Geology Reference
In-Depth Information
Fig. 5.35 Solar radiation as observed and estimated using the SVM model (
ʵ
-SVR with linear
kernel) for the training data set. a scatter plot, b line diagram
Fig. 5.36 Solar radiation as observed and estimated using the SVM model (
ʵ
-SVR with linear
kernel) for the validation data set. a scatter plot, b line diagram
The correlation coef
cient (CORR) statistics were calculated as 0.87 for ANFIS
model and 0.82 for SVM model during the training phase. It can be obviously seen
from Table
5.2
that the ANFIS model approximates the measured values of solar
radiation with a quite high accuracy during training phases than that of SVM model.
However in the validation phase, SVM model has shown better values for some
statistical parameters like ef
ciency and variance values than these of ANFIS
model. It is also found that the prediction capabilities of these two models are lesser
in validation phase than that of ANN models explained in the previous subsection.
The MAPE and MBE for the ANFIS model is lower (0.191 and 0.0003 W/m
2
)
compared to the SVM (0.216 and
1.98 W/m
2
) during training and validation. The
improved MAPE and MBE without signi
−
cant reduction in global evaluation sta-
tistics during training certainly suggest the potential of the ANFIS compared to the
best ANN (ANN-LM). Even though the performance of SVM and ANFIS is less
during validation phase compared to the ANN-LM model, It may be noted that for
an ANN model, the modeller has to perform a trial and error procedure to develop
the optimal network architecture, while such a procedure is not required in
developing an ANFIS and SVM model. Another distinct feature of ANFIS, which
makes it superior to the ANN, is the less number of training epochs required for
convergence which saves computational time of ANFIS in most instants.
