Comparing Implications of ‘Robustness’ and ‘Optimality’ for Decision Support under Uncertainty

Abstract

Conventional simulation‐optimisation approach in decision support under uncertainty aims to find optimal decisions which can meet decision objective(s) under reference scenarios with well‐characterised uncertainty. However, futures often emerge with unexpected circumstances which cannot be predicted in our pre‐specified reference scenarios, and their uncertainty cannot be fully characterised a priori. This article uses robust optimisation based on simulation models as an alternative approach to address such uncertainties of decision support. To show how the two approaches work and how better robust optimisation can be, we formulate a fleet mix problem under uncertainty and implement it in two experiments to compare the results. We conclude that robust optimisation could lead to decisions with probably lower performance than conventional optimisation at imagined future scenarios. However, the advantage of robust optimisation is that the results have better performance in overall and are more reliable over a wide variety of scenarios in the future.