (Agra, Andersson, et al., 2013). Tactical and operational planning problems
must incorporate certain day-to-day events to support scheduling decisions. Thus, time units
usually represent days, half-days, or hours, depending on the characteristics of the problem.
Despite the rounding needs of time lapses, discrete-time formulations offer many benefits.
Because variables have a time unit index, the representation of time-wise value changes in
parameters, such as production/consumption rates, storage capacity, and maximum and mini-
mum inventory levels along the planning horizon, is often easier to model. Discrete-time models
also facilitate the modeling of other real-life aspects, such as multiple berths per port and waiting
time because of berth congestion.
The idea behind discrete-time models can be represented in a time-space network diagram,
where vessels navigate through nodes that represent pairs of time units and ports. At each of these
nodes, the inventory is controlled and the vessel performs one of the following activities: waiting,
operating, or leaving a port. Usually, there is at least one variable to represent vessel activity at
each node. Figure 2 represents a time-space network structure describing the movements of a
vessel between ports along the planning horizon. In this example, the vessel leaves an artificial
origin port (Or) and sail (S) to port 1, where it arrives during period 1 and operates (O) for 2
periods. Thereafter, the vessel sails to port 3 where it arrives during time period 6, waits (W) for 1
period and operates for 2 more periods. Finally, during period 9, the ship sails to port 2, waits for
1 period, operates for another period, and finishes its voyage at an artificial destination port (D).
Certain discrete time models may adopt an extended network to represent the problem in
which the variable representing loading/discharging operation are split into two variables that
indicate whether the ship starts an operation at a port and whether it continues to operate in the
next time unit at the same port. According to Agra, Andersson, et al., 2013, this reformulation
allows one to take advantage of known valid inequalities, which leads to a more computationally
efficient formulation than the traditional discrete formulation found in the literature
(Christiansen et al. 2007; Agra, Andersson, et al., 2013; Papageorgiou et al. 2014). Refer to
Agra, Andersson, et al., 2013 for more details on the extended formulation, which is called FCNF.
Agra, Andersson, et al., 2013 have structured and presented comparison tests that indicate the
superior computational performance of the FCNF formulation when compared with the more

Figure 2. Time-space network representing a movement of a ship in an artificial problem.


MARITIME POLICY & MANAGEMENT 7

common discrete time formulation found in the literature. In this paper, we have selected the
FCNF formulation combined with the valid inequalities presented in Agra, Andersson, et al., 2013
to model COSP because of its capabilities in terms of modeling operational aspects and the best
computational performance for large instances that they have reported. These formulations and
the valid inequalities are presented in Appendix 1.
Regarding the modeling of operational aspects, our implementation is versatile enough to
represent COSP. It allows the representation of important real-life features.

● Heterogeneous fleet, including different ship classes, sailing speeds, and fuel consumption
rates;
● Compatibility between vessels and ports that represents operational aspects such as draft
restrictions, floating production storage and offloading unit (FPSO) mooring arrangements
and vessel class, deadweight limits, and other safety requirements;
● Consecutive offloading operations before discharging;
● Multi-berth ports that enable simultaneous discharge operations;
● Waiting time at ports due to berth congestion;
● Variation in parameters such as the production/consumption rate along time.

In the following section, we present computational experiments considering certain COSP


instances, focusing on the assessment of the benefits arising from the use of a suitable optimiza-
tion-based tool instead of the manual scheduling.

4. Computational tests and results


We tested an instance that represents COSP operations in the Brazilian north and northeast
(NNE) region, as described in Figure 3. Hereinafter, we will refer to this instance as the
reference instance. We have solved the reference instance using the FCNF model including
valid inequalities and considering a planning horizon of 15 days. In addition, the company’s
ship-scheduling team started their decision-making process for the reference instance using
the empirical method. Because the empirical method has a shorter view, it was repeated every
day until the end of the 15 days planning horizon. After completing the comparison using the
reference instance, we have simulated instances considering real-life events that require the
variation of parameters along a planning horizon. The same comparison tests were conducted
for these instances.
We have implemented the FCNF model with valid inequalities using the commercial optimiza-
tion platform AIMMS 3.13 and solved the problem using GUROBI 5.5 on an Intel Core i7 2.40
GHz personal computer. We defined a time limit of 3,600 seconds as the stop criterion for those
instances that did not converge to optimality.

4.1. Reference instance


The reference instance is a typical ship-scheduling scenario encountered everyday by the
scheduling team. The scenario is composed of seven production sites and four onshore
terminals with their real production and demand rates, maximum storage capacities, mini-
mum inventory level, initial stock, port fees, and loading rates. The available fleet is
composed of 11 vessels of different classes and sizes. Each vessel has its own starting time
and origin, storage capacity, waiting and sailing costs per day. We assume vessels are empty
at beginning of planning horizon. Compatibilities between vessels and ports are also
represented to express the safety and operational restrictions previously noted. The time
unit was defined as one day, following the recommendation of the company’ scheduling
team.
8 G. S. D. S. DIZ ET AL.

