Computational Logistics 9th International Conference ICCL 2018 Vietri Sul Mare Italy October 1 3 2018 Proceedings Raffaele Cerulli
Computational Logistics 9th International Conference ICCL 2018 Vietri Sul Mare Italy October 1 3 2018 Proceedings Raffaele Cerulli
Computational Logistics 9th International Conference ICCL 2018 Vietri Sul Mare Italy October 1 3 2018 Proceedings Raffaele Cerulli
com
https://textbookfull.com/product/computational-
logistics-9th-international-conference-
iccl-2018-vietri-sul-mare-italy-
october-1-3-2018-proceedings-raffaele-cerulli/
https://textbookfull.com/product/saint-hot-shots-4-1st-edition-tory-
baker-baker-tory/
textbookfull.com
Ways of Being in Teaching Conversations and Reflections
1st Edition Sean Wiebe
https://textbookfull.com/product/ways-of-being-in-teaching-
conversations-and-reflections-1st-edition-sean-wiebe/
textbookfull.com
https://textbookfull.com/product/the-eastern-arctic-seas-
encyclopedia-1st-edition-igor-s-zonn/
textbookfull.com
https://textbookfull.com/product/werewolves-innocence-complete-bundle-
public-humiliation-shifter-bdsm-fae-marillyn/
textbookfull.com
https://textbookfull.com/product/responsible-business-in-a-changing-
world-new-management-approaches-for-sustainable-development-belen-
diaz-diaz/
textbookfull.com
Environmental Attitudes through Time 1st Edition R. J.
Berry
https://textbookfull.com/product/environmental-attitudes-through-
time-1st-edition-r-j-berry/
textbookfull.com
Raffaele Cerulli · Andrea Raiconi
Stefan Voß (Eds.)
LNCS 11184
Computational
Logistics
9th International Conference, ICCL 2018
Vietri sul Mare, Italy, October 1–3, 2018
Proceedings
123
Lecture Notes in Computer Science 11184
Commenced Publication in 1973
Founding and Former Series Editors:
Gerhard Goos, Juris Hartmanis, and Jan van Leeuwen
Editorial Board
David Hutchison
Lancaster University, Lancaster, UK
Takeo Kanade
Carnegie Mellon University, Pittsburgh, PA, USA
Josef Kittler
University of Surrey, Guildford, UK
Jon M. Kleinberg
Cornell University, Ithaca, NY, USA
Friedemann Mattern
ETH Zurich, Zurich, Switzerland
John C. Mitchell
Stanford University, Stanford, CA, USA
Moni Naor
Weizmann Institute of Science, Rehovot, Israel
C. Pandu Rangan
Indian Institute of Technology Madras, Chennai, India
Bernhard Steffen
TU Dortmund University, Dortmund, Germany
Demetri Terzopoulos
University of California, Los Angeles, CA, USA
Doug Tygar
University of California, Berkeley, CA, USA
Gerhard Weikum
Max Planck Institute for Informatics, Saarbrücken, Germany
More information about this series at http://www.springer.com/series/7407
Raffaele Cerulli Andrea Raiconi
•
Computational
Logistics
9th International Conference, ICCL 2018
Vietri sul Mare, Italy, October 1–3, 2018
Proceedings
123
Editors
Raffaele Cerulli Stefan Voß
University of Salerno Institute of Information Systems
Fisciano, Italy University of Hamburg
Hamburg, Germany
Andrea Raiconi
University of Salerno
Fisciano, Italy
This Springer imprint is published by the registered company Springer Nature Switzerland AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
Preface
While digital transformation and digital innovation are spreading the word and pene-
trating almost every business, logistics is still omnipresent. If we do not move freight or
people, we move bits and bytes, or both or even everything together. We claim that the
data is there, the technology is there, and we need to get our ideas and solution
concepts into running systems. Whether we are emphasizing new buzzwords like
mobility-as-a-service or synchromodality, or whether we are still using existing phrases
like stowage planning or service network design, computational logistics is here to
support these processes and make things better and better. We are solving classical
combinatorial optimization problems related to, for instance, pickup and delivery, we
are thinking about already established concepts of recent years such as e-mobility, and
we are considering new concepts like autonomous vessels. But in all cases we
appreciate the connection to computational tools for solving complex problems in
logistics and supply chain management as well as public transport.
