Clustering techniques applied to a large number of scenarios under
uncertainty allows the selection of a reduced, however, representative set of the
complete set of scenarios. In other words, it allows to select a sample that contains
a smaller amount of elements to the point of sufficiently reducing the total data
volume and obtaining efficiency gains in data processing. The challenge is that the
sample must, above all, be able to preserve the characteristics of the stochastic
process that originated it. To this end, this study proposes a methodology for
selecting stochastic scenarios using the classic Set Covering model, inspired by the
forward selection method proposed by Heitsch and Romisch (2003). Applied in the
calculating of stochastic demand for tools and services for the construction of
offshore oil exploration wells, this approach presents a different scenario
conception from the one used by the authors. The set of scenarios consists of
activity schedules generated from the introduction of uncertainties in the planning
of each activity, which are static, independent and with multiple attributes. A
sensitivity analysis compares the results of the demands calculated with the
scenarios selected by the Set Covering Problem (SCP) and the demand calculated
with all the universe of scenarios. The SCP was solved, in this application, in its
classic version using an exact algorithm and a heuristic algorithm. The results
appoint na unexpressive loss in the final result of the demand calculated with
reduced scenarios and with the complete set of scenarios. The simple first solution
heuristic presented a satisfactory result in relation to the performance gain versus
reliability, and indicates the potential of the method if solved with metaheuristic
and local search algorithms.
|