SIMULATION FOR POLICY EVALUATION, PLANNING AND DECISION SUPPORT
IN AN INTERMODAL CONTAINER TERMINAL
Monaldo Mastrolilli, Nicoletta Fornara, Luca Maria Gambardella, Andrea E. Rizzoli and Marco Zaffalon
IDSIA - Istituto Dalle Molle di Studi sull’Intelligenza Artificiale
Corso Elvezia 36, 6900 Lugano Switzerland
E-mail: {monaldo,nicky,luca,andrea,zaffalon}@idsia.ch
ABSTRACT
Different uses of a simulation tool in an intermodal container
terminal are presented. Initially, the simulation model of the
container terminal of La Spezia (LSCT) in Italy is described.
Then it is shown its calibration and validation. The resulting
model is used as a tool to validate alternative management
policies, such as resource allocation and ship loading and
unloading policies. The terminal management can use such
policies for mid-term planning. Finally, the design of a decision
support system to assist the management in real time decision
making is discussed.
INTRODUCTION
The advent of management information services and data
processing greatly improved the ability of terminal managers to
control the whole process, but still raw data has to be analyzed
and treated to provide some insight on the performance of
terminal operations. Simulation models and Operations
Research techniques have proven to be a reliable and convenient
tool to support the decision-makers in the daily operations in
many cases (Hayuth et al. 1994; Blümel 1997; Bruzzone and
Signorile 1997).
A simulation model of a terminal can provide a valuable tool for
the management, especially to evaluate the performance of new
policies (policy evaluation), to assess the effect of the
implementation of these policies on the terminal state
(planning), and to take operational decisions (real time decision
support).
In this paper we explore these different uses for the simulation
model, in particular with respect to resource allocation (RA) and
loading and unloading (L/U) policies.
The first section is devoted to the description of the model
structure; the following introduces its calibration and validation.
In the third section it is discussed the integration of the RA
module and the L/U scheduling module with the simulator, for
the purpose of evaluating the computer-generated policies. The
latter, computed on the basis of Operations Research techniques
(Gambardella et al. 1996; Gambardella et al. 1998; Zaffalon and
Gambardella 1998; Mastrolilli and Gambardella 1998) are
embedded in the simulation model, which has been calibrated in
the previous step. Their performance can be compared (starting
from the same initial conditions and under the same input
regime) with the performance obtained by the terminal
management (historical data are used to feed the simulator).
The last section discusses the use of the simulation model as a
mechanism to evaluate medium and long term planning
decisions such as the space allocation policies. Furthermore it
discusses the use of the simulation tool as a decision support
tool if real-time data are available.
THE SIMULATION MODULE
The architecture of the simulation tool is based on the partition
of simulation objects between simulation agents and simulation
components. In an intermodal terminal the two flows of
information and of containers are present; the simulation agents
use the flow of information to make decisions on how to direct
the container flow.
The design of the present simulation tool is based on the object-
oriented analysis and design paradigm (Booch 1994).
Simulation agents and components are modeled as objects
which store and exchange information on terminal inputs, states
and outputs and which perform actions. The whole simulation is
the result of the interaction of such single agents, each one
endowed with local knowledge on its actions in response to
other agents behavior (Zeigler 1984; Zeigler 1990) and to
external events (like, trucks and trains arriving at terminal gate;
ships arriving at terminal pier, etc.).
Truck
Pool
QC Op Queue
YC Op Queue
Yard
Region
Ship
QC
YC
empty Shuttle Trucks
Trucks/ Trains/ Shuttle Trucks
ShuttleTrucks
Empty Carrier
Full Carrier
Trucks/ Trains/ Shuttle Trucks
ShuttleTrucks
Figure 1. A representation of the flow of containers.
Figure 1 reports an example of the container flow in the terminal
limited to a quay crane (QC for short; it is a crane used for
loading/unloading a ship) and a yard crane (YC, that is a crane
that moves containers for a given yard area). In the real terminal
there are seven quay cranes, ten yard cranes and ten straddle
carriers (the latter being alternative means, as compared to yard
cranes, to move containers in the yard). Ships, trains and trucks
entering the terminal have a loading list (the containers to
import) and an unloading list (the containers to export). These
lists are used by the yard and ship planners. The ship planner
organizes the L/U operations of a ship. This requires a complex
set of tasks to be considered, like allocating the work shifts for
the QCs; computing the L/U policies in order to respect a ship's
stability constraints; cooperating with the yard planner to assign

yard destinations to the containers to be unloaded and to require
the containers to be loaded; put the quay cranes to work;
supervise loading and unloading operations; collect statistics
and evaluate performance.
On the other hand, the yard planner uses the lists of import and
export containers (produced by the interaction with the ship
planner) to build the schedule solving the job-shop problem
associated with yard crane operations. This also implies a proper
organization of container allocations on the yard in order to
maximize yard crane performance, avoid crane deadlocks and
minimize the time to access containers during storage and
retrieval.
Besides this (high level) management performed by the ship and
the yard planners, there are the local management decisions
taken by "less intelligent" simulation agents such as cranes and
shuttle trucks (every quay cranes has 3 shuttle trucks that bring
containers from the QC to the yard and vice versa) that acts on
the basis of a queue of operations to be performed (where an
operation is a container movement). Figure 2 reports a typical
screen-shot of the terminal during a simulation. A ship is
moored on the west pier (north is to the left of the picture) and it
is being unloaded by two quay cranes QC1 and QC2 (only QC1
is active, though). Containers are to be positioned on yard areas
CA, CB, and CC. On these yard areas the yard cranes YC from 1
to 9 are working.
CALIBRATION AND VALIDATION
Once the simulation model of the terminal is available, it must
be checked in order to verify its capability of reproducing the
real terminal behavior. This is accomplished in two steps,
namely calibration and validation. In general, calibration means
tuning the simulation model parameters in order to match as
close as possible the simulation outputs with the data measured
in the real terminal over a given time interval. A proper
validation phase ensures that the result of the calibration is such
that the simulator reproduces the reality under different
conditions (hence, not only when the data from the calibration
set are used). The validation step is carried out on a validation
set of data. On such a set, the simulator is expected to act in a
way that the real terminal is properly emulated. Of course no
tuning is allowed at this time.
Notice that in the calibration and validation phase, the
simulation module evolves using the same inputs and policies
used by the terminal management over the calibration and
validation periods. For this reason, the resource allocation and
the L/U policies are the ones adopted by terminal managers
during the periods when calibration and validation data were
collected.
In the sequel, we give a short example of validation and
calibration of the simulator. All the data used in the present
paper are taken from the database of the LSCT. The focus is on
a period of two weeks, from 5/11/1998 to 5/24/98. The data
describe the activity of the La Spezia container terminal in great
detail, in such a way that every container movement can be
found. In particular, the database reports the resources (yard
cranes, quay cranes, and straddle carriers) that move the
container, the time of the operation and the origin and
destination of the movement, described by three coordinates
each (bay, row and tier both on the ship and on the yard).
Furthermore, the database tracks the activity of every transport
means that enters and leaves the terminal (ships, trucks, trains).
From this wealth of data, some working shifts for calibration
and some other for validation were extracted. In particular, it is
considered the period that starts at 1 am on the 5/11/1998 and
ends after 10 shifts of 6 hours each (the advantage of such a
choice is that no ships are at the terminal at the selected starting
time, and the system is therefore in a natural initial empty state;
finding the initial empty state is often one of the major problems
for the simulation practitioner).
The L/U process of each quay crane is examined. For this
purpose, the number of moved containers by each quay crane in
Figure 2. A snapshot of the simulation tool interface.