Aiming at the problem of gross errors and abnormal measurements in the deformation monitoring data of concrete dams, a data anomaly identification and reconstruction model is proposed. The association rules are used to quantify the correlation between deformation sequences and water level sequences, and the monitoring data are input into the DBSCAN clustering algorithm to find the abnormal points.The association results are used to classify the data abnormal points into two categories, coarse error points and points reflecting the dam morphology.The points reflecting the dam morphology are retained and the coarse error points are eliminated, and modified wavelet neural network is used to reconstruct the coarse difference data to ensure the integrity of the monitoring sequence. The application results of an arch dam deformation monitoring data show that the model can accurately identify the abnormal values in the monitoring data and can obtain more accurate reconstructed data, providing a new analysis method for the evaluation of measured properties of a dam.