AAU Student Projects - visit Aalborg University's student projects portal
A master's thesis from Aalborg University
Book cover


Influence Diagrams Involving Time, a Framework for Decision Problems Involving Time

Authors

;

Term

4. term

Publication year

2003

Abstract

Denne afhandling undersøger beslutningsanalyse, hvor tid er en eksplicit, målelig faktor. Med afsæt i tidligere projektarbejde opstiller vi en række krav til rammer, der modellerer beslutningsproblemer, der involverer tid (DPIT'er). Vi præsenterer en ramme, Indflydelsesdiagrammer med Tid (IDIT'er), der opfylder disse krav, og som oprindeligt er foreslået i tidligere arbejde. Vi udvider IDIT'er, så de dækker flere tidsaspekter, herunder lokale nyttefunktioner (gevinster/omkostninger), der først realiseres efter den sidste beslutning. Derudover udvikler vi en metode til at løse IDIT'er og finde en optimal strategi. Fremgangsmåden udligner først de asymmetrier, som tidsdimensionen indfører, og reducerer dermed problemet til flere symmetriske delproblemer. Disse delproblemer løses med en beregningsmetode baseret på stærke junction trees med lazy propagation. Vi demonstrerer ramme og løsningsmetode med to eksempler og diskuterer fordele og ulemper ved at bruge én samlet ramme til alle tidsrelaterede forhold frem for flere mindre, specialiserede rammer.

This thesis examines decision analysis where time is an explicit, measurable factor. Building on earlier project work, we define requirements for frameworks that model decision problems involving time (DPITs). We present a framework—Influence Diagrams Involving Time (IDITs)—that meets these requirements and was originally proposed in earlier work. We extend IDITs to capture additional temporal aspects, including local utility functions (payoffs) that occur after the final decision point. We also develop a method to solve IDITs and identify an optimal strategy. Our approach first removes asymmetries introduced by time, reducing the problem to several symmetric subproblems. These subproblems are solved using a computational technique based on strong junction trees with lazy propagation. We illustrate the framework and solution method with two examples, and we discuss the benefits and drawbacks of relying on one comprehensive framework for all time-related issues versus several smaller, specialized frameworks.

[This abstract was generated with the help of AI]