simulation@uflorida.cis.ufl.edu (Moderator: Paul Fishwick) (08/30/89)
Volume: 10, Issue: 7, Wed Aug 30 10:00:46 EDT 1989
+----------------+
| TODAY'S TOPICS |
+----------------+
(1) System Dynamics: A Response
(2) Call for Participation: AISIG90 Conference
(3) Requesting Papers on Discrete Event Simulation
(4) SMPL Report and Questions
(5) Call for Papers: AI, Simulation and Planning
* Moderator: Paul Fishwick, Univ. of Florida
* Send topical mail to: simulation@bikini.cis.ufl.edu OR
post to comp.simulation via USENET
* Archives available via FTP to bikini.cis.ufl.edu, login as
'anonymous', use your last name as the password, change
directory to pub/simdigest.
* Simulation Tools available by doing above and changing the
directory to pub/simdigest/tools.
-----------------------------------------------------------------------------
Date: Thu, 24 Aug 89 06:42 MST
From: CELLIER%evax2@rvax.ccit.arizona.edu
Subject: RE: System Dynamics (Newsletter)
To: Fishwick@bikini.cis.ufl.edu
X-Vms-To: IN::"Fishwick@bikini.cis.ufl.edu"
One of the more modern (1982) bibliographies of applications of System
Dynamics is in my book: "Progress in Modelling and Simulation", Academic
Press. The article was written by J.D.Lebel (chapter 8), pp.119-158.
I totally agree with you on your remark relating to parameter insensitivity.
This statement is a gross simplification. The major problem that I see with
System Dynamics lies in the ease of its applicability. It is so easy to
create a System Dynamics model with 100 parameters which is impossible to
validate on the basis of a usually rather limited set of measurement data.
Consequently, you got two problems: (i) System Dynamics does not recognize its
own event horizon. I.e., once you created the model, the methodology will be
happy to produce trajectory behavior over a ridiculously long time span. Since
most systems (e.g. in business) are inherently time-varying, it is YOUR
responsibility as a modeler to verify for how long your trajectory behavior
may possible be trustworthy. In particular, never use System Dynamics to
extrapolate any states far beyond their measured range, e.g., don't use
System Dynamics to describe a world model in which the pollution assumes a
value which is 100 times larger than any value ever recorded. (ii) A good
curve fit between a few measured values and your model is NOT sufficient to
validate the structure of your model. Any model can be used to curve-fit
any data if it is just made complicated enough (i.e., if enough parameters are
added). A System Dynamics model may be valid to predict the future over a
limited time horizon, but that does not make it valid to conclude facts from
the found model structure. These may be some of the reasons that made System
Dynamics somewhat dubious in some people's mind. System Dynamics can still
be a great methodology if applied wisely, but DON't succumb to the temptation
of falling in love with your System Dynamics model. Blind love usually ends
tragic.
Your reference to chaotic models is interesting. In fact, most chaotic models
are third order or higher Lotka-Volterra type models. The sensitivity comes in
at the peak where the first derivative is impressive, and the second derivative
is impossible. Interestingly enough, these Lotka-Volterra type models can in
fact be looked at as specialized System Dynamics type models (with a small set
of parameters). Due to some peculiarities of the error equation, System
Dynamics works on THESE models about as well as any other methodology, in
spite of the sensitivity problem.
One methodology that overcomes some of the deficiencies of System Dynamics is
Inductive Reasoning (cf e.g. my paper in Simulation 52:3 (Autopilot)).
In this methodology, the model validation process is unseparable from the
model building process. Inductive Reasoning will reject to simulate
(forecast) the system behavior beyond what can be validated from the available
data. It will also give you some insight into the number of parameters that
your model can/should contain. Interestingly enough, the larger the number
of parameters, the smaller is the event horizon. Once you gained the
necessary insight into your model, you can return (if you like) to your
System Dynamics methodology, and create models that are validatable on the
basis of the available data.
