Geoscience Reference
In-Depth Information
Thus, when designing the system, a design that maximizes customer utilities is often
optimal, even though such maximization is not explicitly enforced in the model.
17.6
Conclusions
In this chapter we have focused on a rather specialized sub-field of stochastic loca-
tion models: problems with congestion and static customer assignments. However,
as discussed above, this is a very active and growing field of research. We believe
that the key drivers of this growth are that, on the one hand, SLCIS models do
capture very important trade-offs and stochastic effects that must be taken into
account when designing many real-life systems. On the other hand, these models
retain enough structure to enable exact algorithmic approaches and managerial
insights that may not be available when more complex models (e.g., models with
mobile servers or dynamic customer assignments) are considered.
The variety of SLCIS models considered in the literature is quite bewildering. We
have tried to systematize the models along two dimensions: by customer response
and demand elasticity (leading to our NR/AR/DR/FR classification), and by the key
structural elements of the models, as described in Sect.
17.5
. We believe that this
classification should be useful to future researchers in this field, both with respect
to the importance of clearly spelling out the assumptions for customer behavior and
key model objectives, and with regards to realizing what key difficulties may arise
for a given model type.
Many open questions remain, as should be clear from the preceding sections.
The assumptions made with respect to queueing behavior in many models are
quite restrictive and could likely be generalized using the approximation approaches
described in Sect.
17.2.3.2
. The assumptions underlying NR models or AR models
with distance-only utility are questionable and could lead to under-performance of
the resulting system (especially with respect to the realized demand). The reliance of
many authors on heuristic approaches without the ability to benchmark the resulting
solutions versus the optimal ones is not comforting given the strategic nature of
decisions underlying SLCIS models. In short, many ways to improve on the existing
models remain to be explored. We hope that some of these improvements will be
investigated in the next generation of SLCIS models.
Finally we would like to mention that many of the issues that have been
explored in the SLCIS context (customer response, elastics demand) are still
waiting to be addressed in the models with mobile servers/dynamic customer
assignments. As noted earlier, these models involve a different level of complexity,
with the underlying queueing systems being much less tractable. Nevertheless,
the assumptions regarding customer behavior and response are very important and
deserve further study.
