/* ---- Google Analytics Code Below */

Friday, March 04, 2016

Industrial Simulation vs Agent Models

Had thought about this difference for some time.  A new examination brought this paper up.  A good technical, practitioner eye view.  Here DES means Discrete Event systems (Classic Industrial Monte Carlo simulation) and ABS means Agent Based Systems.  My other, older, observation is that ABS systems are very hard to maintain and augment, their use of the 'agent' model does not simplify their useful design.

Guidelines for Design and Analysis in Agent-Based Simulation Studies
by Parastu Kasaie and W David Kelton

‘ ... As originally DES-modelers, self-taught and still new to the world of ABS modeling, we have occasionally observed a gap between traditional simulation theory (mostly developed for DES) and current practice of ABS models in the literature. These include the lack of justification for choosing ABS models over simpler modeling paradigms; a general tendency for development of “realistic,” heavily detailed ABS models despite the tradeoff between complexity of analysis and transparency of findings; modelers’ personal internal assumptions and mechanisms for agents’ behavior and interactions without explicit validation and consideration of alternatives; instances of excessively parameter-rich ABS models requiring specialized calibration techniques (not generally discussed for DES models); unclear definition of errors in ABS applications as relevant to different sources of modeling mistakes vs. natural stochastic uncertainties; specific challenges faced in validation of ABS models due to their multi-level structure; as well as further special issues in experimentation and sensitivity analysis of these models.’  ... " 

More from W David Kelton.

No comments: