Decision Programming for Multi-Stage Optimization under Uncertainty

Research output: Working paperScientific

2 Downloads (Pure)

Abstract

Influence diagrams are widely employed to represent the structure of discrete multi-stage decision problems under uncertainty. In this paper, we develop the \textit{Decision Programming} modeling framework which extends influence diagrams by admitting several types of constraints and formulations to which optimal solutions can be computed with mixed-integer linear programming solvers. In particular, Decision Programming makes it possible to (i) omit the usual 'no forgetting' assumption where earlier decisions need not be known when making later ones; (ii) accommodate several types of deterministic and chance constraints, including those based on risk measures such as Conditional Value-at-Risk; and (iii) incorporate multiple objectives and calculate all non-dominated strategies. In the context of project portfolio selection, Decision Programming can be viewed as an extension of Contingent Portfolio Programming (Gustafsson and Salo, 2005) to problems whose scenario probabilities depend endogenously on project decisions. We provide illustrative examples and evidence on the computational performance of Decision Programming formulations.
Original languageEnglish
Publication statusSubmitted - 21 Oct 2019
MoE publication typeNot Eligible

Fingerprint Dive into the research topics of 'Decision Programming for Multi-Stage Optimization under Uncertainty'. Together they form a unique fingerprint.

  • Cite this