value 1 gave the best predicting performance. Finally, the comparison between these two models indicated that RBFNN yielded a more accurate prediction for both training sets and testing sets. The average MSE, RMSE and correlation coefficient R produced by RBFNN model are 0.0049, 0.0703 and 0.99 in predicting the training sets and 0.0193, 0.1389 and 0.991 for the prediction of testing sets.