Abstract: Present dam safety monitoring has shortcomings of weak continuous spatio-temporal monitoring ability and small feed-control range of single measuring point. In view of this, a dam performance evaluation RFM (Recency Frequency Magnitude) model with weakened subjective interference is developed on the basis of fully mining the dam’s prototype monitoring data. First, the concept of “middle type” and “bottom type” monitoring sequence is proposed based on the strong periodicity of dam behavior. Second, K-means clustering algorithm is introduced to classify monitoring sequence adaptively. Finally, the safety evaluation system of dam behavior is established on the basis of defining the project health status represented by various categories in line with the RFM index scoring system. The application of RFM model is illustrated with the horizontal displacement monitoring data of a dam as an example. The project example demonstrate that the evaluation of this model is reasonable and objectively reflects the service state of the dam, and also effectively reduces experiential activities in the evaluation process.

Key words: dam monitoring, time series, RFM model, self-adaption, clustering algorithm, RFM indicator score, deformation