Here,
the related parameters can be referred to literatures in [14,15].
The comparison results of the LS-SVM model and the
traditional models are shown in Fig. 3. The prediction
performance of traditional models is not stable because the
prediction results of the three types of packings show the
obvious difference. In contrast to that, the LS-SVM gives
better generalization than the traditional models, with all the
RMSE of about 7 %. Generally, the packing constants are
obtained from experiments with specific conditions. Even
with the same type of packing, these constants cannot adapt
themselves to the packings with different diameters, or
different operation conditions in packed towers [4,14].
Consequently, the traditional empirical equations may appear
large fluctuation when they are employed for the prediction of
different packings. Fortunately, the LS-SVM model does not
have these restrictions. The training set includes various types
of packing data, which makes the LS-SVM model acquire
more information during learning. Therefore, the good and
stable prediction of the LS-SVM model can be achieved.