Adaptive Genetic Algorithms - Computational Intelligence in Time Series Forecasting - page 312

9
Adaptive Genetic Algorithms
9.1 Introduction
The genetic algorithms, or in general the various evolutionary computations, have
been introduced to the reader in Chapter 5 along with their important
implementation aspects. Genetic algorithms (GAs) are often described as a
gradient-free, robust search and optimization technique, where the search direction,
unlike a gradient-based optimization method, is not biased towards a local
optimum, but, at the same time, GAs can also be applied to an ill defined complex
problem for optimization. However, the above advantages of GAs may be totally
jeopardized because of the extremely long run time required for a complex
optimization problem. Furthermore, even at the end of an extremely large number
of generations the solution obtained from the GA run may be completely
unacceptable. This being the main motivation why GA researchers are constantly
trying to improve GAs in order to obtain an acceptable solution within a reasonable
number of generations of a GA run. With the above objectives in mind, the present
chapter furnishes a few important possibilities, collected from various publications,
for the improvement of a standard GA run.
The most typical features of genetic algorithms (GAs) are:
x genetic representation or encoding of data to be optimized
x initial population of encoded data
x control parameters of the algorithm
x fitness function.
In practical applications, the adequate selection of GA features substantially
influences its performance; and, vice versa , the non-adequate selection of GA
features might lead to nonacceptable problem solutions. To prevent the latter
situation, this was a challenging task of researchers in the 1980s, who tried to
implement various practical concepts to facilitate the feature selection process. In