Inversion method of support vector machine in predicting single pile settlement

摘要

There exist certain similarities of environmental parameters and the settlement with variation of the bearing capacity in the same region. Considering of the data set, we increase two variable attributes as the indirect data, and regard the original data as the direct data. With the help of support vector machine (SVM) and independent component analysis, the SVM model in predicting settlement is built by learning from the data of one pile in the same region; meanwhile, the SVM model (Envfea_SVM) which reflects the features of the environment is also built by the inversion SVM. Then, the indirect data of the other pile is input to the Envfea_SVM model, and the corresponding environment feature will be acquired. At last, the feature and the direct data of the other piles are considered as the input, and the SVM model of predicting settlement can be built. The results of the experiments show the Envfea_SVM is effective, and the LearnPre-S model we proposed could predict very precisely.