xiaorong xiang, ryan kennedy, gregory madeynom/papers/ads05_kennedy.pdf · 4 v & v for...
TRANSCRIPT
Verification and Validation of Agent-based Scientific
Simulations
1
Xiaorong Xiang, Ryan Kennedy, Gregory MadeyComputer Science and Engineering
University of Notre Dame
Steve CabanissDepartment of ChemistryUniversity of New Mexico
2
OverviewIntroduction
Concepts of Verification and ValidationResearch Objectives and Methods
A Case StudyApply Verification and Validation Methods to the Case StudyConclusionFuture Work
3
Model Verification & Validation (V & V)V & V
Verification: get model rightValidation: get right model
The cost and value influence confidence of model acceptance level
*Adapted from Sargent: “Verification andValidation of Simulation Models”
4
V & V for Agent-based SimulationAgent-based modeling is a new approachDifferent than Queuing Models
Entities: large number of heterogeneous active objects vs. passive objects
Space: continuous or discrete grid space vs. network of servers and queues
Interactivity: high vs. lowActive components: agents vs. queues and serversGoal: discovery vs. design and optimization
Few literature to date address the formalized methodology for V & V of Agent-based Simulations
5
What and HowResearch objective
Generate guidelines or a formalized methodology for V & V of Agent-based Simulations
HowNOM project as a case studyEvaluate and adapt the formalized V & V techniques in industrial and system engineering for DESIdentify a subset of these techniques that are more cost-effective for Agent-based Simulations
6
NOM Agent-based Simulation ModelNSF funded interdisciplinary project
Understanding the evolution and heterogeneous structure of Natural Organic Matter (NOM)E-science exampleChemists, biologists, ecologists, and computer scientists
Agent-based stochastic model Web-based simulation model
7
NOMWhat is NOM?
Heterogeneous mixture of molecules in terrestrial and aquatic ecosystems
Why study NOM?Plays a crucial role in the evolution of soils, the transport of pollutants, and the global carbon cycleUnderstanding NOM helps us better understand natural ecosystems
8
The Conceptual Model IAgents
A large number of moleculesHeterogeneous properties
Elemental compositionMolecular weightCharacteristic functional groups
BehaviorsTransport through soil pores (spatial mobility)Chemical reactions: first order and second orderSorption
9
Stochastic Synthesis: Data Model
Pseudo-Molecule
Elemental Functional Structural
Composition
CalculatedChemicalProperties
and Reactivity
LocationOriginState
10
The Conceptual Model IIStochastic Model
Individual behaviors and interactions are stochastically determined by:
Internal attributesMolecular structureState (adsorbed, desorbed, reacted, etc.)
External conditionsEnvironment (pH, light intensity, etc.)Proximity to other molecules
Length of time step, ∆t
Space2D Grid Structure
Emergent propertiesDistribution of molecular properties over time
11
Implementations
12
V & V of the NOM ModelExamples of V & V techniques
Face validityAnimationGraphical representation
TracingInternal validityHistorical data validation (calibration sets and test sets)Sensitivity analysisPrediction validationComparison with other modelsTuring test
13
V & V of NOM Simulation Model
*Adapted from Sargent: “Verification and Validation of Simulation Models”
14
Face Validity
15
Internal Validity I
0
500
1000
1500
2000
2500
3000
3500
4000
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25
Seed
Num
ber
of M
olec
ules
at t
he e
nd s
tep
AfterBefore
16
Internal Validity II
Number of Molecules
1775.0
1725.0
1675.0
1625.0
1575.0
1525.0
1475.0
1425.0
1375.0
1325.0
1275.0
1225.0
Freq
uenc
y60
50
40
30
20
10
0
Std. Dev = 100.79 Mean = 1473.9
N = 450.00
17
Model-to-Model Comparison ICompare the model with validated oneCompare the model with non-validated oneDifferent implementations
Different programming languagesDifferent packages
Different modeling approachesPredator-Prey model
Agent-based approach vs. System Dynamics approach
Powerful method for ABS
18
Model-to-Model Comparison IIFeatures Alpha Step No-flow ReactionDeveloping Group University of New
Mexico, chemistsUniversity of Notre Dame, computer scientists
Programming language Pascal Java (Sun JDK 1.4.2)
Platforms Delphi 6, Windows Red hat Linux cluster
Running mode Standalone Web based, standalone
Simulation package None Swarm, Repast libraries
Animation None Yes
Spatial representation None 2D grid
Second order reaction Random pick one from list
Choose the nearest neighbor
First order with split Add to list Find empty cell nearby
19
Model-to-Model Comparison III
20
Model-to-Model Comparison IV
21
Model-to-Model Comparison V
22
Conclusion and Future WorkV & V Case StudyModel-to-Model Comparison is Powerful
Collect and evaluate more statistical dataCompare simulation results against empirical dataTweak V & V methodsGenerate guidelines and methodology for V & V of agent-based simulation models
Questions or Comments?
23