Abstract: A method of landslide displacement prediction based on least squares support vector machine (LSSVM) is proposed to overcome the shortcomings of traditional prediction for Dahua landslide in Lancang River basin as a case study. The original sequence is decomposed and reconstructed into trend item and fluctuation item based on time series analysis and ensembled empirical mode decomposition (EEMD). The displacement of trend item is affected by internal factors of landslide,and the least square method and polynomial equation are used for fitting prediction. The displacement of fluctuation item is affected by periodic factors such as reservoir water level,rainfall and groundwater level. The grey correlation degree method and kernel principal component analysis (KPCA) are combined to screen and reduce the dimension of input factors,and then PSO-LSSVM coupling model is used for predicting the fluctuation item. Finally,the cumulative predicted displacement is obtained by adding the predicted displacement of trend item and fluctuation item,and the model prediction accuracy is quantitatively analyzed. Results demonstrate that the EEMD-KPCA-PSO-LSSVM combined model has a sound prediction performance with higher prediction accuracy than traditional BP neural network,LSSVM and other single models.

Key words: water-induced landslide, displacement prediction, EEMD, KPCA, LSSVM