Abstract: Kalman filter is widely applied to dam deformation prediction.However,the identification of parameters to the model,especially the state and observation noise covariance matrices,is derived mostly from the experience of engineering or expert knowledge.Therefore,a self-learning method was proposed for parameter identifying,in which the parameters of Kalman filter are determined by the combination of Monte Carlo and rejection sampling algorithm from history data.More precisely,the state noise sorted out from training ones is evaluated by samples,whose observations approximate actual value completely,and the observation noise is determined by calculating the difference of the aforementioned noise and overall error.The experiment result shows that the proposed method is more accurate than other congener ones,and it’s more applicable to dam deformation prediction.

Key words: Monte Carlo,Rejection sampling,Kalman filter,Parameter self-learning,Dam deformation prediction