Subgrade settlement prediction based on Support Vector Machine

Due to traditional ballastless track settlement prediction algorithms have large error and can´t accurately forecast settlement after work, a new method using Support Vector Machine (SVM) to forecast ballastless track settlement of high-speed railway is proposed in this paper. Firstly, build a SVM model and calculate the dual model. Then, mapping it to a higher dimension space by kernel function. At last solve and validate the model by an example. By comparing with the traditional forecasting algorithms and BP neural network, the results show that SVM can obtain high prediction precision and good generalization capability in few training samples comparing to other algorithms, provide a more secure and reliable solution for ballastless track settlement.