Information Technology Reference
In-Depth Information
3
Related Work
There are a few papers that propose to apply multi-agent systems in the domain of con-
tainer terminal management and optimisation. Thurston and Hu [1] proposes to use a
multi-agent system to automate container terminal operations. They focus on the load-
ing process only. Like us, they have agents for each of the machines (quay cranes,
straddle carriers). While their work is promising, the paper is early work: it outlines
the approach and reports on an early Java prototype. No experimental evaluation is re-
ported, and (from personal contact with the author) it appears that no subsequent work
has been done. Rebollo
et al.
[2] also propose to automate container terminal oper-
ations using a multi-agent system. Again, the paper is abstract: it provides a system
architecture, but does not provide details about how the individual agents would op-
erate. Implementation appears to have been in progress, but we have not found any
subsequent papers describing the resulting implementation, or results from evaluation.
Note that such work, which attempts to address all of the problems of a container ter-
minal, is quite ambitious, and is in danger of needing to make simplifying assumptions
that render it inapplicable to real ports. Our approach is firstly to not attempt to auto-
mate a terminal, but to provide decision support; and secondly, to deal with parts of the
problem separately, while trying to avoid oversimplifying the problem.
Other work that seeks to apply agents to container terminals has been more modest
in scope, seeking either to tackle part of the problem only, or to simulate but not to
control. An example of the latter is Henesey
et al.
[3], which describes a simulation
tool (“SimPort”). Unlike the earlier described work, they do not aim to automate the
operation of a container terminal, but instead to provide a tool that can be used to
analyse the performance of (static) policies. This analysis can then be used to select
static policies to implement.
In addition to agent-based approaches, there have been other (non-agent-based) ap-
proaches that aim to tackle various aspects of the management and optimisation of
container terminals (see [4,5] for recent surveys). A common limitation of such work,
which is often based on operations research techniques, is that it computes solutions
up-front, but does not address well the dynamic nature of the problem.
The work of Chen
et al.
[6] is in some ways quite close to our work on optimis-
ing container moves (Section 5), and, indeed, their formalisation was the starting point
for our work. However, they make a number of assumptions that are unreasonable in
practice, and which we have relaxed. In particular, they assume that the processing of
loading and unloading is handled separately. Additionally, they assume a three-stage
process and do not provide for “buffering” where, for example, a Quay Crane can un-
load a second or third container even though the first container has not yet been taken
to the yard.
Considering now the yard allocation problem (which we cover in Section 6), there
is a range of work that tackles this problem. Zhang
et al.
[7] consider the problem of
allocating inbound and outbound containers to blocks in the yard, within the context of
an overall planning hierarchy for a container terminal. A two-stage process first deter-
mines the number of containers that are to be assigned to each block to evenly balance
the overall workload. This is followed by the allocation of individual containers to spe-
cific blocks, without considering the specific placement of containers within each block.
