Geology Reference
In-Depth Information
Fig. 5.21 Variation of gamma statistic (
ʓ
) for the data corresponding to different combination of
input data sets [
17
]
5.4.2 Non-linear Model Construction and Testing Results
This section includes the modelling results from nonlinear models like LLR, ANNs
with different algorithms, hybrid models like SVM and ANFIS. The section also
includes the daily solar radiation modelling results obtained from wavelet based
models like NW, W-ANFIS and W-SVM at the Brue catchment.
5.4.2.1 Daily Solar Radiation Modelling with LLR and ANN Models
In this section of the study, two types of models were constructed and tested for
predicting daily solar radiation (Local Linear Regression and Arti
cial Neural
Network). The neural networks were trained using the Broyden-Fletcher-Goldfarb-
Shanno (BFGS) algorithm, the Conjugate Gradient Algorithm and the Levenberg
-
Marquardt training algorithm. The nonparametric procedure based on LLR models
does not require training in the same way as that of neural network models. The
optimal number of nearest neighbours for LLR (principally dependent on the noise
level) was determined by trial and error method and 16 nearest neighbours were
implemented. Then, the performance of the developed LLR technique was com-
pared with the neural network models using three global statistics (correlation
ef
ciency, root mean squared error and mean bias error) as shown in Table
5.1
.
Figures
5.22
a, b and
5.23
a, b show resulting plots (both scatter and line plot) of the
computed (using LLR model with p = 16) and observed daily solar radiation during
the training and validation periods. The estimated solar radiation using the LLR