Figure 3. Map of Brazilian onshore terminals, refineries, and production sites in the NNE region.

Table 1. Data for the reference instance.


Production Initial stock Storage capacity Minimum level Production rate Days to inventory upper bound
site (m3) (m3) (m3) (m3) (days)
COARI 6,145 44,177 2,300 4,841 7.9
XAREU 23,799 40,000 4,000 1,698 9.5
GUAMARE 88,465 125,381 17,204 6,500 5.7
MACEIO 7,445 14,514 2,225 800 8.8
ARACAJU 45,463 147,000 13,000 6,500 15.6
TNC 30,515 63,636 6,716 4,200 7.9
PIRANEMA 5,000 37,000 – 1,800 18

Table 1 presents the main parameters related to production sites in the reference instance. Such
information and fleet availability are available for the scheduling team every day and based on this
information, the team makes the ship-scheduling decision.
According to the empirical method, to define the time of each offloading operation at the
production sites, the scheduling team gives priority to those sites where the inventory level is
closer to the maximum capacity. Inventory levels at onshore terminals are checked only on
demand or when advised by the onshore terminal inventory management team.
In the reference instance, according to the empirical method adopted by the scheduling team,
the production site GUAMARE has the highest priority to receive an offloading operation (5.7
days to reach maximum level). Thereafter, COARI and TNC (both with 7.9 days to reach
maximum level) have the second highest priority. To schedule these offloading operations, the
team searches for vessels that are compatible with these production site ports and are available to
reach the production site in time.
MARITIME POLICY & MANAGEMENT 9

4.2. Results comparison


After the end of the planning horizon, we compared the results obtained with both methods for
the reference instance according to four aspects: planning horizon, inventory level at production
sites, inventory level at onshore terminals, and transportation costs.

4.2.1. Planning horizon


The ship-scheduling decisions for day 1 according to empirical method is described in Table 2.
Notice that the planning horizon is no longer than 6 days (the time of the latest offloading
scheduled in MACEIO). The XAREU production site is 9.5 days from reaching the inventory
upper bound, and there is no offloading operation scheduled at that site. The scheduling team will
consider this offloading operation and the remainder of the problem the next day.
Figure 4 presents a snapshot of the inventory level evolution and offloading operations in the
COARI production site, when we use the FCNF model and the empirical method to solve the
reference instance in the first time-unit of the horizon. We divide Figure 4 in four parts a, b, c,
and d from the left to the right and from top to bottom. Figure 4a represents the inventory level at
COARI production site, while the Figure 4(c) shows the time of each offloading operation and the
amount of crude oil offloaded each time. Both information represent the solution provided by the
FCNF model. In Figure 4(b), we show the projection of inventory level at the same production site
obtained by the scheduling team (using the empirical method) on the first day of planning
horizon. Figure 4(d) presents the moment and amount of product of the first offloading scheduled
using the empirical method. Notice that when using the FCNF model, the offloading operations

Table 2. Ship-scheduling decision based on empirical method.


Days inventory Offloading date Onshore Days inventory
Production site upper bound (days) Vessel (time unit) Cargo size (m3) terminal upper bound (days)
COARI 7.9 Marta 3 15,000 Manaus 11.0
XAREU 9.5 9.5
GUAMARE 5.7 Elka Aristotle 0 70,000 Temadre 16.4
MACEIO 8.8 Aliakmon 6 10,000 Manaus 21.3
ARACAJU 15.6 15.6
TNC 7.9 Socrates 4 40,000 Suape 17.4
PIRANEMA 17.8 17.8

Figure 4. Inventory level forecast at COARI production site using the FCNF model and empirical method to support scheduling
decisions on the first day of planning horizon.
10 G. S. D. S. DIZ ET AL.

are planned for a 15-day planning horizon, which is much longer than the planning horizon
considered in the empirical method (about 7 days ahead). In the COARI production site, for
example, the FNCF model provides complete offloading scheduling with three operations. The
empirical method only anticipates the next offloading operation, providing much less visibility for
all of the teams involved in the offloading operations.

4.2.2. Inventory level at production site


When we analyze the inventory management at production sites, the empirical method did not
allow the inventory level to attain the maximum level in any of the seven production sites. This
occurs because this method was developed focusing on avoiding disruptions in the crude oil
production process. Figure 5 presents the inventory level at the COARI production site at the end
of the planning horizon, after applying the empirical method for the 15 days of the planning
horizon. Notice that the inventory level is nearly the same as in the FCNF model suggestion. This
result confirms that empirical method provides considerable focus on inventory control in
production sites.

