Intelligent prediction of landslide displacements based on optimized empirical mode decomposition and K-Mean clustering

According to the deformation characteristics of step-like landslides in the Three Gorges Reservoir area, a new method for predicting the landslide displacement is proposed. The monitoring displacements of points ZG118 and XD-01 in Baishuihe landslide are taken as example analysis. By using the empirical mode decomposition with soft screening stop criteria (SSSC-EMD), the cumulative displacement-time curves and the influencing factor time series are adaptively decomposed into multiple intrinsic mode functions (IMF). The K-Means clustering method is adopted to cluster and accumulate IMFs. The displacement components (including the trend, periodic and stochastic displacements) and the influence factor components (including high-frequency and low-frequency factors) which contain physical meanings are obtained. The trend displacements are fitted by the least square method. The periodic and stochastic displacements are predicted by combating fruit fly optimization and least squares support vector machines (FOA-LSSVM) model. Finally, the cumulative prediction displacement is found to be the addition of the three component prediction values. The results show that the proposed (SSSC-EMD)-K-Means-(FOA-LSSVM) model has the capability of predicting the displacement variation of step-like landslides. The prediction accuracy of this model is higher than those of traditional SVR and LSSVM models. Furthermore, the single factor analysis is performed by changing the length of the training, and it is positively correlated with the prediction accuracy.