Prediction of Landslide Displacement Using EMD-PSO-ELM with Multiple Factors

Abstract

This study demonstrates a model for the prediction of active landslide displacement based on the extreme learning machine (ELM) with multiple factors. The particle swarm optimization (PSO) model is selected to optimize the parameters of ELM. Firstly, the landslide displacement sequence which has been monitored is divided into several components developed by the empirical mode decomposition (EMD). Secondly, from the analysis of the basic characteristics of a landslide, this research acquires a series of main influencing factors. Thirdly, each landslide displacement component respectively is predicting by the multi-factor PSO-ELM model. Then, all landslide displacement components are added up as the forecasting result. The model is first trained and then evaluated by using data from a case study of shuping landslide triggered by seasonal rainfall in China. Performance comparisons of EMD-PSO-ELM model with PSO-ELM model are presented. The experimental results illustrate that the multi-factor EMD-PSO-ELM model can efficiently measure the landslide displacement behavior.