4.2.3. Inventory level at onshore terminal


In contrast to what we observed concerning inventory management at production sites, the
empirical method provides minimal focus for inventory level control at onshore terminals.
Figure 6 presents a snapshot of inventory evolution and discharging operations at SUAPE onshore
terminal during the planning horizon. We divide Figure 6 in four parts Figure 6(a–d) from left to
right and from top to bottom. Figure 6(a,c) presents, respectively, the inventory level evolution at
SUAPE onshore terminal and discharging operations (time and crude oil quantity), both accord-
ing to the solution provided by FCNF model. Note that the FCNF model provides a complete
supply scheduling for SUAPE onshore terminal, which is sufficient to maintain the inventory level
at that terminal between the bounds along the entire planning horizon. Figure 6(b,d) presents,
respectively, the inventory level evolution and discharging operations (time and amount of
product discharged at SUAPE onshore terminal) proposed by scheduling team (using the empiri-
cal method). The empirical method suggested the scheduling of two cargos discharging on days 8
and 9. The inventory level reached 46,000 m3 on day 8, 14,000 m3 lower than the minimum
operational level of 60,000 m3. The first cargo designated to supply SUAPE was scheduled during
day 1, and the destination was decided according to historical operations. When the team
observed that the inventory level at SUAPE would decrease to the minimum level before the

Figure 5. A view of inventory level at COARI using FCNF model and empirical method at the end of planning horizon.
MARITIME POLICY & MANAGEMENT 11

Figure 6. Inventory management at SUAPE onshore terminal using the FCNF model and empirical method to support
scheduling decisions.

arrival of the first cargo, the time had passed. They immediately scheduled a second cargo to
SUAPE; however, its best estimated time of arrival in SUAPE was also on day 8. The decision
made by the scheduling team represented at least two inefficiencies: the inventory level at SUAPE
decreased below the minimum operational level and an increased waiting cost of the second vessel
scheduled to that terminal due to berth congestion.

4.2.4. Transportation costs


In addition to the longer planning horizon and the focus on the entire problem, the FCNF model
provides a cost-optimized ship-scheduling decision, minimizing bunker consumption, port
expenses, and waiting costs. Although the empirical method does not consider the transportation
cost in the ship-scheduling decision, for the sake of comparison, we have calculated the total cost
of all of the scheduling performed during the 15 days horizon. It considers bunker consumption,
port expenses, and waiting costs involved in the transportation operation (known as demurrage
cost).
The empirical method suggested nine offloading operations during the planning horizon and
an estimated cost of US$ 1.71 million. Meanwhile, the FCNF model suggested 10 offloading
operations and an associated cost of US$ 1.49 million. It represents a reduction of 13% in
operational costs.
Therefore, the use of the FCNF model provides a higher quality ship-scheduling decision, for a
longer planning horizon, a wider visibility of the inventory control (at production sites and
onshore terminals), lower transportation costs and an integrated analysis of the entire scheduling
scenario.

4.3. Special cases considering variation of parameters


An important issue regarding COSP is that many of the parameters used to develop the schedul-
ing scenario may vary along the planning horizon. The variation of parameters performed
emulates many real-life situations as described in Section 2. The challenge here is how to schedule
the fleet when one cannot rely on average values to estimate parameters using the empirical
method.
To demonstrate how such situations influence the ship-scheduling decision and how FCNF
formulation address these, we have created a new set of instances based on the reference one in
which we simulate some real-life events.
12 G. S. D. S. DIZ ET AL.

● Instance 1: Planned berth shutdown, interrupting discharge operations for a predefined


period;
● Instance 2: Reduction of storage capacity and unavailability of offloading station in a
production site; and
● Instance 3: Increase in production rate.
● Instance 4: Planned refinery maintenance shutdown.
● Instance 5: Increase in consumption rate.
● Instance 6: Multiple maintenance operations in production sites, making offloading station
unavailable in several production sites.

Table 3 presents the main results obtained by the company’s scheduling team using the empirical
method to solve the new set of instances.
Note that in five of seven instances, there were incidences of demurrage cost. Usually, this
waiting time (and cost) is related to berth congestion at the port. The scheduling team guaranteed
the non-violation of maximum inventory bound at production sites in all the instances tested,
which means that there was no disruption in the crude oil production process. Another important
observation is that in six of seven instances there were violations of the minimum inventory level
at the onshore terminals. It confirms that the empirical method gives considerable attention to
inventory control at the production sites, instead of having a complete view of the whole system.
Table 4 presents in detail the transportation costs and the cost reduction obtained, when
solving the same set of instances using the FCNF model. Note that transportation costs decrease
in all instances tested and that the reduction reaches nearly 30% in the third instance. It is
important to emphasize that there was no incidence of demurrage cost and no minimum
inventory bound violation at onshore terminals in any of the instances tested when we use the
FCNF model to support the ship-scheduling decision.
These tests showed how important it is to predict certain events to have an efficient ship-
scheduling program. Additionally, the use of a mathematical model to support the ship-

Table 3. Main results obtained using empirical method to scheduling the fleet.
Empirical Method
Sailing Demurrage Port cost Maximum inventory bound Minimum inventory bound violation
Instances cost (US$) cost (US$) (US$) violation at production sites (m3) at onshore terminals (m3)
Reference 706,950 25,000 980,000 – 14,000
1 706,950 75,000 980,000 – 14,000
2 829,650 – 1,100,000 – 2,000
3 999,150 25,000 1,100,000 – 14,000
4 570,800 25,000 1,080,000 – –
5 853,200 – 1,200,000 – 21,600
6 813,500 175,000 1,080,000 – 9,900

