Adaptive hybrid machine learning model for forecasting the step-like displacement of reservoir colluvial landslides: a case study in the three Gorges reservoir area, China

摘要

Abstract Establishing an accurate and dependable displacement prediction model is essential to building an early warning system for landslide hazards. This study proposes an adaptive hybrid machine learning model for forecasting the step-like displacements of reservoir colluvial landslides. Firstly, candidate factors are determined based on the landslide deformation response. Then, the cumulative displacement and candidate factors are decomposed using the optimized variational mode decomposition algorithm. Second, the sensitivity analysis of the gray wolf optimizer-based kernel extreme learning machine (GWO-KELM) models to each factor component is analyzed using the PAWN method. Then, the factors are optimized based on the analysis results. Third, based on the optimized factors, GWO-KELM models of different displacement components are established and integrated to predict the cumulative displacement. The Baishuihe landslide was taken as an example. The raw data of its three monitoring sites were employed to verify the performance of the proposed model. The results indicate that the model can decompose the cumulative displacement and factors with the adaptively determined parameters. In addition, the model performed well over a three-year prediction of the landslide displacement.