Abstract: There are often abnormal measurements in the original observation sequence of dam safety monitoring, which greatly affects the reliability and accuracy of dam safety monitoring data analysis.Therefore, based on the analysis of the abnormal values characteristics and the advantages and disadvantages of traditional anomaly detection methods, this paper studied the detection methods of abnormal values in monitoring data from the local and overall perspectives.Firstly, aiming at the defects of multiple local anomaly coefficient methods requiring data with long sequence and equal time interval, a local change anomaly coefficient method(LV)and a collaborative discrimination strategy of local method and overall method were proposed.Furthermore, the density clustering algorithm(DBSCAN)was introduced, and a LV-DBSCAN anomaly detection method considering the overall and local characteristics of the data was proposed.Taking the downstream displacement monitoring data of two vertical measuring points of a concrete gravity dam as an example, the detection accuracy of different methods on different types of data sets was compared and analyzed.The results showed that the LV-DBSCAN method proposed in this paper has wider applicability, higher accuracy and lower misjudgment rate.