Table 4. Reduction in transportation costs when using FCNF model to support ship-scheduling decision.
FCNF model
Sailing cost Demurrage cost Port cost Transportation cost Relative decrease in
Instances (US$) (US$) (US$) reduction (US$) transportation Cost (%)
Reference 487,050 – 1,000,000 224,900 13
1 487,050 – 1,000,000 274,900 16
2 487,050 – 1,000,000 442,600 23
3 487,050 – 1,020,000 617,100 29
4 524,400 – 840,000 311,400 19
5 652,800 50,000 960,000 390,400 19
6 579,850 25,000 960,000 503,650 24
MARITIME POLICY & MANAGEMENT 13

Table 5. Computational results using FCNF model.


Instances Objective function (US$) Lower bound (US$) Optimality gap (%) Computational time (seconds)
Reference 1,487,050 1,365,800 8.2 3,600
1 1,487,050 1,487,050 0.0 591
2 1,487,050 1,369,700 7.9 3,600
3 1,507,050 1,382,600 8.3 3,600
4 1,364,400 1,364,400 0.0 696
5 1,662,800 1,493,050 10.2 3,600
6 1,564,850 1,564,850 0.0 704

scheduling decision is vital to avoid loss of money due to transportation costs and disruption in
production.
Table 5 summarizes the computational aspects of the instances tested using the FCNF model.
Every have 34,725 variables (of which 8,010 are integers) and 33,764 constraints. Notice that
optimality was not reached for all instances within the time limit of 3,600 seconds. Nevertheless,
as can be seen in Table 4, the solutions obtained using the FCNF model presented a considerable
cost reduction for all instances evaluated.
The decomposition approach of the empirical method makes the problem manually tractable,
and because it is repeated every day, it relies on updated information. However, that approach
provides no visibility for the entire process and is time-wise short-sighted. Therefore, when using
the empirical method, one cannot consider the variation of parameters for the entire scheduling
scenario, and the ship-scheduling decision is jeopardized for having a short-sighted view of the
problem. In contrast, the use of the FCNF model to support ship-scheduling decisions allows a
more complete analysis of the entire COSP scenario, avoiding unnecessary costs and the violation
of inventory operational bounds. Additionally, a wider visibility of the supply chain operations is
very important for every player involved in the crude oil flow. An operational schedule composed
of a longer planning horizon and the anticipation of offloading and discharging operations would
be very useful. This schedule would help in the maintenance planning of the fleet, production sites
and onshore terminals and/or in every other activity that could affect or be affected for offloading
and discharging operations.

5. Conclusion
A MIR problem, named COSP and encountered by a Brazilian petroleum company, has been
addressed in this paper. We presented the empirical method adopted in the company to support
ship-scheduling decisions in COSP. The FCNF formulation associated with certain valid inequal-
ities has been used to solve a set of real-life instances of COSP. We compare the results with the
solutions provided by the empirical method adopted by the company. The tests also check the
FCNF model performance when encountering real-life events that are represented by variation of
parameters along the planning horizon.
We have shown that the use of the FCNF model to support ship-scheduling decisions in COSP
significantly improved the quality of the solution. Excluding the inventory control at production
sites, where the empirical method had comparable results, the FCNF model indicated better results in
every other aspect analyzed in this paper, particularly in those instances with a variation of
parameters. As shown, this model provides more visibility for every team involved in the offloading
operations and forecasts the inventory level at every port (production sites and onshore terminals)
for the entire planning horizon. This model avoids unnecessary costs, such as demurrage costs, and
considerably reduces the total transportation costs of the entire fleet. The average reduction in costs
was approximately 20%. In addition to the wider visibility of problem and the quantitative results, the
use of the FCNF model to support the scheduling decision also greatly reduces the time and manual
14 G. S. D. S. DIZ ET AL.

effort from the scheduling team when compared with the empirical method. This model allows the
team to spend their time performing more analytical activities and less manual work.
This paper showed that the use of an optimization-based support system for a ship-scheduling
decision in a MIR problem is a success, considerably improving the quality of decisions. The main
achievements of the use of the FCNF model are the cost reduction and the visibility of longer
planning horizon that enables one to tackle the problem of variation in parameters and provides
better planning of every activity that affects offloading operations. For future research, it would be
interesting to investigate the optimization of MIR problems considering stochastic events that
impact maritime transportation, such as bad weather that influences sailing, waiting and operating
time, unexpected variation in the crude oil production rate and/or the consumption rate of
refineries. It would be interesting to study how robust optimization could be applied to represent
uncertainty in operational time in ports and/or waiting time in onshore terminals, for example.

Disclosure statement
No potential conflict of interest was reported by the authors.

Funding
This work was not supported by any Research agency.

