STUDY ON DEFORMATION CHARACTERISTICS AND DISPLACEMENT PREDICTION OF RESERVOIRLANDSLIDE – TAKING HONGYANZI LANDSLIDE AS AN EXAMPLE

Reservoir landslide is a common type of geological hazard in the reservoir area, which may cause significant risks to the normal operation of reservoirs and hydropower stations. In this study, the Hongyanzi landslide in the Pubugou reservoir area in the southwest of China was monitored for a long time series, and its deformation characteristics and inducing mechanism were analyzed, and the deformation prediction was carried out by using a variety of models. The main results are as follows:

The water level regulation of Pubugou Reservoir and the deformation rate curve of Hongyanzi landslide are divided in detail, and the response law of landslide deformation under the conditions of reservoir water level fluctuation and rainfall is deeply explored by using the methods of process analysis and quantitative analysis, and the internal mechanism is analyzed. It is found that: (1) the decline of reservoir water level is the main inducement of landslide deformation, and when the decline rate is continuously greater than 0. 5 m/d, the landslide will accelerate to decline; (2) the main rainfall period is inconsistent with the landslide deformation period, and the rainfall has little effect on the landslide; (3) The rapid decline of reservoir water level leads to the increase of water level difference inside and outside the slope, and the increase of seepage pressure, which is the main reason for the increase of landslide deformation.

The PCA-SSA-SVR and PCA-SSA-KELM models are constructed to predict the daily displacement increment of landslides. The comparative experiments show that the MAE of the two models is 1. 126 and 1. 233, the RMSE is 1. 921 and 1. 901, and the R2 is 0. 949 and 0. 950, respectively. Compared with the other 7 unoptimized models, the fitting and generalization abilities of these two models are significantly improved, but there are some limitations in capturing data details and short-term fluctuations. Compared with the PCA-SSA-SVR and PCA-SSA-KELM models, the GRU model has better prediction performance, and its MAE, RMSE and R2 are 0. 924,1.259 and 0. 978, respectively.It successfully captures the details and short-term fluctuations of the data, but it is complex to implement, difficult to tune and has a high time cost.