The International Conference on Computational Logistics (ICCL) is a forum where
recent advances on the topic are presented and discussed. This volume offers a
selection of 32 peer-reviewed papers out of well over 70 contributions submitted to the
9th International Conference on Computational Logistics (ICCL 2018), held in Vietri
sul Mare, Italy, during October 1–3, 2018. The papers show various directions of
importance in computational logistics, classified into five topic areas reflecting the
interest of researchers and practitioners in this field. The papers in this volume are
grouped according to the following parts:
1. Maritime Shipping and Routing:
As a major mode in freight transportation we see considerable work in the area of
maritime shipping. The ICCL has been strong in addressing issues of maritime
shipping as it is also observed here. The papers in this area address various prob-
lems arising, among others, in maritime inventory routing, fleet deployment, as well
as the upcoming topic of autonomous shipping. Moreover, inland waterways are
addressed, too.
2. Container Handling and Container Terminals:
Moving from the waterside into ports, in this part we have papers focussing on the
relocation of containers within the yard area of container terminals, stowage
planning, vessel capacity, as well as related key performance indicators.
3. Vehicle Routing and Multi-modal Transportation:
Being a classical area of research, vehicle routing problems still need quite a bit of
attention, especially when they become richer and include more and more practical
constraints. Additionally, concomitant factors such as those of pricing services, as
well as sensing vehicle coverage, etc., have received more attention recently and are
considered here.
VI Preface
Program Committee
Panagiotis Angeloudis Imperial College London, UK
Claudia Archetti University of Brescia, Italy
Buyang Cao Tongji University, China
Francesco Carrabs University of Salerno, Italy
Carmine Cerrone University of Molise, Italy
Raffaele Cerulli University of Salerno, Italy
Stefano Coniglio University of Southampton, UK
Francesco Corman ETH Zurich, Switzerland
Ciriaco D’Ambrosio University of Salerno, Italy
René De Koster Erasmus University Rotterdam, The Netherlands
Karl Doerner University of Vienna, Austria
Jan Fabian Ehmke Otto-von-Guericke University Magdeburg, Germany
Kjetil Fagerholt Norwegian University of Science and Technology,
Norway
Yingjie Fan Erasmus University, Rotterdam, The Netherlands
Paola Festa University of Naples Federico II, Italy
Monica Gentili University of Louisville, USA
Rosa Guadalupe Gonzalez Universidad de Los Andes, Chile
Ramirez
Luis Gouveia University of Lisbon, Portugal
Peter Greistorfer Karl-Franzens-Universität Graz, Austria
Irina Gribkovskaia Molde University College, Norway
Hans-Otto Günther TU Berlin, Germany
Richard F. Hartl University of Vienna, Austria
Leonard Heilig University of Hamburg, Germany
Sin C. Ho Chinese University of Hong Kong, SAR China
Rune Møller Jensen Information Technology University of Copenhagen,
Denmark
Herbert Kopfer University of Bremen, Germany
Eduardo Lalla-Ruiz University of Hamburg, Germany
Jasmine Siu Lee Lam Nanyang Technological University, China
Gilbert Laporte HEC Montréal and GERAD, Canada
Janny Leung Chinese University of Hong Kong, SAR China
Ridha Mahjoub LAMSADE, University Paris-Dauphine, France
Vittorio Maniezzo University of Bologna, Italy
Rudy Negenborn Delft University of Technology, The Netherlands
Dario Pacino Technical University of Denmark, Denmark
Julia Pahl University of Southern Denmark, Denmark
VIII Organization
Additional Reviewers
The Cost of Continuity in the Collaborative Pickup and Delivery Problem . . . 239
Margaretha Gansterer, Richard F. Hartl, and Sarah Wieser
Exact Methods and Heuristics for the Liner Shipping Crew Scheduling
Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 363
Valerio Maria Sereno, Line Blander Reinhardt, and Stefan Guericke
1 Introduction
The Maritime Inventory Routing Problem (MIRP) is a combinatorial problem
where one has to manage the voyage of vessels and the inventory of ports along
a finite planning horizon. This problem consists of routing and scheduling a het-
erogeneous fleet of vessels for transporting one or more products from production
to consumption ports. Each port has a storage capacity and a production or con-
sumption rate that may vary along the planning horizon. A vessel must arrive at
a production port to load a certain quantity of product before the port becomes
full, and similarly, it must deliver the product to a consumption port before
inventory becomes empty. Vessels can differ between them by capacity, operat-
ing costs, and traveling time between each port. The objective usually aims to
minimize the traveling and operational costs. Most of the MIRP discussed in
the literature are particular variations based on real scenarios, where different
assumptions and side constraints are considered.
