Prediction model oriented for landslide displacement with step-like curve by applying ensemble empirical mode decomposition and the PSO-ELM method

Abstract

For landslides characterized with “step-like” deformation curves, the accelerations of the deformation during the rainy season are destructive for both residents and infrastructure; therefore, it is essential to perform displacement prediction. The aim of this study is to present a computational intelligence approach that adopts ensemble empirical mode decomposition (EEMD) and extreme learning machine (ELM) method optimized by particle swarm optimization (PSO) to conduct displacement prediction. First, the cumulative displacement was decomposed by the EEMD method to obtain the trend and periodic components. The trend displacement was predicted by setting previous displacement data as an input variable. The external triggering factors were also decomposed by EEMD into several subsequences. Subsequences with periodic characteristics were selected as the input datasets to forecast the periodic displacements using an ELM model optimized by PSO (PSO-ELM). Finally, the total displacement was obtained by adding the two predictive components to validate the model performance. The Baishuihe landslide in the Three-Gorges area of China was selected as an example; long-term monitoring records from monitoring site ZG118 were utilized to validate the model. The results revealed that the prediction accuracy can be improved by deleting any random components in the triggering factors. The correlation coefficient and the root mean square error between the measured and predicted displacements were 0.996 and 7.62 mm, respectively, thus indicating satisfactory calculation accuracy for the trained model. Therefore, under the premise of available monitoring data, the PSO-ELM model was effective in forecasting landslide displacements with step-like curve in this region.