A knowledge-guided landslide deformation prediction approach based on SVR

Abstract

Predicting the deformation based on landslide multi-mode monitor data is a critical issue of reliable data mining and comprehensive knowledge discovering of landslide for early warning. Due to the complex changes of multi-mode monitoring data and interaction effect caused by geological and geomorphological, hydrological, and anthropogenic factors, most of the deformation prediction methods cannot obtain consistent deformation from multi-mode data. Aiming at this problem, a knowledge guided deformation prediction of landslide Based on SVR is presented, which includes the following aspects: firstly, a sensitivity coefficients is defined which reflects the sensitive degree induced by multiple influencing factors; secondly, the k-means clustering is implemented to discover the mechanism knowledge rules; finally, the deformation is predicted by support vector regression under the guiding of priori rules.; This method trades progressive deformation of landslide as the evolution induced by the variation of precipitation, hydrological conditions and other factors versus time. Both the landslide deformation mechanism and the relationships among different influencing factors are considered in the proposed method. In experiments, typical monitoring datasets including deformation data, rainfall data and the water level change of reservoir in Baishuihe landslide of China are adopted to evaluate the performance of the proposed method.