Abstract: To address the problems of high frequency of false alarms and low manual intervention in the traditional dam measurement point safety monitoring by using a single model for prediction alarms, this paper proposes a combined model for dam deformation prediction based on weight assignment and modification.A variety of common prediction models and data evaluation indicators are used as the research objects, and the prediction results are processed by introducing the residual correction technique, establishing the prediction model evaluation indicator system and determining the objective and subjective weights of the indicators, and ranking and selecting the models based on the objective and subjective weights.The model weights are modified by real-time feedback of the evaluation indexes and alarm status, and then the dam measurement points are alarmed.The experiment proved that the warning feedback algorithm effectively reduces the frequency of false warnings and warning levels, and makes the results more consistent with the dam's sexual state.