Information Technology Reference
In-Depth Information
switch between suppliers and components when one supplier fails to supply them. This
new strategy, Multi-Attempt Massive Strategy with Prediction of Gaps with Supplier
Switching (MAMS-PG-SS), combines the prediction of gaps with a strategy that looks
for substitutable components. It determines the set of components that can be substi-
tuted for every component that the agent can order. For instance, if Socrates cannot
obtain memory 1GB from supplier Queenmax, then it looks for substitutable compo-
nents in the following order: memory 1 GB (MEC), memory 2 GB (Queenmax) and
finally memory 2 (MEC). This list indicates the order in which the agent will try to sat-
isfy the number of components needed. The agent keeps track of the number of attempts
to every supplier and if it reaches a threshold value the agent switches supplier or in the
appropriate case supplier and type of component. This process is carried out until the
agent satisfies the quantity needed of every component. The strategy works as expected
when a supplier cannot provide the specific component, and the agent has to look for a
replacement. Since it works by substituting the supplier or the type of component, the
number of PCs of each model assembled in a TAC day will be different. In some cases,
the agent will be left with a number of unused components in its inventory. This occurs
when the agent has ordered a greater number of CPUs than the number of motherboards
of one specific type (Pintel or IMD) in its attempt to find substitutable components and
manufacture alternative PC models. We run a series of 100 games to compare this strat-
egy against all the previous ones developed, while introducing a SouthamptonTAC-like
behaviour (TestAgent) to observe the results. However, as we haven't had the chance to
test this strategy against other real agents and all our strategies operate in a similar way,
the results as shown in Fig. 5(b) are not conclusive.
6
Conclusions
This paper presented the Socrates trading agent and the strategies employed in the TAC
SCM 2004 competition. Socrates is a production-driven agent which attempts to keep
factory utilisation to the maximum throughout the game. As such it is plagued by a
number of problems as has been described, more so as customer demand is not taken
into consideration when making orders to the suppliers or scheduling production. We
hope to address some of these problems for the TAC SCM 2006 competition. In partic-
ular, we are working on a strategy that concentrates equally on the customers' side as
well as on the supplier side. We would like to explore the idea of using variable scoring
functions to deal with the supplier side. We are also considering the scheduling prob-
lem under a new light, using a daily schedule as well as a dynamic global production
schedule that takes into account the current inventory of components and the component
ActiveOrders
in deciding which customer orders can be satisfied.
References
1. TAC SCM specification. Available at http://www.sics.se/tac/.
2. Sun Java Center J2EE Patterns.
http://java.sun.com/developer/technicalArticles/J2EE/
patterns/, March 2001.
