Abstract: Due to many factors affecting the landslide，it is difficult to create an accurate kinetic model， and also it is hard to predict the short-term deformation of large-scale slope by using the traditional dynamic model to meet the demand for high-precision early warning of large-scale slopes. The short-term deformation prediction model for large-scale slope was established under the cooperation of regression analysis and robust Kalman filter. Fitting value was adopted to replace the data containing gross errors for filtering and prediction operation， which solves the problem of Kalman filter lacking for anti-interference to gross errors. The engineering case based on monitoring data of large-scale slope in Jinduicheng Open-pit Molybdenum Mine showed that both of the prediction models are effective， but the precision and robustness of the short-term deformation prediction model of large-scale slope under the cooperation of regression analysis and robust Kalman filter are better than the traditional robust Kalman filter model.

Key words: Large-scale slope, Deformation forecast, Gross error, Regression analysis, Robust Kalman filter