Because the construction environment of high-speed railway subgrade is very complex, settlement monitoring data is often unequal time-interval. For Least Squares Support Vector Machine (LS-SVM) has strong nonlinear fitting capability, the equal time-interval settlement time series of subgrade were obtained by using the method of equal time-interval interpolation to a settlement-time relation function which was established by LS-SVM, and then the settlement prediction model of high-speed railway subgrade was established based on LS-SVM. The above prediction model and a combination method of BP neural network and Grey Theory were respectively used to predict subgrade settlement of 5 subgrade settlement monitoring sections at Shangyu north train station of Hangzhou-Ningbo passenger dedicated line. The comparison between prediction results and field test data shows that the accuracy of short time-interval LS-SVM prediction model is much higher and more stable than that of the combination method of BP neural network and Grey Theory, and the extrapolation prediction results of the former are more credible than those of the latter.%高速铁路路基的施工环境复杂,沉降监测数据往往是不等时距的.鉴于最小二乘支持向量机拥有强大的非线性拟合能力,使用最小二乘支持向量机建立沉降与时间的关系函数,以等时间步长插值得到路基的等时距沉降时间序列,建立基于最小二乘支持向量机的高速铁路路基沉降预测模型.分别运用给出的预测模型和BP神经网络与灰色理论联合方法对杭甬铁路客运专线上虞北站5个路基沉降监测断面进行路基沉降预测,并与现场实测数据对比.结果表明,短时距的最小二乘支持向量机预测模型比BP神经网络与灰色理论联合方法的预测精度高,预测结果更稳定,外推预测沉降更可靠.
展开▼
