Geoscience Reference
In-Depth Information
the network weights are based on the gradient method and the error signals are
corrected and adjusted. Back propagation is used for explaining the correction of
network behaviour, which is opposite to the weight communication between syn-
apses (Wiszniewski
1983
).
3 Genetic Algorithm
In 1960, Rechenberg presented the basic idea of evolutionary algorithms, where
GA can be derived from. This is, in fact, a computerised search method, which
is based on the optimisation algorithms, genes and chromosomes, founded in
Michigan University by Professor Holland (Holland et al.
1989
) and then further
developed (Freisleben and Merz
1996
).
In this algorithm, due to being derived from nature, stochastic search processes
are used for optimisation and learning problems (Sheta and Turabieh
2006
). In
nature, chromosome combinations will produce better generation. Among these
mutations occurring within the chromosomes it may improve the next generation.
GA solves these problems by using this concept (Sivanandam and Deepa
2010
).
Overall operations of this algorithm are; fitting, selecting, combining and
mutating (Ravagnani et al.
2005
). In the algorithm process, an initial population
of chromosomes is selected for the creation of a new and possibly better genera-
tion. Each chromosome has various arrays that should be optimised. After creat-
ing the initial population of merit (cost consumption) for each chromosome in the
population the calculation is based on the objective function. The major parts of
the chromosomes that are too costly are left behind and the chromosomes that are
sufficiently cost for evaluation are to be kept to produce the next generation of
children. Among them, there are a number of elite chromosomes, which are con-
sidered to be low-cost, are chosen and remain untouched for the next generation.
To determine the number of chromosomes needed to integrate, parents are selected
to produce offspring. Two chromosomes are selected as parents when they are
combined. Sometimes randomly genes are changed; a mutation occurs and enables
the algorithm to search for a wider area. In other words, new generation can be
created by reproductive processes of combining gene and mutation. This process
must be repeated many times to achieve convergence and create an optimal solu-
tion (Haupt and Haupt
2004
).
4 Height Interpolation Methods
The main purpose of using the known point height interpolation is to determine
the heights of the unknown's middle points. In 2004, Yang examined different
methods for interpolation according to the accuracy and applicability by using
Surfer 8.0 software (Yang et al.
2004
). These methods can be divided into different
