Evolutionary Computation - Computational Intelligence in Time Series Forecasting - page 209

nozzles and, using a genetic evolution process, ends with finding the optimal
nozzle shape. The solution concept used by them was termed evolutionary
strategy .
Evolutionary strategy also relies on the mechanism of evolutionary
computation, but it uses it in an original way. In contrast to genetic algorithms,
which aim at solving discrete and integer optimization problems, the objectives of
evolutionary strategies are more focused on solving the problems of continuous
parameter optimization. The evolutionary strategy achieves this through the search
from one population of solutions to another, rather than like genetic algorithms
searching from individual to individual. Also, the evolutionary strategy uses
selection, recombination, and mutation as separate genetic activities for generating
a new solution ( i.e. the new generation), which is actually the major difference
with the genetic algorithms.
The basic idea of evolution strategies relies on the hypothesis that, during
biological evolution, the laws of heredity have been developed for rapid
phylogenetic adaptation . This is actually a considerable improvement of the
genetic algorithm concept, which traditionally does not consider the effects of
genetic procedures on the phenotype. The presumption for coding the variables in
the evolution strategy is the realization of a sufficiently strong causality effect ( i.e .
that small changes in the cause must create small changes in the effect).
The climax of the theory of evolution strategy is the discovery of an evolution
window , stating that evolutionary progress takes place only within a very narrow
band of the mutation step size. This fact indicates the need for a rule of self-
adaptation of the mutation step size. These genetic operators were taken straight
from biological evolution and rely strongly on the principle of mutation. In the
problem at hand, a mutation was simply a small change in the overall make-up of a
jet nozzle.
In their experiments, Rechenberg and Schwefel tested the performance of the
evolved jet nozzles after every mutation. After many repeated trial runs of this
kind, they succeeded in producing a jet nozzle that was better than any of the jet
nozzles at that time available on the market. It is remarkable that, for jet nozzle
optimization, no mathematics dealing with fluid dynamics and propulsion was
taken into account. For the experiments, some nozzles available on the market
were taken and evolved further in order to produce, with every evolutionary step, a
better problem solution.
5.4.1 Applications to Real-world Problems
Evolutionary strategies, instead of a step-by-step search for a single problem
solution, from the very beginning deal with a set of potential problem solutions.
The strategies start with a set of initial solutions and improve them through
repeated evolutionary steps until the best solution has been found. After every step,
the degree of improvement is evaluated using some fitness criteria. Before
initiating the next evolutionary step, a decision is made as to what genetic
operators should be selected. Two such operators are dominant here, i.e . mutation
and crossover, whereby mutation is the most frequently used because it offers
prospective changes in the problem solution. The crossover operator, however,