Authors:
Jérémy Boes
;
Pierre Glize
and
Frédéric Migeon
Affiliation:
Université Paul Sabatier, France
Keyword(s):
Complex Systems Modeling, Multi-Agent Systems, Self-tuning, Self-composition.
Related
Ontology
Subjects/Areas/Topics:
Agent Based Modeling and Simulation
;
Application Domains
;
Automotive Industry
;
Complex Systems Modeling and Simulation
;
Formal Methods
;
Mobile Software and Services
;
Non-Linear Systems
;
Sensor Networks
;
Simulation and Modeling
;
Simulation Tools and Platforms
;
Software and Architectures
;
Telecommunications
;
Wireless Information Networks and Systems
Abstract:
Many methods for complex systems control use a black box approach where the internal states and mechanisms of the controlled process are not needed to be known. Usually, such systems are tested on simulations before their validation on the real world process they were made for. These simulations are based on sharp analytical models of the target process that can be very difficult to obtain. But is it useful in the case of black box methods? Since the control system only sees inputs and outputs and is able to learn, we only need to mimic the typical features of the process (such as non-linearity, interdependencies, etc) in an abstract way. This paper aims to show how a simple and versatile simulator can help the design of systems that have to deal with complexity. We present a generator of models used in the simulator and discuss the results obtained in the case of the design of a control system for heat engines.