As the nature of the problem is very complex, and maritime transporta-
tion involves substantial costs, it is crucial to use optimization techniques for
c Springer Nature Switzerland AG 2018
R. Cerulli et al. (Eds.): ICCL 2018, LNCS 11184, pp. 3–16, 2018.
https://doi.org/10.1007/978-3-030-00898-7_1
4 M. W. Friske and L. S. Buriol
proposed. The approach used was able to obtain good solutions, including two
new best-known values. However, for the hard (large) instances, the proposed
algorithm cannot achieve feasible solutions.
This work aims to extend the work of [7], applying the Relax-and-Fix frame-
work for solving the FCNF model. In this case, we use the problem described
in [12], modeling it as the FCNF similar to [2]. The R&F is a matheuristic
that consists in dividing the problem into subproblems, solving them iteratively
by relaxing and fixing subsets of variables to obtain a feasible solution for the
problem. For more explanation about R&F, we recommend the book of [13]. We
also implemented valid inequalities based on knapsack sets, and MIP-based local
searches were used for improving the solution quality. The objective of this work
is to verify if a relatively simple algorithmic approach such as R&F can obtain
good solutions for a hard problem, without using problem-specific algorithms,
as proposed by [11].
The remainder of this work is organized as follows. Section 2 describes the
problem and presents the FCNF formulation with the valid inequalities. Section 3
describes the R&F and MIP-based local search procedures. Computational
experiments are presented in Sect. 5. Finally, Sect. 6 presents the conclusions
and future works.
2 Problem Formulation
In this section, we present the FCNF model for the MIRP, that is based on
the model described in [12]. This model considers a deep-sea MIRP, where the
traveling times are much larger than the operating times. Each vessel must visit
different ports such that production ports do not reach their maximum capacity,
and discharge ports do not achieve an empty inventory. The discharging ports
pay different revenue values according to the quantity discharged by vessels.
In some cases, it is not possible to supply loading and discharging demands
due to the limited fleet size and the large traveling times between ports. In this
case, it is possible to consider the inventory as soft constraints, where the excess
or lack of the product can be sold to or bought from simplified spot markets.
However, the quantity sold to or bought from each port is limited.
In the FCNF, the loading of vessels is modeled as a commodity that flows
along the nodes, represented by port-times. Let V be the set of vessels, J the
set of ports, and T the set of time periods, with T = |T |. Ports are split into
subsets J P for production or loading ports, and J C for consuming or discharging
ports. Ports are grouped in production regions RP and discharging regions RC .
