A Graphical Language for Modeling Stochastic Programming Leo Lopes - - PowerPoint PPT Presentation

▶
a graphical language for modeling stochastic programming
SMART_READER_LITE
LIVE PREVIEW

A Graphical Language for Modeling Stochastic Programming Leo Lopes - - PowerPoint PPT Presentation

A Graphical Language for Modeling Stochastic Programming Leo Lopes University of Arizona Robert Fourer Northwestern University INFORMS Annual Meeting 2003, Atlanta, GA Problem Statement We wish to develop a


slide-1
SLIDE 1

INFORMS Annual Meeting 2003, Atlanta, GA

A Graphical Language for Modeling Stochastic Programming

Leo Lopes University of Arizona Robert Fourer Northwestern University

slide-2
SLIDE 2

INFORMS Annual Meeting 2003, Atlanta, GA

Problem Statement

  • We wish to develop a modeling language for

Stochastic Programming that:

✁

Helps decision makers and analysts communicate

✁

Helps the development of models from start to finish

✁

Helps document existing models

✁

Helps insert optimization models into information technology infrastructures

✁

Is formal and complete

slide-3
SLIDE 3

INFORMS Annual Meeting 2003, Atlanta, GA

Usage Scenarios

  • Elicitation
✁

An OR consultant is developing a model for a business client

  • Forward-Engineering
✁

A team of OR experts is exploring approaches for modeling a problem

  • Reverse Engineering
✁

An OR expert is taking over maintenance of a model from another OR expert

slide-4
SLIDE 4

INFORMS Annual Meeting 2003, Atlanta, GA

Context: People

  • OR Expert
✁

Has expertise in developing, solving, and analyzing models

✁

Has extensive technical training

  • IT Expert
✁

Know what information is available where

  • Decision Maker
✁

Knows about the problem in intimate detail

slide-5
SLIDE 5

INFORMS Annual Meeting 2003, Atlanta, GA

Context: Decision Framework

Make Decision Observe Uncertainty Make Decision Observe Uncertainty Make Decision Observe Uncertainty Make Decision

…

slide-6
SLIDE 6

INFORMS Annual Meeting 2003, Atlanta, GA

Context: Deliverable

min c

T x

E f , x s . t Ax

  • b
slide-7
SLIDE 7

INFORMS Annual Meeting 2003, Atlanta, GA

Summary of Tradeoffs

  • Decision
✁

Which graphical language to produce?

  • Objective
✁

Create a language that is as simple, complete, and formal as possible

  • Constraints:
✁

We need to support a class of sophisticated mathematical models

✁

We need to provide different views suitable for different audiences

slide-8
SLIDE 8

INFORMS Annual Meeting 2003, Atlanta, GA

Summary of Solution

  • Use the Unified Modeling Language (UML) as

a base

✁

Proven graphical language for object-oriented systems engineering

✁

Has been extended to fields as varied as cryptography and Business Analysis

✁

Widely used and strongly supported by the industry

✁

Currently being taught to many OR, IT and Business students.

slide-9
SLIDE 9

INFORMS Annual Meeting 2003, Atlanta, GA

Aspects of a Stochastic Programming Problem with Recourse

  • Decision Model
✁

What are the decisions and constraints at each stage?

  • Dynamic Model
✁

What type of uncertainty is present at each stage?

✁

How does the uncertainty affect the decision process?

  • Interface Model
✁

What are the inputs and outputs of this model?

slide-10
SLIDE 10

INFORMS Annual Meeting 2003, Atlanta, GA

Problem: Interface Model Issues

  • Organize all the model components
✁

Sets

✁

Variables

✁

Parameters

  • Communicate the model requirements to IT

professionals

slide-11
SLIDE 11

INFORMS Annual Meeting 2003, Atlanta, GA

Solution: Class Diagram

distance

slide-12
SLIDE 12

INFORMS Annual Meeting 2003, Atlanta, GA

Problem: Dynamic Model Issues

  • What is being modeled as uncertain?
  • At what point is it observed?
  • How does it affect the decision model?
slide-13
SLIDE 13

INFORMS Annual Meeting 2003, Atlanta, GA

Solution: Stochasticity Diagram

slide-14
SLIDE 14

INFORMS Annual Meeting 2003, Atlanta, GA

Stochasticity Diagram (continued)

slide-15
SLIDE 15

INFORMS Annual Meeting 2003, Atlanta, GA

Problem: Decision Model Issues

  • What level of detail do we wish to represent?
  • Can we (should we) produce a notation that is

isomorphic to algebra?

  • Simplicity x Ambiguity
  • Aggregations
slide-16
SLIDE 16

INFORMS Annual Meeting 2003, Atlanta, GA

Solution: Element Diagram

Assignment.hours

slide-17
SLIDE 17

INFORMS Annual Meeting 2003, Atlanta, GA

TSP Element Diagram

slide-18
SLIDE 18

INFORMS Annual Meeting 2003, Atlanta, GA

Element Diagram (continued)

slide-19
SLIDE 19

INFORMS Annual Meeting 2003, Atlanta, GA

Summary

  • We have developed a language that:
✁

Covers all structural aspects of Stochastic Optimization Problems

✁

Is formal and formally correct

✁

Is flexible, appropriate for computer implementation, and adaptable to different levels of expertise

  • In the paper
✁

Formal treatment of syntax, literature review and related work

✁

Complete examples from the literature

slide-20
SLIDE 20

INFORMS Annual Meeting 2003, Atlanta, GA

Future Work

  • Is it useful?
✁

Human testing

✁

Is implemented / supported by software vendors

✁

Is taught