Prediction of landslide sharp increase displacement by SVM with considering hysteresis of groundwater change

The displacement-time curves of reservoir landslide mostly show step-by-step growth characteristics. The sharp increase of the displacement plays a significant role in the evolution process of the step-like landslide, so it is extremely important to accurately predict the suddenly change displacement of landslide. In order to overcome the shortcomings existed in the current displacement monitoring methods and prediction models for the mutation displacement of landslide, this paper proposed a hybrid machine learning displacement prediction model based on Support Vector Machine, including Support Vector Classification(SVC) and Support Vector Regression (SVR), optimized by Particle Swarm Optimization (SVC-PSO-SVR) and considering the hysteresis of groundwater level change. The Majiagou No. 1 Landslide in the Three Gorges Reservoir (TGR) Area, whose deformation shows typical step-by-step tendency, was illustrated. Base on the deep displacement data of Majiagou landslide from January 2016 to December 2017, the landslide deformation pattern can be divided into two states: stability and acceleration. Firstly, the SVC model was adopted to predict the time range of acceleration state. Then, considering the lag fluctuation of groundwater level, the concept of equivalent reservoir level was proposed. Finally, based on SVC model, the sharp increase displacement of landslide was predicted using PSO-SVR model. The proposed SVC-PSO-SVR model yielded the root mean square error (RMSE) of 0.08827 mm and the mean absolute percentage error (MAPE) of 0.02105 mm. Besides, the prediction accuracy of SVC can attain 96.49% (55 / 57). The results show that the model can accurately predict the time range and displacement of acceleration deformation section of Majiagou landslide. The proposed model is of great significance for landslide prediction and early warning.

Automated summary

This paper proposed a hybrid machine learning displacement prediction model based on Support Vector Machine, optimized by Particle Swarm Optimization and considering the hysteresis of groundwater level change, to accurately predict the sudden change displacement of landslide. The model showed great significance for landslide prediction and early warning by achieving high prediction accuracy and effectively capturing the deformation pattern of landslides.