Geoscience Reference
In-Depth Information
parameters values (KFEXs), (2) the KF scheme with the parameter values optimized
by GA (OPTMs) and (3) no CP scheme (NOCPs).
27.3
Micro-Genetic Algorithm (Micro-GA)
27.3.1
The Optimization Process
The micro-GA, suggested by
Goldberg
(
1989
)and
Krishnakumar
(
1989
)isa
small-population genetic algorithm with reinitialization while standard GAs mostly
use large populations to achieve diversity upon “convergence”. It requires less
computational time than standard GAs (
Krishnakumar
,
1989
;
Lee et al. 2005
;
Wa n g
et al. 2010
). The procedural details of micro-GA were described by
Carroll
(
1996
),
Liong et al.
(
2005
), and
Wang et al.
(
2010
). With a small population there will be
rapid convergence to a possible suboptimal solution, by generating new population
members as soon as a convergence has been achieved in a GA cycle.
The micro-GA has been demonstrated to yield marked improvement over
conventional large-population GAs. Although the range of application of micro-
GA is becoming extensive, its applicability has yet to be explored fully, and is
certainly needed for atmospheric design problems where computational requirement
is enormous.
For experiments in this study, the micro-GA is initialized with a random
sample of individual solutions, with the population size set to five following
Lee
et al.
(
2005
)and
Wang et al.
(
2010
). The chromosomes are generated based on a
tournament selection method in order to select parent genes on which the uniform
crossover operation is applied. The micro-GA does not have mutation operations,
and the algorithm stops when the prescribed number of generations (100 in this
study) is reached. The ranges of
T
c
and
c
to search are set to
600
s
T
c
3600
s
s
1
c
0:1
s
1
, respectively. The chromosome length is set to 10;
and
0:0001
3000=2
10
/
s
1
(i.e.,
thus, the precision for
T
c
and
c
is about 3 s (i.e.,
and
0:0001
0:0999=2
10
), respectively.
27.3.2
Fitness Function
The fitness function to be optimized is defined by using a QPF skill score -
the equitable threat score (
ETS
)(see
Hamill 1999
;
Lee et al. 2006
;
Ya n g a n d
Tung 2003
). The forecast amounts are computed from the 24-h accumulated
total (grid resolved + parameterized) precipitation at a forecast period of 12-36 h
(from 1200 UTC 30 to 1200 UTC 31 August 2002). The 24-h observations of
total precipitation are based on the hourly precipitation data from 615 Automatic
Weather Stations (AWSs) of the Korean Meteorological Administration (KMA).
Search WWH ::
Custom Search