Abstract: The anomaly recognition of monitoring data is the premise and foundation of online monitoring of dam operation safety.It is difficult to achieve efficient and accurate recognition by a single identification method, while the RREW model is easy to miss the data sequence with poor regularity and single step type, and the calculation efficiency is low.To this end, a 1D-VGG data anomaly recognition model based on convolutional neural network was proposed.And the dam safety data anomaly recognition model cluster consisting of model libraries such as statistical regression model, robust regression model, 1D-VGG model and discriminant criteria such as Pauta criterion and MZ criterion was established.Then the matching mechanism between different data types and anomaly recognition models and early warning criteria was constructed.The engineering verification showed that the 1D-VGG data anomaly recognition model had good recognition effect on data sequences with different sequence lengths and different step proportions, and can effectively make up for the shortcomings of traditional regression model and robust regression model.The anomaly recognition model cluster constructed by the above three models and two criteria can realize online accurate and rapid identification of massive data anomalies, and provide reliable data support for online monitoring of dam safety.