ORCID
Gustavo Souto dos Santos Diz http://orcid.org/0000-0002-6426-4270
Silvio Hamacher http://orcid.org/0000-0002-3980-0342

References
Agra, A., H. Andersson, M. Christiansen, and L. Wolsey. 2013. “A Maritime Inventory Routing Problem: Discrete
Time Formulations and Valid Inequalities.” Networks 62 (4): 297–314. doi:10.1002/net.v62.4.
Agra, A., M. Christiansen, and A. Delgado. 2013. “Mixed Integer Formulations for a Short Sea Fuel Oil Distribution
Problem.” Transportation Science 47 (1): 108–124. doi:10.1287/trsc.1120.0416.
Al-Khayyal, F., and S.-J. Hwang. 2007. “Inventory Constrained Maritime Routing and Scheduling for Multi-
Commodity Liquid Bulk, Part I: Applications and Model.” European Journal of Operational Research 176 (1):
106–130. doi:10.1016/j.ejor.2005.06.047.
Bausch, D. O., G. G. Brown, and D. Ronen. 1998. “Scheduling Short-Term Marine Transport of Bulk Products.”
Maritime Policy & Management 25 (4): 335–348. doi:10.1080/03088839800000057.
Bremer, W. M., and A. N. Perakis. 1992. “An Operational Tanker Scheduling Optimization System: Model
Implementation, Results and Possible Extensions.” Maritime Policy &Amp; Management 19 (3): 189–199.
doi:10.1080/751248660.
Cho, S.-C., and A. N. Perakis. 2001. “An Improved Formulation for Bulk Cargo Ship Scheduling with a Single
Loading Port.” Maritime Policy & Management 28 (4): 339–345. doi:10.1080/03088830010002755.
Christiansen, M. 1999. “Decomposition of a Combined Inventory and Time Constrained Ship Routing Problem.”
Transportation Science 33 (1): 3–16. doi:10.1287/trsc.33.1.3.
Christiansen, M., K. Fagerholt, B. Nygreen, and D. Ronen. 2007. “Maritime Transportation.” Handbooks in
Operations Research and Management Science 14: 189–284.
Christiansen, M., K. Fagerholt, B. Nygreen, and D. Ronen. 2013. “Ship Routing and Scheduling in the New
Millennium.” European Journal of Operational Research 228 (3): 467–483. doi:10.1016/j.ejor.2012.12.002.
Christiansen, M., K. Fagerholt, and D. Ronen. 2004. “Ship Routing and Scheduling: Status and Perspectives.”
Transportation Science 38 (1): 1–18. doi:10.1287/trsc.1030.0036.
Dauzère-Pérès, S., A. Nordli, A. Olstad, K. Haugen, U. Koester, M. P. Olav, G. Teistklub, and A. Reistad. 2007.
“Omya Hustadmarmor Optimizes its Supply Chain for Delivering Calcium Carbonate Slurry to European Paper
Manufacturers.” Interfaces 37 (1): 39–51. doi:10.1287/inte.1060.0276.
MARITIME POLICY & MANAGEMENT 15

Engineer, F. G., K. C. Furman, G. L. Nemhauser, M. W. P. Savelsbergh, and J. H. Song. 2012. “A Branch-Price-


And-Cut Algorithm for Single Product Maritime Inventory Routing.” Operations Research 60 (1): 106–122.
doi:10.1287/opre.1110.0997.
Fagerholt, K. 2004. “Designing Optimal Routesin a Liner Shipping Problem.” Maritime Policy & Management 31
(4): 259–268. doi:10.1080/0308883042000259819.
Flatberg, T., J. Havardtun, O. Kloster, and A. Lokketangen. 1999. “Combining Exact and Heuristic Methods for
Solving a Vessel Routing Problem with Inventory Constraints and Time Windows.” Ricerca Operativa 29 (91):
55–68.
Fodstad, M., K. T. Uggen, F. Rømo, A. Lium, and G. Stremersch. 2010. “Lngscheduler: A Rich Model for
Coordinating Vessel Routing, Inventories and Trade in the Liquefied Natural Gas Supply Chain.” The Journal
of Energy Markets 3 (4): 31–64. doi:10.21314/JEM.2010.049.
Furman, K. C., J. Song, G. R. Kocis, M. K. Mcdonald, and P. H. Warrick. 2011. “Feedstock Routing in the
Exxonmobil Downstream Sector.” Interfaces 41 (2): 149–163. doi:10.1287/inte.1100.0508.
Grønhaug, R., and M. Christiansen. 2009. “Supply Chain Optimization for the Liquefied Natural Gas Business.”
Innovations in Distribution Logistics 619: 195–218.
Meng, Q., and T. Wang. 2010. “A Chance Constrained Programming Model for Short-Term Liner Ship Fleet
Planning Problems.” Maritime Policy & Management 37 (4): 329–346. doi:10.1080/03088839.2010.486635.
Papageorgiou, D. J., G. L. Nemhauser, J. Sokol, M. Cheon, and A. B. Keha. 2014. “Mirplib – A Library of Maritime
Inventory Routing Problem Instances: Survey, Core Model and Benchmark Results.” European Journal of
Operational Research 235: 350–366. doi:10.1016/j.ejor.2013.12.013.
Perakis, A. N., and W. M. Bremer. 1992. “An Operational Tanker Scheduling Optimization System: Background,
Current Practice and Model Formulation.” Maritime Policy & Management 19 (3): 177–187. doi:10.1080/
751248659.
Persson, J. A., and M. Göthe-Lundgren. 2005. “Shipment Planning at Oil Refineries Using Column Generation and
Valid Inequalities.” European Journal of Operational Research 163 (3): 631–652. doi:10.1016/j.ejor.2004.02.008.
Rocha, R., I. E. Grossmann, and M. V. P. De Aragão. 2013. “Cascading Knapsack Inequalities: Reformulation of a
Crude Oil Distribution Problem.” Annals of Operations Research 203 (1): 217–234. doi:10.1007/s10479-011-
0857-8.
Song, J., and K. C. Furman. 2013. “A Maritime Inventory Routing Problem: Practical Approach.” Computers &
Operations Research 40 (3): 657–665. doi:10.1016/j.cor.2010.10.031.
UNCTAD-United Nations Conference on Trade and Development. 2013. “Review of maritime transport.” Accessed
June 9 2014. Unctad.Org/En/Publicationslibrary/Rmt2013_En.Pdf

