Posted by yaman barlas <
ybarlas@boun.edu.tr>
Dear Fadl;
We recently developed an algorithm and software that tries to do the following:
- Given a dynamic 'pattern' (a 'shape', not a couple of numbers), the algorithm
can 'recognize' the shape and classify in the proper 'pattern class'
- We developed a software that can integrates the above algorithm and any given
VENSIM model.
- This means that the software can for instance:
i- do automated extreme condition testing (validation) whereby the user states
a set of conditions and the expected dynamic patterns that that the SD model
'should' yield. You check a few minutes (or hours) later and the software can
tell you which of those extrem conditions resulted in 'pass' and which resulted
in 'fail'. (So in a nutshell, this is Reality Check of Vensim, done on
qualitative patterns rather than numbers, - what I call in my papers 'indirect
structure testing')
ii- more interestingly, you can specify a set of parameters and ranges and a
class of 'desired pattern' and come back (a few hours, days or weeks!) later to
find out what set of parameter values -if any- yielded your desired pattern and
in what order of 'degree of match'.
iii- you may of course apply the above to parameter calibration and ultimately
to policy design...
We are still in the process of testing and improving the algorithm and the
software.
I presented the algorithm a couple of years ago (co-author Korhan Kanar).
More up to date: We plan to present and demonstrate the below software at the
ISDC conference in Boston this summer. You may want to attend the session. I do
not yet know the schedule. We hope to distribute the software free to all
interested for research and teaching purposes in a couple of months...
Yaman Barlas
---------------------------------------------------------------------------
Yaman Barlas, Ph.D.
Professor, Industrial Engineering Dept.
Bogazici University,
34342 Bebek, Istanbul, TURKEY
Fax. +90-212-265 1800. Tel. +90-212-359 7073
http://www.ie.boun.edu.tr/~barlas
SESDYN Group:
http://www.ie.boun.edu.tr/labs/sesdyn/
-----------------------------------------------------------------------------
AUTOMATED DYNAMIC PATTERN TESTING, PARAMETER CALIBRATION AND POLICY IMPROVEMENT
Suat Boð
Koc University, Dept of Industrial Engineering
34450 Sariyer, Istanbul – Turkey
Tel: ++90 212 338 1676
sbog@ku.edu.tr
Yaman Barlas
Bogazici University, Dept. of Industrial Eng.
34342 Bebek, Istanbul - Turkey
Tel: ++(90) 212 359 7073, Fax: ++(90) 212 265 18 00
ybarlas@boun.edu.tr
System Dynamics (SD) is a special type of simulation modeling where output
validity refers to validating the patterns of dynamic behaviors, such as
oscillations, growth or decline. The developers and users of these models (the
decision makers and people affected by decisions based on such models) are all
rightly concerned with whether a model and its results are “valid.”
Structural model validity and validation have long been recognized as one of
the main issues in system dynamics. This concern is addressed through pattern
recognition and testing in this paper. Another issue in dynamic simulation
methodology is parameter calibration; assuming that the structure of simulation
model constructed by the user is valid. Parameter calibration is the
minimization of an error function which is a measure of the correspondence
between numerically calculated output patterns and the respective real behavior
patterns. We offer a software that does automated parameter calibration with
respect to a given (desired) dynamic pattern. This particular feature can also
be used in policy improvement design.
Keywords: Dynamic Pattern Recognition, Structure validity Testing, Parameter
Calibration
Posted by yaman barlas <
ybarlas@boun.edu.tr>
posting date Thu, 2 Jun 2005 15:16:46 +0300