sigma workshop part 1: pattern-oriented testing and analysis
DESCRIPTION
SIGMA Workshop Part 1: Pattern-oriented Testing and Analysis. Gönenç Yücel , Can Sücüllü SESDYN Research Group Boğaziçi University, Istanbul. Similarity of dynamic behaviors. Comparing/evaluating model outcomes Calibration Validation Behavior analysis Scenario and policy analysis - PowerPoint PPT PresentationTRANSCRIPT
1
SIGMA WorkshopPart 1: Pattern-oriented Testing and AnalysisGönenç Yücel, Can SücüllüSESDYN Research GroupBoğaziçi University, Istanbul
Similarity of dynamic behaviors• Comparing/evaluating model outcomes
• Calibration• Validation• Behavior analysis• Scenario and policy analysis
• Similarity in terms of what?• Pattern characteristics
Problem identification/description◦ Oscillating inventory◦ Exponentially growing resource consumption◦ Declining sales
Model calibration and behavior validation
Sensitivity analysis◦ Numeric sensitivity vs. behavior sensitivity
Policy analysis
Thinking in patterns
Page 10.00 25.00 50.00 75.00 100.00
Time
1: Noname 4
Page 10.00 25.00 50.00 75.00 100.00
Time
1: Noname 4
Page 10.00 25.00 50.00 75.00 100.00
Time
1: Noname 4
4
Patternwise Comparison
23.07.2014 BATS - Can Sücüllü & Gönenç Yücel - ISDC 2014 Delft
5
Behavior Analysis and Testing Software (BATS)
Overview of BATS with External Parties
7
BATS
USER
Excel
View theResults
Control theSoftware
Parse Commands
ReceiveSimulation results
ImportData-series
ExportData-series
8
• Excel• txt, csv, tab• Drawn by the user• Vensim results files (.vdf)• Vensim models through Model Docking Window
Data Importing Methods
Set parameter values
Model Docking Window of BATS
Send simulation results to Workspace in order to further analyze the model output using data analysis features
Send commands; Run
Exporting of simulation results from Vensim models
Overview of Features of BATS
Data ImportingLoad From FileModel Docking WindowDraw
Data PreparationSplitSelectExponential SmoothingMoving Average< Trend >
Data AnalysisClassifyTrendAutocorrelationAutocorrelation TestSpectral DensityAmplitude EstimationCrosscorrelationSummary StatsGraphical Comparison
Model AnalysisHypothesis TesterBehavior Space ClassifierBehavior Class Mapper
Data VisualizationPlot
*
**
**
Graphical Comparison
11
BATS Hypothesis Tester
Extreme conditions are imposed
parameter values
Model is run and output is obtained
ISTS Algorithm is called
IF hypothesized behavior <-3 FAILED
IF hypothesized behavior >-3 PASSED
IF multiple >-3 INCONCLUSIVE
Behavior Space Classifier
13
Settings
BATS Behavior Space Classifier
Parameter 1 Range 1
Parameter 2 Range 2
Outcome Variable
For each cell, a
simulation run is made
Model output is
taken from Vensim ISTS
Algorithm is called
25 likelihoods
are obtained
The highest value is
entered in the
Classifier Grid cell
Param1=1Param2=9
oscct
Behavior Class Mapper
15
Settings
BATS Behavior Class Mapper
For each point, a
simulation run is made
Model output is
taken from Vensim ISTS
Algorithm is called
25 likelihoods
are obtained
The likelihood value for pre-specified class is plotted on the
graph
Pre-specified class is sshgr
(s-shaped-growth)
-3.99Param1=8Param2=16
Identified Usage Modes of BATS• Structure-oriented behavior validity testing
• Extreme condition tests
• Behavior pattern validity testing• Barlas’ Multi-Step procedure and Graphical Inspection
• Sensitivity analysis• Behavior pattern sensitivity with respect to parameter changes
• Model calibration• Parameter calibration based on behavior patterns
• Policy design• Policy parameter specification based on behavior patterns
• Policy analysis• Policy structure specification based on pattern characteristics of model behaviors
18
• Current version of BATS can be downloaded from the following link (Only for Windows for now. Mac version will be coming soon)• Website:
http://www.ie.boun.edu.tr/labs/sesdyn/projects/bats/index.html• Download links:
http://www.ie.boun.edu.tr/labs/sesdyn/projects/bats/bats_v1_11_win.ziphttp://www.ie.boun.edu.tr/labs/sesdyn/projects/bats/bats_v1_11_win.rar
• Please, keep in mind that this is a beta version.
• And if you give it a try, do not forget to send us some feedback that can help us to improve it.
Gönenç Yücel, Assist. Prof. Boğaziçi UniversityIndustrial Engineering [email protected]
For further questions
Thank you…