Geoscience Reference
In-Depth Information
Table 9
Results of interpolation using ANN
No. of neurons in hidden layer
RMSE from first series (m)
RMSE from second series (m)
CHC (2)
CHC (2)
5
0.708
0.711
10
0.815
0.705
Fig. 2
Comparison of results of different interpolation methods
In NN interpolation, the control points are used for training and a series of
checkpoints (1) for validation and for testing/evaluating; checkpoints (2) are
used for the perceptron network with a hidden layer of 5 neurons and 10 neurons
(Saati et al.
2008
; Karabork et al.
2008
), considering the first period size and with
momentum 0.7 Ns, the following results have been obtained (Table
9
).
In order to compare and evaluate different methods of interpolation, the results
of the current methods and AI techniques are collected from the RMSE through
the series of checkpoints (2). The reason for this is that the AI techniques at the
series of checkpoints (1) in the optimisation process of interpolation parameters
and for the network validation in GA can be used. Therefore, to ensure that the
results of the optimisation process are valid, a series of checkpoints (2) is used
as the independent checkpoints. Consequently, the RMSE rates obtained from the
conventional interpolation methods and AI techniques are compared with a series
of checkpoints (2), which is represented in Fig.
2
as a line graph.
As Fig.
2
shows, in regards to the width of the checkpoints (1) region with
16 m altitude, the triangulation method does not produce appropriate results.
