Information Technology Reference
In-Depth Information
direction. Based on the above ideas, he proposed genetic algorithm to simulate
evolution process of structures.
In genetic algorithm, structure is first coded in the form of character string,
and each character string is called an individual. Then a set of character strings,
which is called a population, are iteratively operated. Each iteration is called a
generation, which contains the process of keeping superior structure in
population and that of information transform among structural and stochastic
character strings. Similar to natural evolution, genetic algorithm gets good
chromosome by operating gene in chromosome, so as to solve related problem.
Similar to nature, genetic algorithm knows nothing about the problem to be
solved. It just needs to evaluate each chromosome, and selects chromosome
according to evaluation value, so that the chromosome with higher value has
more chances to propagate. In genetic algorithm, a bit string serves as a
chromosome, while single bit serves as a gene. Population is randomly generated,
and each individual in population is a bit string. Each individual is endowed with
a numerical value, which is called fitness. An individual with low fitness is
deleted while one with high fitness then is selected for further operations. The
common genetic operators mainly include reproduction, crossover, mutation and
inversion.
Comparing with traditional optimization algorithm, genetic algorithm has the
following features:
(1) Genetic algorithm does not directly operate parametric variables, but utilizes
related codes of parametric variables;
(2) Genetic algorithm's searching starts with one point which is from population
but not from problem space;
(3) Genetic algorithm utilizes fitness value, not differential coefficient or other
information;
(4) Genetic algorithm utilizes probabilistic transition rules, not deterministic rules.
Genetic algorithm can solve the hard problem through coding technique and
breed mechanism to simulate the complex phenomenon. In particular, it is not
constrained by the restriction hypothesis of search space and does not require the
searching space to be continuous and derivative; it can get global optimal
solution with high probability from discrete, multiple-extremum and noising
high-dimensional problem. In addition, genetic algorithm's inherent parallelism
makes it very suitable for massively parallel computing. At present, genetic
