As the RBF kernel function of the LS-SVM model
utilized in the paper, the regularization parameter γ and the
width coefficient σ play an important role in
establishing a LS-SVM regression model. Generally, the
parameter γ is used to control the function fitting errors.
With a larger value of γ , the better fitting model will be
obtained, also with longer training time. However, if the value
of γ is too large, it will cause an over-fitting problem, and
make the generalization ability degrade. A smaller value of
σ can obtain a better fitting model. However, if σ is too
small, it also causes an over-fitting problem. In this study, the
traditional cross validation method is adopted to determine the
optimal parameters of the model [11]. Considering the 10-flod
cross validation costs more computational time than 5-fold
with the same training data, the 5-fold cross validation is used
to search the optimal parameters. The range of γ and σ 2 is
[100, 1000] and [1, 100], respectively. Therefore, the main
steps of establishing the flooding velocity prediction model
based on LS-SVM can be shown in Fig. 2.
As the RBF kernel function of the LS-SVM model
utilized in the paper, the regularization parameter γ and the
width coefficient σ play an important role in
establishing a LS-SVM regression model. Generally, the
parameter γ is used to control the function fitting errors.
With a larger value of γ , the better fitting model will be
obtained, also with longer training time. However, if the value
of γ is too large, it will cause an over-fitting problem, and
make the generalization ability degrade. A smaller value of
σ can obtain a better fitting model. However, if σ is too
small, it also causes an over-fitting problem. In this study, the
traditional cross validation method is adopted to determine the
optimal parameters of the model [11]. Considering the 10-flod
cross validation costs more computational time than 5-fold
with the same training data, the 5-fold cross validation is used
to search the optimal parameters. The range of γ and σ 2 is
[100, 1000] and [1, 100], respectively. Therefore, the main
steps of establishing the flooding velocity prediction model
based on LS-SVM can be shown in Fig. 2.
การแปล กรุณารอสักครู่..
