Optimisation of
electromagnetic
devices
285
COMPEL ± The International
Journal for Computation and
Mathematics in Electrical and
Electronic Engineering,
Vol. 18 No. 3, 1999, pp. 285-297.
# MCB University Press, 0332-1649
The optimisation of
electromagnetic devices using
niching genetic algorithms
J.A. Gallardo and D.A. Lowther
Electrical and Computer Engineering Department, McGill University,
Montreal, Quebec, Canada
Keywords Electromagnetics, Genetic algorithms, Optimization
Abstract The use of niching genetic algorithms can provide a method of a more widespread
search of the design space for a device than more conventional methods. It provides, in effect, a
breadth first rather than a depth first search. Thus several alternative designs may be evaluated
in parallel.
Introduction
The optimisation of electromagnetic devices is becoming more and more
important, especially now that it is possible to predict the performance of a
device to a high degree of accuracy (Neittaanmaki et al., 1996). While it is
possible to obtain sensitivity information from analysis systems, this is by no
means routine in conventional programs and thus the information available
tends to be only the values of the performance with little information about
their rate of change with parameter variations. This makes it difficult, but not
impossible, to use deterministic methods to search for an optimal solution and
has led to considerable work in the use of stochastic systems instead (Alotto et
al., 1998). Genetic algorithms fall into this latter class of systems. In general,
such systems move towards an optimum by a random process, which always
has a finite possibility of actually moving away from the desired answer. This
feature is useful in that it provides a method for extracting the process from a
local minimum (Uler and Mohammed, 1996; Uler et al., 1994).
The basic genetic algorithm looks for a single minimum in the design space.
However, depending on the constraints imposed on the design and the cost
functions to be minimised, there may, in fact, be several solutions and the
designer may well want to be presented with all these possibilities before
choosing a final design (Ramberger and Russenschuck, 1998). The question
then is how can several minima in a particular design space be identified
simultaneously? One proposed solution (Sareni et al., 1997a; 1997b) is to use a
genetic algorithm which explores the design space with several populations, or
``niches'' rather than just one. The niches are designed to span the design space
and individuals are assigned to niches using a Hamming distance calculation
based on a binary string which encodes the parameters of the device. These
This work was supported in part by a grant from the Natural Science and Engineering
Research Council of Canada.
The current issue and full text archive of this journal is available at
http://www.emerald-library.com