Appendix 1
Agra, Andersson, et al., 2013 propose a reformulation of a discrete time model as Fixed Charge Network Flow
(FCNF) model, which they claim that it provides tighter linear relaxations when compared with the other discrete
time models found in literature. Here we describe the mathematical formulation of the FCNF model.

Sets
N set of all ports indexed by i and j;
T set of time periods indexed by t;
V set of vessels indexed by v;
NP set of loading ports indexed by i and j;
ND set of discharge ports indexed by i and j.

Parameters
CTijv sailing cost of vessel v between ports i and j;
CWv waiting cost of vessel v for each time period;
CPiv port cost of port i for vessel v;
o(v) position of vessel i in the beginning of planning horizon;
d(v) artificial end node for each vessel;
Tijv sailing time of vessel v between ports i and j;
Bit number of berths available in port i during time period t;
Qv maximum amount of product to be (un)loaded at one time period of ship v;
L0v inventory on board of vessel v in the beginning of the planning horizon;
16 G. S. D. S. DIZ ET AL.

Kv vessel capacity;
Dit demand rate in the discharge port i for each time period t;
Pit production rate in the load port i for each time period t;
SMXit Upper bound of inventory level in port i for each time period t;
SMNit Lower bound of inventory level in port i for each time period t;
S0i Inventory level at port i in the beginning of the planning horizon.

Variables
oivt where oivt ∈ {0,1}
equal to 1 if vessel v is operating (loading/unloading) in port i during time period t and 0 otherwise,
oAivt where oAivt ∈ {0,1}
indicates whether ship v starts to operate at port i in period t
oBivt where oBivt ∈ {0,1}
indicates whether ship v continue to operate at port i in period t
xijvt where xijvt ∈ {0,1}
equal to 1 if vessel v left port i to port j during time period t and 0 otherwise,
wivt where wivt ∈ {0,1}
equal to 1 if vessel v is waiting outside berth in port i during time period t and 0 otherwise,
f Xijvt that indicates the load on board ship v when traveling from port i to port j, leaving port i in period t,
f OAivt that indicates the load on board ship v when starting to operate at port i in period t and has not operated in
period t-1,
f OBivt that indicates the load on board ship v before continuing to operate at port i in period t after having
operated in time period t-1,
fWivt that indicates the load on board ship v while waiting during time period t at port i.
qivt where qivt ∈ R
quantity (un)load from/to vessel v in port i during time period t,
sit where sit ∈ R
inventory level at port i during time period t.

Routing constraints
X X
xoðvÞjvt ¼ 1; "v 2 V; (1)
j2N [fdðvÞg t2T
X X
xidðvÞvt ¼ 1; "v 2 V: (2)
i2N [foðvÞg t2T

Constraints (1) and (2) guarantee that every ship leaves from its artificial origin port and finishes the voyage at its
artificial destination port.

Sailing, waiting, and operating time constraints


X
xjiv;tTjiv þ wiv;t1 ¼ wivt þ oAivt ; "v 2 V; i 2 N; t 2 T; (3)
j2N [foðvÞg

X
oAivt1 þ oBivt1 ¼ oBivt þ xijvt ; "v 2 V; i 2 N; t 2 T; (4)
j2N [fdðvÞg

oAivt ; oBivt 2 f0; 1g; "v 2 V; i 2 N; t 2 T; (5)

oAivt þ oBivt ¼ oivt ; "v 2 V; i 2 N; t 2 T; (6)


X
oivt  Bit ; "i 2 N; t 2 T: (7)
v2V

Constraint (3) indicates the ship arrival at one port and (4) shows when the ship sailing from one port. Constraint
(5) indicates the variables oAivt and oBivt are binary. Constraint (6) provides the link between the old and the new
operating variables. Constrain (7) gives berth restriction at each node and consequently waiting time.
MARITIME POLICY & MANAGEMENT 17