A port-time (i, t), i ∈ J , t ∈ T represents a possible operation of a vessel in
port i at time t. Source node o(v) represents the starting point of vessel v, and
sink node d(v) is the end of vessel route. Binary variable xijvt defines if a vessel v
travels from port i to port j, departing at time period t. Parameter Tijv denotes
the traveling time between port i and j by vessel v. When vessel v arrives at
v
specific node (i, t), it can wait for one time-period (binary variable wit is set to 1),
A
or it can start to operate (binary variable oivt is set to 1). When vessel v starts
6 M. W. Friske and L. S. Buriol
operating, continuous variable fitv represents the amount discharged (or loaded)
at node (i, t). The consumption rate of port i in each time period t is denoted
by Dit , and sit is the inventory of port i in time t. Parameter s0i corresponds
to the initial inventory of port i. Continuous variable αit represents the amount
of product bought from the simplified spot market. After started to operate, a
vessel can continue to operate in the same port, using variable oB ivt = 1, or it
can leave the port, traveling to another port or the sink node using a traveling
arc xijvt .
Figure 1 illustrates the FCNF model for a discharging port i ∈ J P and one
vessel v ∈ V, with T = 4. A possible route of the vessel is highlighted.
In Fig. 1 each port-time (i, t) is divided into three layers. The top layer coor-
dinates the traveling of vessels between ports, including dummy source and sink
nodes. The middle layer coordinates the operation of vessel at the port, and the
bottom layer represents the transfer of product between vessel v and port i.
In the FCNF model, when a vessel arrives at a port, it must operate (dis-
charging or loading) before departing to another destination. Also, after ending
the operation, it is not allowed to wait at the same port. These constraints are
not considered in the original formulation of [12]. However, they are implicit and
useful for a good solution in a real scenario.
Applying a R&F Approach to a FCNF Model of a MIRP 7
Objective function (1a) maximizes the revenue Rit of the unloaded product
at discharging ports, subtracting arc costs Cijv used by each vessel. The third
term is an additional value that induces vessels to operate as soon and as few
times as possible. Variable ovit is equal to one when vessel v operates at port i
in time t. The penalization value Pit for using spot markets is accounted in the
last term of the equation.
Constraints (1b) fix the source arc of each vessel v. It assumes that vessel
v travels from the source node o(v) to its initial port jv0 , departing from time
period 0. The traveling time between the source node and the initial port is the
first time t0v in which vessel v becomes available. Constraints (1c) impose that
all vessels end its route, reaching the sink node d(v). Constraints (1d) and (1e)
are the flow balance of each vessel along the nodes. Constraints (1f) define that
if a vessel is operating, it started to operate in the current time period, or it
continues to operate from the previous time period.
Constraints (1g)-(1h) represent inventory balance of each vessel,
where Δi = 1 if i ∈ J P , and Δi = −1 if i ∈ J C . Variable fijvt X
is the load
W
on board of vessel v before traveling from port i to port j in time period t, fivt
OA
is the load on board of vessel v before waiting at port i in time t, fivt is the
OB
load on board of vessel v before starting to operate at port i in time t, and fivt
is the load on board of vessel v before continuing to operate at port i in time t.
Constraints (1i) define the initial inventory of each vessel, where parameter L0v
corresponds to the initial inventory.
Side constraints (1j) and (1k) impose that each vessel must depart from a
loading region to a discharging region at full capacity and must depart from a
discharging region to a loading region empty. Constraints (1l) limit to Bi (num-
ber of berths) the number of vessels that can operate simultaneously at a node.
The inventory balance at ports is defined by (1m). Constraints (1n) and (1o)
limit the daily amount and the cumulative amount of product that can be bought
from or sold to the simplified spot market, respectively. Constraints (1p) impose
the inventory limit of each port to be between the lower limit Simin and upper
limit Simax , and (1q) define the initial inventory for each port. Constraints (1r)–
(1u) limit the flow of vessel v to the capacity Qv along the arcs, while (1v) impose
a minimum Fimin and maximum Fimax amount of product that can be loaded or
discharged by vessel v when operating in each port i and time period t. Finally,
(1w) and (1x) define the scope of binary variables.
Inequalities (2) impose a minimum number of vessel departures, while (3) impose
a minimum number of loadings at port i.
