Abstract: In view of the monotony and nonlinearity of landslide displacement time series, the landslide displacement is decomposed into trend item and deviation item which are dynamically predicted by curvilinear regression-BP neural network model. The Shuping landslide in the Three Gorges reservoir area is taken as a case study. In this method, the trend item of displacement time series is extracted by curvilinear regression model and the deviation of curvilinear regression model is approximated by BP neural network model based on factors which influence displacement fluctuations. Then the prediction values of trend displacement and deviation displacement are superposed to obtain the total displacement prediction value. The results indicate that the prediction model can reflect the key role of dynamic change of impact factors in the displacement development. The average relative error of the prediction is 3.3%, indicating that the model can effectively improve the precision of prediction results.

Key words: landslide, displacement prediction, impact factors, curvilinear regression, neural network