GENERATION OF EFFICIENT DYNAMICAL MODELS FOR
SIMULATION OPTIMIZATION
J. J. Ramos, M. Gutierrez, R. Buil, M.A. Piera, M. Narciso
JuanJose.Ramos@uab.es
Dpt. de Telecomunicaciones e Ingenier´ia de Sistemas
Universitat Aut`onoma de Barcelona.
ABSTRACT
Many engineering applications call for the optimization of a hybrid (discrete/continuous)
dynamic system. Examples include the design of operating procedures for process startup,
shut-down and changeovers. Simulation optimization is one of the most popular
approaches to improve the use of simulation models as a tool to obtain in a short time the
best (optimal or quasi-optimal) decision variables values that minimize a certain objective
function. The time expended to obtain the results depends on, among other important
factors, the simulation time required to evaluate the system dynamics evolution according
to di®erent operation alternatives. This work presents how the PML modelling language
can deal with the system complexity in order to set a simulation model which is e±cient
in the sense of minimizing the computation time required to evaluate the state trajectory
over a period of simulation time.
Keywords: object-oriented modelling, simulation, simulation optimization.
1 INTRODUCTION
World-wide market competition, high product quality
requirements, together with random demands instead
of steady demand, are some key-factors which have
forced industry to change the conventional production
planning methodology towards new heuristic based
methodologies, which could cope with a large amount
of decision variables inherent to present production
architectures. The exact optimal solution of a production
planning and/or operation problem is quite
complex and di±cult, maybe impossible to obtain.
Present industrial processes usually include the operation
of hybrid (discrete/continuous) dynamic system.
An emerging class of problems are those in which the
optimal trajectories are characterized by a sequence
of switches and/or jumps at events, some of which
are dependent on the state of the system satisfying
certain conditions (state or implicit events), and it is
necessary to search over alternative operation policies
(sequences of events) to find the optimal one.
Simulation becomes a highly valuable tool in complex
system configuration settings. However, simulation
limitation arises out of an inability to evaluate
more than a fraction of the immense range of available
options. Simulation-optimization methodology is a
input
output
optimization
strategy
Simulation model
Figure 1: Simulation-Optimization Approach.
new approach that tries to combine evaluation methods
(simulation) and search methods (optimization)
to provide solutions that should be achieved quickly
and reliably.
Simulation optimization can be defined (Carson and
Maria, 1997) as the use of search methods to find input
parameter settings that improve selected output
measures of a simulated system (Figure 1 illustrates
this approach). Mainly, in the absence of tractable
mathematical structures, this approach uses an optimization
package to arrange the simulation of a
sequence of system configurations so that an optimal
or near optimal system operation could eventually
be obtained (this work is not focused on the
.....