For discharging ports, we can consider three inequalities sets:
⎛ ⎞
⎝ xjiv,t−Tjiv + wiv,l−1 + oB
iv,l−1
⎠≥
v∈V j∈J ∪{o(v)} t∈T
t∈T Dit − Simax + Simin
(4)
K
D
t∈T it − Simax + Simin
oA B
ivt + oiv,l−1 ≥ (5)
v∈V t∈T
K
Dit − Simax + Simin
oivt ≥ t∈T
(6)
Fimax
v∈V t∈T
Time horizon
First
iteration Interval 1 Interval 2 Interval 3 Interval 4 Interval 5
Fixed
block Integer block Relaxed block End block
Second
iteration Interval 1 Interval 2 Interval 3 Interval 4 Interval 5
Last
iteration Interval 1 Interval 2 Interval 3 Interval 4 Interval 5
and a part of the model that was omitted is now considered in the model as
relaxed block. The problem is then solved again by the MIP solver. The algo-
rithm continues iterating until all intervals have been removed from the end-block
and integrality constraints are reintroduced to the variables of all intervals. At
this point, a solution to the original problem is obtained.
In the Relax-and-Fix strategy, solving each interval up to optimality does not
necessarily leads to an optimal solution for the original problem. In this case, we
use MIP relative GAP and time limit as stopping criteria in each iteration, as
suggested in [17], for accelerating the search. Initially, the MIP relative GAP is
set to a positive value, which is linearly decreased along the iterations such that
in the last iteration, the MIP relative GAP is set to 0.1%.
According to Fig. 2, there is an overlap between the blocks at each iteration
This strategy is used to avoid infeasible solutions after some part of the model
was fixed and integrality constraints are re-introduced to relaxed binary variables
[13]. However, for the MIRP even using the overlap, port-time inventory bounds
can be violated. It occurs when no vessel can reach a port at specific times due
to the previously fixed routing decisions and the spot markets variables are not
sufficient to avoid lack or surplus of inventory. To handle this issue, we introduce
nonnegative auxiliary variables βjt , θjt , j ∈ J , t ∈ T . They transform the port-
time inventory capacity in soft constraints, where the variables βjt , θjt work as
an unlimited spot market either for buying and selling product amounts, being
highly penalized in the objective function. Equation (1m) is reformulated as
follows:
sit = si,t−1 + Δj (Dit − fitv − αit − βit + θit ), ∀i ∈ J , t ∈ T (7)
v∈V
Applying a R&F Approach to a FCNF Model of a MIRP 11
Note that the auxiliary variables are used to prevent the solver from stopping
prematurely, but if some of these variables are positive at the end of R&F, the
solution for the original problem remains infeasible.
4 Improvement Phase
After the R&F algorithm obtained a solution, an improvement phase starts. It
is used for improving the quality of solution and in some cases for removing
infeasibilities due to the use of auxiliary variables βjt and θjt .
In the proposed improvement phase, we use three MIP-based local search
strategies. MIP-based local search is a technique that combines heuristic and
mathematical programming, being used in several works, including MIRPs [11,
15]. It consists of first fixing all integer variables with the current solution values.
Systematically a subset of these variables is unfixed, and a solver is used to
solve the subproblem. Then, the variables are fixed with the newly obtained
values, and a new subset is selected to be optimized. All continuous variables
are kept free to be optimized in all iterations. The definition of each subset
defines different strategies for MIP-based local search. The proposed strategies
are:
optimized. Then, a next vessel with the same time interval is optimized. The
algorithm iterates between all time intervals and all vessels, m|V| steps in a
complete iteration. The search stops when no improvement is achieved in one
complete iteration or if the time limit is reached.
5 Computational Results
This section presents the results obtained by solving the Maritime Inventory
Routing Problem with the solution approach described in Sects. 3 and 4. As
in [11], we solved the model as a minimization problem, turning negative the
objective function (1a) for comparison purposes. The algorithms were imple-
mented using CPLEX 12.5 C++ API and compiled with the optimization param-
eter −O3. Experiments were carried out on an Intel Core i7-3632QM computer
running at 2.2 GHz on a single core, with 8 GB RAM.