Loading constraints
0  qivt  Qv oivt ; "v 2 V; i 2 N; t 2 T; (8)

P X
fjiv;tT jiv
þ fiv;t1
W
¼ fivt
W
þ fivt
OA
"v 2 V; i 2 N; t 2 T; (9)
j2N[foðvÞg
X
OA
fiv;t1 þ fiv;t1
OB
þ qiv;t1 ¼ fivt OB þ X
fijvt ;
j2N[fdðvÞg (10)
"v 2 V; i 2 NP[foðvÞg ; t 2 T;
X
OA
fiv;t1 þ fiv;t1
OB
 qiv;t1 ¼ fivt OB þ X
fijvt ;
j2N[fdðvÞg (11)
"v 2 V; i 2 ND[foðvÞg; t 2 T;
foXðvÞjvt ¼ L0v xoðvÞjvt ; "v 2 V; j 2 N [fdðvÞg; t 2 T: (12)

The quantity to be (un)loaded (qivt) must be less than the maximum amount of product to be (un)loaded of ship v
(constraint 8). The flow conservation constraints (9)–(12) ensure the load on board balance along every arc of the
structure. The variables upper bounds and non-negativity constraints are expressed in (13)–(17):

0  fijvt
X
 Kv xijvt ; "v 2 V; i 2 N [foðvÞg; j 2 N [fdðvÞg; t 2 T (13)

0  fivt
OA
 Kv oAivt ; "v 2 V; i 2 N; t 2 T; (14)

0  fivt
OB
 Kv oBivt ; "v 2 V; i 2 N; t 2 T; (15)

0  qivt  Qv oivt ; "v 2 V; i 2 N; t 2 T; (16)

0  fivt
W
 Kv wivt ; "v 2 V; i 2 N; t 2 T: (17)

Inventory control constraints


X
si;t1 þ qivt ¼ Dit þ sit ; "i 2 ND; t 2 T; (18)
v2V
X
si;t1 þ Pit ¼ qivt þ sit ; "i 2 NP; t 2 T; (19)
v2V

SMNit  sit  SMXit ; "i 2 N; t 2 T; (20)

si0 ¼ S0i ; "i 2 N; (21)

xijvt 2 f0; 1g; "v 2 V; i 2 N [foðvÞg; j 2 N [fdðvÞg; t 2 T; (22)

oivt ; wivt 2 f0; 1g; "v 2 V; i 2 N; t 2 T; (23)

sit ; qivt 2 R"v 2 V; i 2 N; t 2 T: (24)


The inventory level at each port is controlled during every time unit of the planning horizon at load and discharge
ports (constraints 18 and 19). Constraint (20) gives the operational range within inventory levels must be, and
constraint (21) gives initial inventory at each port. All binary variables are stated in constraints (22) and (23) and
constraint (24) gives the continuous variables.

Objective function
X X X X XXX XXX
min CTijv xijvt þ CPiv oivt þ v wivt :
CW (25)
v2V i2N [foðvÞg j2N [fdðvÞg t2T v2V i2N t2T v2V i2N t2T

The minimization function (25) contains transportation costs, operation costs, and waiting costs.
18 G. S. D. S. DIZ ET AL.

Valid inequalities and Branching strategy


To further strengthen the FCNF formulation, we have added some valid inequalities also proposed in Agra,
Andersson, et al., 2013. For the sake of clarity, we will show only the valid inequalities that we have implemented
to strengthen the FCNF model. The demonstration of how to get to these valid inequalities is explained in Agra,
Andersson, et al., 2013.

Knapsack inequalities
The idea here is to look at the flow in and out of a port over a pre-defined time interval T = [k,l] ⊆ T. One must define a
set of interval T v ⊆ T in which is a subset of the time periods in T in which ship v is assumed to operate at port i. Also
define T +v = {t ∈ Tv: t + 1 ∉ Tv} as the time periods in T followed immediately by the departure of ship v of port i and T

v = {t ∈ Tv: t—1 ∉ Tv} as the time periods in T in which ship v starts operating at i. Thus for an arbitrary Q > 0, the
following Chvátal–Gomory inequalities are valid for XFCNF. Constraint (26) is valid for every loading port.
! P
P P Kv B P P Kv P Qv Pit þS ik
Q oiv;tþ1 þ Q xijv;tþ1 þ Q oivt  t2T Q i;l1 "i 2 NP: (26)
v2V t2Tvþ j2N [fdðvÞg t2Tv t2TnTv

