Information Technology Reference
In-Depth Information
13.10 Evolutionary Strategy
Evolutionary strategy(ES) is a method to solve parameter optimization problems
by simulating natural evolution process. Early evolution treatment is based on a
population consisting of many individuals and an operator: mutation.
Evolutionary strategy operates mainly on individual-level. It can be simply
described as follows:
1. The defined problem is to find a n-dimensional real vector
x
, such that the
maximum value of function
(
x
):
R
n
ŗ
R
is obtained.
2. The initial population of parental vector is randomly selected from the
feasible range of each dimension.
3. Offspring vector is created by adding parental vector with Gaussian random
variable with mean square deviation of zero.
4. Create next generation's parents by selecting vectors according to minimum
error principle.
5. The process terminates, if the standard deviation of vectors keeps unchanged,
or there is no effective calculation method.
F
13.11 Evolutionary Programming
Evolutionary programming(EP) is to search a computer program, which is coded
as an individual, with high fitness in the space consisting of all possible computer
programs. In evolutionary programming, there are several thousands computer
programs used to evolve. Evolutionary programming is first proposed in 1962 by
L.J. Fogel, A.J. Owens and M.J. Walsh.
Evolutionary programming focuses on the change of behaviors of species.
The representation of evolutionary programming is naturally oriented to task-
level. Once a representation of fitness is determined, mutation operation can be
defined based on such a representation and related offspring are created based on
concrete parental behaviors.
Evolutionary programming starts with an initial population consisting of
randomly generated computer programs, which consist of the functions suitable
for problem space. Thus functions can be normal arithmetic function, normal
programming operation, logic function or particular function. Each
individual(computer program) in population is evaluated by fitness value, which
is related to particular field.
