Geology Reference
In-Depth Information
Fig. 5.44 Complexity versus
combined error
—
case study:
solar radiation modelling
The sensitivity of the model with change in inputs used for training is assessed
varying the inputs in the range of
30 % to +30 %. The change in output with
percentage changes in different inputs is averaged and plotted as in Fig.
5.45
. The
slope of the sensitivity lines corresponding to each model in the Fig.
5.45
was taken
as the measure of the sensitivity of the model. The overall model utility index
Overall Model Utility indexcan be calculated once the error and sensitivity values
are available. Figure
5.46
shows the variation of the sensitivity of the models with
the complexity. The sensitivity-complexity hypothesis is as the complexity of the
model increases the sensitivity increases. It has found that the sensitivity increases
with complexity linearly with R
2
value of 0.2874. One can note from Fig.
5.46
that,
the sensitivity of the SVM model is very high and comparable to that of W-SVM,
even though the modelling time and thus complexity are observed less. The highest
value of sensitivity in solar radiation modelling is observed with W-SVM model
when we used the daily data from the Brue catchment, followed by models like
SVM, W-ANFIS, LLR, ANN-LM and AW model.
Based on the information the modeller can tabulate the overall model utility in
terms of complexity, error and sensitivity. Table
5.3
shows different models used in
daily solar radiation modelling and their overall utility indices. The sensitivity and
errors are expressed in relative fraction of the highest values. The valve of the
−
Fig. 5.45 Sensitivity curves
for different data based
models in solar radiation
modelling