For testing the algorithms, we use the “Group 1” instances available in the
MIRPLIB [1]. The instances name present their characteristics. For example,
instance “LR2 22 DR2 22 VC3 V10a” means that there exists 2 loading regions
(LR), and in each region there is two loading ports, two discharging regions
(DR), each of them with two ports, three vessel classes (VC), and a total of ten
available vessels (V), at least one for each vessel class. The letter at the end of
the name is used for differentiating instances with the same size. Each instance
was tested with time horizons of 45 and 60 time-periods, where each time-period
corresponds to one day.
5.2 Parametrization
Some parameters should be defined for the proposed algorithm. Although the
tested instances have different sizes and there are differences in their character-
istics, we defined a more general parametrization to be tested. Table 1 presents
the parameters and the tested values.
For the MIP-based local search procedures, parameter “Total time limit (s)”
is divided equally for each strategy. If the first strategy ends before the time
limit, the remaining time is equally divided for the other two strategies and so
on. The improvement phase uses the default value of the CPLEX solver for the
MIP relative GAP.
The valid inequalities were added to the model a priori for all combinations
of T , such that T includes either the first time period or the last time period,
i.e. T = {1, . . . , t}, or T = {t, . . . , T}, t ∈ T .
Applying a R&F Approach to a FCNF Model of a MIRP 13
Parameter Value
Relax-and-Fix Number of time periods per 15
interval (T /n)
Size of model (in time periods) 45 Instances prefix “LR1 1”
at the first iteration 30 Instances prefix “LR1 2” or “LR2”
Overlap (%) 50
Time limit (s) per iteration 600 Instances prefix “LR1” or
“LR2 11”
1200 Instances prefix “LR2 22”
Initial MIP GAP (%) 5
Local Search Time limit (s) per iteration 120
Total time limit (s) 7200
Number of time intervals m 3
1
http://www.cpubenchmark.net/.
14 M. W. Friske and L. S. Buriol
Decir, por tanto, que del fuego proceden todos los seres y que no
existe cosa alguna independiente de ese origen, es caer en un delirio
que nuestra inteligencia rechaza como contrario á las pruebas que
nos da el testimonio de nuestros sentidos; y admitir como primordial
materia el fuego porque se cree conocerlo perfectamente, y recusar
al mismo tiempo la existencia de otros seres que con suficiente
claridad se muestran ante nosotros, me parece inconsecuencia y
error grave. Para conocer la verdad, ¿qué mejor guía que nuestra
razón auxiliada por los sentidos, los cuales nos hacen distinguir lo
falso de lo verdadero y nos dan principio de certeza? Además, ¿por
qué hemos de negar la existencia de todas las cosas y admitir
solamente la del fuego, ó negar que éste exista y dar como reales
solamente los demás cuerpos? Parece que afirmar cualquiera de
estos dos extremos es incurrir en igual demencia.
712. Quapropter qui materiem rerum esse putarunt...
Así, los que han afirmado que el conjunto del Universo tiene su
fundamento en el fuego y los que han entendido que en el aire se
encuentra el origen de todos los seres, lo mismo que aquellos otros
que han sostenido que en el agua reside el principio creador, ó bien
que la tierra puede germinar todas las cosas ó determinarse en las
infinitas diferenciaciones corpóreas existentes, según mi parecer,
todos han caído en grave error, no menos que aquellos otros que
suponen combinaciones dualísticas formadas con los elementos de
todas las cosas, y al fuego juntan el aire, y á la tierra el agua; y
también, por último, los que entienden que de esos cuatro
materiales, fuego, tierra, aire y agua se han podido producir todos
los seres.
Sin duda notarás que en muchos de estos versos míos hay varios
elementos ó letras simples comunes á numerosas palabras, y, sin
embargo, ni los versos ni las palabras tienen igual significado y
sonido igual: varía el valor de las letras sólo al cambiar éstas de
orden. Y como los elementos primordiales de las cosas en mayor
número son que las letras, pueden producir mayor suma de seres
diferentes.