Multi-Agent Model to Multi-Process Transformation A Housing Market Case Study

1
Multi-Agent Model to Multi-Process
TransformationA Housing Market Case Study

• Gerhard Zimmermann
• Informatik
• University of Kaiserslautern

2
Overview

1. Introduction
2. The Case Study
3. The Multi-Agent Model
4. The Transformation Process
5. Experiments
6. Conclusion

3
1. Introduction

• Research goal
• User activities in computer simulations of the
  built environment
• Problems to be solved
• Observation of user behavior and preferences
• Dynamic formal model of user activities
• Integration of user activity model with built
  environment model
• Mapping of the model into a simulator
• Validation of model and simulator
• Definition of experiments
• Execution and evaluation of experiments

4
Open Questions

• Can the simulation of user activities provide us
  with knowledge that we did not enter into the
  simulator?
• Can the simulation of user activities make the
  design and decision process
• more efficient?
• produce better results?
• produce more reliable results?

5
2. The Case Study
town
market
property
household discontent
city
activity center
household discontent
free property
household content
6
decision process for each household
offered property profile
offered property profile
household profile
current property profile
offered property profile
household preferences
utility metric
criteria
ideal property utility
current property utility
offered property utility
current satisfaction
offer satisfaction
content
take
discontent
reject
7
diversity and dynamics

• different household types have different
  preferences and profiles that change dynamically
• aging, marriage, birth, job change, ...
• households migrate into or out of town
• all properties have different profiles that
  change over time
• aging, renovation, market value, ...
• new properties are added to, old ones taken from
  the market
• construction, land use changes, ...
• ideal application for agents!

8
3. The Multi-Agent Model

• Agent
• Autonomy autonomous dynamic entity that can
• Pro-activeness act based on its own state,
  parameters, behavior,
• Reactivity react to messages from other agents
  or the environment,
• Social ability send messages to other agents
  or the enviroment
• Intelligenz knowlege, learning, belief,
  emotion, ...
• Multi-Agent System
• System of many different communicating agents
  that exist, act, and react concurrentlyin an
  environment

9
more definitions

• Multi-Agent Model
• Model of a specific multi-agent system
• Modeling Technique
• Technique for creating models,
• based on an informal or formal modeling language
• Multi-Agent System Computer Simulation
• Implementation of a multi-agent model,
• based on an implementation language

10
simulator production process
system of agents
reality
modeling process
modeling language
multi-agent model
implementation process
implementation language
multi-agent simulation
11
model refinement
model
model stucture
model behavior
aggregation hierarchy
communication structure
agent behavior
object types
channels messages
finite state machines
SDL block types
SDL channels signals
SDL processes
code
SDT run-time system
computer
12
agent behavior

• extended finite state machine (EFSM)
• state transition

state1
trigger
message condition timer
activity
message
state2
13
state transition graph
14
multi-agent model
15
MTC message transition chartsoffer handshake
scenario
newData
16
4. The Transformation Process
17
example transformation
Problem Description Need accept and react to
offers
18
5. Experiments

• Problem missing statistical data on
• household preferences and profiles
• property profiles
• assignment of household types to property types
• Makeup of data
• 16 household types (individual number and income)
• type distribution according to German census data
• corresponding property types
• random distribution of locations
• rent and value in relation to Sozialwohnungen
• Starting assignment
• initialization run with all households migrating
  into town

19
initialization

• Initial conditions
• 500 households, 550 properties
• initial random urgencies
• stable conditions
• after 65 days 471 households found a property
• 3600 offers with 1..3 properties each / 7 offers
  per household
• 1700 empty offers
• 1800 acceptOffer messages
• 1350 rejected acceptOffer

20
initialization
21
50 new properties at day 150
22
6. Conclusion

• Multi-agent models provide a good mapping of
  diverse active individuals (humans or objects)
• Autonomous processes provide a goodimplementation
  of multi-agent models
• Modeling and implementation can be efficiently
  supported by software engineering processes
• Modeling and implementation can be efficiently
  supported by computer tools
• The housing market case study demonstrates this
  efficiency and shows human behavior