For discharging ports, one has to part the subset T into three disjoints sets R0v, R1v and R2v. R2v are the time periods
in T where the ship can start operating, R1v are the time periods in which the ship can continue to operate and R0v
are all others time periods in T. Also define T2v = {t ∈ R2v: t—1 ∉ R2v } as the first time periods in T that the ship
can start operate and T1v = {t ∈ R1v ∪ R2v: t—1 ∉ (R1v ∪ R2v)} as the first time period in T that the ship can either
start or continue to operate. The inequalities (27) are valid for every discharge port.
X X   X   X X   X  
Qv Kv A Kv kv
t2R0v
o ivt þ t2R1v Q
o j2N[foðvÞg t2R2v Q
x ijv;t-Tjiv þ t2Tv2 Q
wiv;t1
v2V Q ivt

   & P ’ (27)
X kv B t2T Dit  Si;l1 þ Sik
þ 1
t2Tv Q
o ivt  "i 2 ND;
Q

Mixed Integer rounding inequalities


A set of simple mixed rounding inequalities are obtained from the basic-mip set of the form: {(s,y) ∈ R1+ x Z1:
s + y ≥ b} and applied for every loading port.

s þ fy  b; (28)

where:
sik
s¼ ;
Q
0 1
X X Kv X X Kv X Qv
y¼ @ oB
þ xijv;tþ1 þ oivt A;
v t2T þ
Q iv;tþ1 j2N [fdðvÞg t2T Q t2TnT
Q
v v v

P
b¼ P þSi;l1
t2T it
Q;

f ¼ b  b:

Wagner–Whitin constant capacity lot-sizing reformulations


The Wagner–Whitin valid inequalities are derived from the relaxation of a Lot-Sizing Constant Capacity set
(LSCC).
For any discharge port i they are formulated as below:

X
t X
t

~si;k1 þ Q ~oiu  ~ iu "k 2 T; t 2 T; k  t and "i 2 ND;
D (28)
u¼k u¼k
MARITIME POLICY & MANAGEMENT 19

~sit  0; ~oit 2 f0; 1g"t 2 T and "i 2 ND; (29)

where: P P n o
~sit ¼ sit  SM
it ; ~
qit ¼ qivt ; ~oit ¼ oivt ; SM
it ¼ Si if t ¼ 0 and Sit ¼ max Sit ; Si;t1  Dit
0 M M
if t 2 T,
v2V v2V
 ¼ maxfQv : v 2 V g.
~ it ¼ Dit  SM þ SM and Q
D i;t1 it

For loading ports, the valid inequalities are described as:


X
t X
t

^si;k1 þ Q ~oiu  Piu "k 2 T; t 2 T; k  t and "i 2 NP; (30)
u¼k u¼k

Si  Sit  ^sit  Si  Sit ; "t 2 T and "i 2 NP; (31)

~oit 2 f0; 1g; (32)

where:

^sit ¼ Si  sit ; Si ¼ maxfS0i ; maxt2T Sit g:

The two-level lot-sizing structure presented in the FCNF formulation provides some new valid inequalities. For
discharge port these valid inequalities are:
X
t X
t

ei;k1 þ K ~oAiu  ~ iu ; "i 2 ND; "k 2 T; t 2 T and k  t;
D (33)
u¼k u¼k

eik  0; ~oAiu 2 f0; 1g;

where:
P OB P
eit ¼ ~fi;tþ1
OB
þ ~sit ; ~fitOB ¼ fivt  ¼ maxfKv : v 2 V g; ~oA ¼
K iu oAivt : ;
v2V v2V

A similar valid inequality is developed for loading ports:


P
t P
t
 ~oAiu 
i;k1 þ K
eLP Piu ; "i 2 NP; "k 2 T; t 2 T and k  t; (34)
u¼k u¼k

where:
X
f OB þ ^sit ; f OB ¼ Kv  oB  f OB :
it ¼
eLP iv;tþ1 ivt ivt ivt
v2V

Inequalities for lot-sizing with start-up relaxations


These inequalities are related with the idea of start-up costs, that is, a cost associated with the first period of an
interval of sets-up (Agra, Andersson, et al., 2013). In the FCNF problem, oAivt can be interpreted as a start-up
variable and can be used to derive valid inequalities. It is important to highlight that these inequalities are stated
considering only one berth per port (Bit ¼ 1), which means that ~oAit ; ~oit 2 f0; 1g.

~oAit  ~oit "t 2 T and "i 2 ND; (35)

~oAit  ~oit  ~oi;t1 "t 2 T and "i 2 ND; (36)

X
t
~ it "t 2 T and "i 2 ND;
~oit  O (37)
u¼1

where:
X X X
t
~oAit ¼ oAivt ; ~oit ¼ ~ it ¼
oivt ; andO D 
~ it =Q:
v2V v2V u¼1
20 G. S. D. S. DIZ ET AL.

Given a time interval [k,l]:


X
pi X
σ i;ki þ ð~oi;kþj1 þ ~oAi;kþj Þ  pi "i 2 ND; (38)
j¼1 jj¼kþj

where:

X
l X
t
pi ¼ ~ it  O
δit ; δit ¼ O ~ i;t1 and σ it ¼ ~ it  0 and σ i0 ¼ 0:
~oiu  O
t¼k u¼1

The entire formulation presented above is discussed in more detailed in Agra, Andersson, et al., 2013.

You might also like