Abstract: The deformation monitoring sequence of concrete dam has abnormal mutation due to uncertain factors such as failure of observation facilities or manual collection, which increases the difficulty of subsequent analysis and comprehensive evaluation. Therefore, on the basis of empirical wavelet transform (EWT) theory and local outlier factor (LOF) recognition mechanism, the boxplot method is introduced to redefine the judgment threshold of the detection model, which effectively avoids many interference factors caused by human subjective setting. Based on this, an outlier diagnosis and treatment system is constructed to objectively restore the characteristics of monitoring information. The engineering example shows that the proposed scheme has accurate and excellent signal recognition ability of outlier subset and complete theoretical support, which is suitable for the early optimization of the deformation monitoring signal of concrete dam driven by complex internal and external multi-source environment.