A Landslide Displacement Prediction Method of Particle Swarm Optimization Combined with Support Vector Machine Regression Based on Recursive Feature Elimination Selection

Affected by the seasonal precipitation and changes in water levels of reservoirs, parts of mountains will generate landslides in which the cumulative displacement-time curve of some landslides exhibits obvious “step type” dynamic deformation characteristics. In the light of that displacement, a landslide displacement prediction model was proposed, which was combined the particle swarm optimization (PSO) algorithm with the support vector regression (SVR) based on the recursive feature elimination (RFE) algorithm. and the research was performed combined with the Xinpu landslide. Firstly, the methods of outlier removal and missing value filling for landslide displacement data were explored. Outlier rejection was adopted a median-based method with ensemble empirical mode decomposition, and missing value filling was adopted a statistical variable-based method. Then, the “step type” cumulative displacement of landslides was split into the trend and period terms by the exponential smoothing method. The displacement of the trend term was fitted by Fourier curve. For the displacement of the periodic term, the influence factors with high correlation with the displacement of the periodic term were selected based on the SVR-RFE method, and the prediction model for the displacement of the periodic term was established. Subsequently, the prediction model parameters were optimized by the PSO algorithm. Finally, the predicted displacement of the period term was superimposed on the predicted displacement of the trend term to obtain the final predicted cumulative displacement of landslides, in which the goodness of fit was 0.999, the root mean square error was 9.974 mm, and the average absolute error was 7.037 mm. Comparing the proposed model with the GSCV-SVR model and the GA-SVR model, the proposed model was of a strong predictive capability for sudden displacement and was suitable for risk warning during periods of accelerated displacement changes of “step type” displacement of landslides.