Francois Cellier
Associate Professor
University of Arizona
Cellier%ECEVAX@RVAX.CCIT.Arizona.Edu
Cellier@Arizevax.Bitnet
Looney::Cellier (Span)
FCellier (Nasamail)
------------------------------
Date: Thu, 24 Aug 89 10:06:11 -0400
From: Paul Fishwick <fishwick@fish.cis.ufl.edu>
To: simulation@ufl.edu
Subject: AISIG90 Call for Participation
CALL FOR PARTICIPATION
----------------------
The Fifth Annual AI SYSTEMS IN GOVERNMENT CONFERENCE
----------------------------------------------------
Date: May 1990
Place: Washington, DC
Submission Deadline: Oct. 1, 1989
* Conference Chair *
Dr. Barry G. Silverman
Institute for AI
George Washington University
Net: barry@gwusun.gwu.edu
* Simulation and Modeling * Point of Contact:
Dr. Richard Modjeski
Phone: (703)-756-1685
Net: modjeski@alexandria-emh2.army.mil
------------------------------
Posted-Date: Mon, 28 Aug 89 11:56:12 +0200
From: gatech!cs.utexas.edu!uunet.uu.net!prlb2!pirotte@bikini.cis.ufl.edu
Date: Mon, 28 Aug 89 11:56:12 +0200
To: bikini.cis.ufl.edu!simulation@cs.utexas.edu
Subject: discrete event simulation
Cc: uunet.UU.NET!pirotte@cs.utexas.edu
I teach a course at the Univ. of Brussels (5th year in engineering
curriculum) in which a few lessons are devoted to an introduction to
discrete event simulation. To illustrate this part, I give assignments
which consist in reading and discussing papers from the technical
literature describing simulation experiments.
To renew my stock of papers (I do not have an easy access to the
specialized literature), I am interested in receiving a copy
of technical papers from conferences and journals that describe a
particular application (construction of a model, etc.) and the solution
with simulation of a problem in the application domain.
Paper copies can be sent at the following address:
A. Pirotte
Philips Research Lab.
2 avenue Van Becelaere
1170 Brussels, Belgium
Thank you very much!
------------------------------
Date: Tue, 29 Aug 89 18:15 N
From: Patrick Van Renterghem / Transputer Lab <PVR%AUTOCTRL.RUG.AC.BE@CUNYVM.CUNY.EDU>
Subject: SMPL report/questions
To: simulation-maillist@ufl.EDU
X-Vms-To: IN%"simulation-maillist@ufl.edu"
Comments: From: Patrick Van Renterghem, State University of Ghent
References: > The Transputer Lab, Grotesteenweg Noord 2, +32 91 22 57 55
Keywords: > B-9710 Ghent-Zwijnaarde, Belgium, Europe. Fax: +32 91 22 85 91
Dear Simulationists,
Several weeks ago, I placed this message on this bulletin board, hoping
that someone would provide me with some answers for my questions. But there
was not one reply, so I am posting the message again. Please reply to me if
you know answers to the questions.
Patrick
--------------------------------------------------------------------------
This is a short write-up on the porting of the SMPL program (which is
available via ftp as described on this mailing list) to our Microvax II/VMS
and to a transputer system.
* Our C compiler complains about things like '=+' and '=*'. Although this can
be easily solved by replacing them by '= +' and '= *', I would recommend to
give your equal sign and expressions some more space. Eliminating blanks
decreases readibility.
* time () and pause () are already in the VAX library. I have replaced the smpl
time and pause routines by smpl_time and smpl_pause.
* Once these problems were solved, the program ran on the Microvax. We then
ported the package to a single transputer system, mainly because we expected
it to run a lot faster on a T800 transputer than on a MicroVAX or a PC/AT.
And indeed it does. We used 3L's Parallel C and an Inmos B004 board with a
T800 and 2 MBytes of slow memory. Naturally, our intention is to get the
application running on several transputers. We know all about lockstep and
time-warp approaches, but does anyone have an implementation of time-warp that
we can use ?
