LANDSLIDE DISPLACEMENT PREDICTION MODEL BASED ON TIME SERIES AND MIXED KERNEL FUNCTION SA-SVR

This paper puts forward a landslide displacement prediction model. The model is based on time series decomposition and hybrid kernel function SA-SVR. It makes progresses on solving problems of being difficult to quantitatively predict step-type landslide deformation. Firstly,based on the principle of time series decomposition,exponential smoothing is used repeatedly to decompose the cumulative displacement of the landslide into trend displacement and periodic displacement to make the decomposed trend displacement smoother and keep the accuracy of periodic displacement forecast. At the same time,the third-order polynomial of K-flod cross-validation is used to predict the trend displacement avoiding the problems in polynomial prediction. The prediction is easy to overfit and the predicted value deviates from the true value. Based on the properties of SVR kernel function,the mixed kernel function with strong generalization ability and learning ability was constructed as the kernel method of SVR model. The landslide inducing factor is taken as the input vector of SVR model. The simulated annealing(SA)is used to optimize the parameters of SVR model using the mixed kernel function. Thus the SA-SVR model with mixed kernel function is established to predict the periodic displacement. Finally,the trend displacement and periodic displacement are combined to get the predicted total displacement. The paper takes the Baijiabao landslide in the Three Gorges reservoir area as an example,selects the data of ZG325 monitoring point from January 2012 to September 2020,and uses the ZG324 monitoring point as auxiliary verification. The results show that compared with the conventional SVR prediction model,the simulated annealing(SA)performs well in parameter optimization,and the hybrid kernel function is more sensitive to the SVR model,which can greatly improve the prediction accuracy and has high application and promotion value.