Abstract: The existence of abnormal values in dam observation data tends to reduce its reliability,and is bound to lower the training and prediction accuracy of the displacement monitoring model.In order to solve such problems,a time series prediction method for concrete dam displacement based on reliability identification and excavation of observation data is proposed in this paper.By comparing and analyzing the disposal effect of abnormal data in displacement monitoring,this research proves the necessity of identifying abnormal values within dam observation time series and effectively interpolating them.For this reason,variational mode decomposition algorithm is introduced to decompose the processed time series into several feature-protruding subsequences,and the extracted time series are reconstructed after obtaining the replacement value of lacking data by training the bidirectional long short-term memory network-based model.The deep learning-based interpolation model for concrete dam observation data is constructed,and then the displacement prediction model of concrete dam based on defective observation time series data is established.Taking a concrete gravity dam as an example,this method is utilized to identify the lacking data and correlations of observation time series,and then interpolate the anomalies,so that the problem of interpolating the missing data can be transformed into the problem of predicting the missing values.The comparative analysis shows that this method has strong robustness and high-precision on data missing interpolation and forecasting ability.