* A number of questions remain:
- are there any known bugs in smpl ?
- is it public domain (can we pass it on ?) ?
- would it be possible to have the smpl/PC and mtr () programs as well ?
- does anyone have experience with parallelization of discrete event
simulation on a distributed memory machine ?
*****************************************************************************
* Patrick Van Renterghem, BITNET: pvr@bgerug51.bitnet *
* R&D Assistant, EDU: pvr%bgerug51.bitnet@cunyvm.cuny.edu *
* State University of Ghent UUCP: mcvax!bgerug51.bitnet!pvr *
* Belgium JANET: PVR%earn.bgerug51@earn-relay *
* *
* Automatic Control Lab/The Transputer Lab, Tel: +32 91 22 57 55 ext. 313 *
* State University of Ghent, Fax: +32 91 22 85 91 *
* Grotesteenweg Noord 2, *
* B-9710 Ghent-Zwijnaarde, Belgium *
*****************************************************************************
------------------------------
Date: Tue, 29 Aug 89 22:55 MST
From: ZEIGLER%evax2@rvax.ccit.arizona.edu
Subject: call for papers
To: fishwick@fish.cis.ufl.edu
X-Vms-To: IN::"fishwick@fish.cis.ufl.edu"
ANNOUNCEMENT AND CALL FOR PAPERS
AI, SIMULATION AND PLANNING IN HIGH AUTONOMY SYSTEMS
March 26 - 27, 1990
Sponsoring Agencies: The University of Arizona, College of Engineering and
Mines, Department of Electrical and Computer Engineering, Martin Marietta Data
Systems and McDonnell Douglas Corporation.
In Cooperation With: AT&T, Bell Canada, IEEE Computer Society, Information
Machines, NASA-Ames Research Center, Rand Corporation and Siemens Corporate Research.
Increasing the autonomy of systems in automation and robotics is a key element
in system engineering with such goals as:
- reducing the need for human intervention and supervision in remote or
hazardous environments.
- relieving humans of attending to complex procedures not directly related
to their primary objectives.
- providing knowledgeable assistance in executing higher level decision
making functions.
- increasing the rate of decision making beyond that reasonably supportable
by a human controller.
At the intersection of computers, control, information, and management, design
for high autonomy requires the tools of AI and Simulation to successfully
integrate decision making and physical layers. Typical issues raised include
design stage testability, multi-abstraction model/knowledge bases,
discrete/continuous and symbolic/numeric interfaces, self-embedded model
construction, and self-planning under behavior constraints.
The conference will feature invited and contributed papers in the technical
areas of: simulation-evaluated planning and scheduling, qualitative reasoning,
device modelling for operations/diagnostics/repair, knowledge-based simulation
and design, multi-agent computer architectures and parallel simulators,
discrete event dynamic systems, quality assurance issues, intelligent control,
model-based perception, model reuse and evolution, embedded learning and
adaptation, and related topics.
Papers describing applications are also solicited in such areas as autonomous
vehicles, telerobotics, factories of the future, design support environments,
mission planning support, logistics, wargaming, and in particular, novel
applications which fuse modelling paradigms, e.g. combined
cognitive/social/natural models and combined neural/symbolic processing.
Persons wishing to present papers should submit four copies of an abstract.
Abstracts should be not more than three double-spaced pages, including figures
and representative citations. Abstracts must be received not later than
Friday, November 17, 1989. Mail abstracts to The Office of Engineering
Professional Development, University of Arizona, Box 9 Harvill Building,
Tucson, AZ 85721, (602) 621-3054 or FAX: (602) 621-1443. Accepted papers will
be determined by December 15, 1989. Proceedings will be published in a form
permitting wide distribution. Selected papers may be published in a special
issue of a major journal.
------------------------------
END OF SIMULATION DIGEST
************************