Abstract: The influence factors of concrete dam deformation are complex， and monitoring data contain a lot of uncertain information， so forecasting models with different modeling methods have various prediction effects. In view of this situation， a combined forecasting model can make up the limitation of a single model. In the modeling process， how weights could reflect the changes of monitoring information in real-time will be the key to the modeling method for combination mode. According to the optimum theory， considering the interaction between unknown quantities and weight coefficients of the individual model， by use of the distributed global optimization ability of shuffled frog leaping algorithm (SFLA)， an optimal modeling method with SFLA is established. Long-term monitoring data of concrete dam deformation indicate that this method has good forecasting effect， simplifis the modeling style of a combined model with variable weights， improves long-term forecasting ability of the model， and provides a new approach to the analysis of concrete dam deformation.