Abstract: To address the issue of weak mechanical interpretation in the time-series decomposition model of step-type landslide displacement, we propose a decomposition method incorporating sliding R_nl step-point detection and improved weighted moving average method to modify step-term displacement. Both the Nishihara creep constitutive model and a self-adaptive improved genetic algorithm model were utilized. The proposed method was applied to decompose the displacement time series of Baishuihe landslide. The results of the proposed method were compared with those of the MK Test, sliding t test, and the Bayes test, demonstrating that the sliding R_nl step-point detection yields more accurate and applicable results. Furthermore, the displacement time series decomposition results were also compared with those obtained from quadratic moving average time series decomposition, cubic exponential smoothing time series decomposition, and VMD time series decomposition. The findings reveal that our proposed decomposition method effectively addresses irregular displacement and enhances the mechanical interpretation of the landslide trend term. Additionally, the introduction of the most critical step-term displacement in landslide displacement prediction enhances the specificity of analysis and prediction. In conclusion, our decomposition model holds significant engineering value and serves as a valuable reference for time series prediction.

Key words: landslide displacement, time series decomposition, step term displacement, creep model, genetic algorithm, slide test