Computational Intelligence: An Introduction - Computational Intelligence in Time Series Forecasting - page 6

x genetic programming , which is an extension of genetic algorithms to the
population in which the individuals are themselves computer programs
x evolutionary strategies , which deal with “ evolution of evolution ” by
modelling the strategic parameters that control variations in evolutionary
process
x evolutionary programming , which models adaptive evolutionary
phenomena
It is interesting to note that the algorithms of evolutionary computation listed
above, although being structurally similar, have still been quite independently
developed by different researchers without any contact between them.
Genetic algorithms , the first evolutionary algorithms, have been widely studied
across the world and predominantly used for optimum random search. The basic
version of genetic algorithm, originally proposed by Holland (1975), models the
genetic evolution of a population of individuals represented by strings of binary
digits. Based on this model, genetic evolution is simulated using the operations of
selection , crossover , and mutation and monitoring and controlling the simulation
performance using the fitness function .
Genetic programming , developed by Koza (1992), extends the original version
of genetic algorithms to the space of programs by representing the evolving
individuals through individual programs to be evolved. While evolving the
programs, genetic programming for each generation qualifies their fitnesses by
measuring the performances. The qualifying one is used to find out the programs
that at least approximately solve the problem at hand.
Evolutionary strategies have been formulated by Rechenberg (1973) for the
direct solving of the engineering optimization problems. This is performed by
emulation of the evolutionary process of self-optimization of biological systems in
the given environments. It is similar to the case in biological evolutionary
processes. Schwefel (1975) extended the concept of initially formulated
evolutionary strategies and developed the evolution of evolution strategy . In the
latter, the individuals are represented by genetic building blocks and by a set of
parameters related to the strategy and these are used to determine the behaviour of
individuals in the given environment. The strategic parameters are simultaneously
evolved while evolving the genetic characteristics of individuals. During the
evolutionary process, the mutation operator is strictly permitted only if it directly
improves the fitness value.
Evolutionary programming was introduced by Fogel et al . (1975) using the
concept of finite-state automata . In contrast to genetic algorithms, the algorithm
deals with the development of adequate behavioural models , rather than of genetic
models . Evolutionary programming was developed to simulate the adaptive
behaviour of some real-life phenomena and by selecting the set of optimal
behaviours using the fitness function as a measure of success. The substantial
operative difference to genetic algorithms is that evolutionary programming does
not use the crossover operator.