A DEVS-BASED SIMULATION APPROACH FOR STRUCTURE VARIABLE HYBRID
SYSTEMS USING HIGH ACCURACY INTEGRATION METHODS
T. Pawletta, S. Pawletta
Wismar University of Technology, Business and Design
Research Group of Computational Engineering and Automation
PF1210, Philipp-Mueller-Str., D-23952 Wismar, Germany,
t.pawletta@mb.hs-wismar.de
Keywords: Variable Structure Systems, Hybrid Systems, Modular
Hierarchical Systems, DEVS, Simulation Methods, Numerical
Integration
ABSTRACT
This paper introduces a DEVS-based simulation approach
for structure variable hybrid systems using advanced integration
algorithms of recent numeric libraries. It starts
with a brief description of a DEVS-specific modeling formalism
which supports structure dynamics on the aggregation
level of modular hierarchical systems. Then the
design, generation and execution of a corresponding simulation
model is discussed. It is shown, how the simulation
model is able to handle complex structural changes
at runtime dynamically and how it solves differential or
differential-algebraic equations using algorithms of numeric
libraries without further modifications. References
to the realization of the proposed method and its application
on a real engineering system are given.
1. INTRODUCTION
Complex dynamic systems are often characterized by the
feature that their temporal behaviour contains elements
that are modeled most appropriately by structural changes,
i.e. input-output relations between system components as
well as system components themselves may vary in time.
Examples of such systems are given in [2, 7, 12, 8]. Furthermore,
the problem for which a model was developed
may require a structure variable modeling solution. One
example is the problem of adaptive modeling depth [4],
where system parts need more detailed models in critical
situations while for the rest of simulation simpler approximations
are accurate enough. Another application
are structure optimizing experiments [9, 5], where an optimal
system structure is to find under certain constraints.
The idea of structure variable modeling was introduced
for modular hierachical discrete event systems
This work was funded by the German Research Foundation under
project KONDISK no. la724/8-2.
(DEVS) by Zeigler [13]. After that several approaches
are proposed for structure variable modeling. The most
suitable approach for engineering applications is the direct
extension of the aggregation level – coupled system
or network layer in DEVS formalism – by a dynamic
description to specify structural changes [3, 7, 1, 8]. It
supports a real modeling from the viewpoint of technical
units (e.g. plant and controller). In traditional modeling
concepts where this feature is lacked all structure changing
operations have to be modeled in the dynamics describing
atomic systems. This leads to complex atomic
systems and their restricted reuse.
Complex engineering applications are often hybrid
systems, which are characterized by discrete event and
continuous dynamics. A DEVS-based formalism for nonstructure
variable hybrid systems and associated abstract
simulator algorithms were introduced by Praehofer in [10] 1
while Barros proposed an extension of the variable structure
modeling approach on the aggregation level to heterogeneous
flow systems in [1]. Barros approach fulfills
most requirements to model structural dynamics in hybrid
systems, but is restricted to the use of simple integration
algorithms at simulation runtime. All this hybrid
system concepts assume that the model specification is
transformed in an one to one manner to a modular hierachical
simulation model. Such an executable simulation
model contains all necessary information to handle
structural changes on the component aggregation level.
But this causes numerical problems, because the continuous
system description is distributed among different program
objects. High accuracy integration algorithms in
recent numerical libraries require a closed representation
of the continuous behavior specification. For most of this
algorithms it is not useful to adapt them to modular computing
models 2, because of the numerical stability and
the computation performance. On the other side for struc-
1Of course the approach supports structural changes in atomic systems
to model state discontinuities or equation switching of continuous
quantities.
2The term computing model is used equivalently for an executable
simulation model.
.....