In order to avoid the phenomenon of over fitting and low prediction accuracy of conventional time series model which does not consider the non-linear and random environmental quantity, an integrated empirical mode decomposition(EEMD)method was used to decompose the actual monitoring seepage flow of the water measuring weir. Multiple groups of stable eigenmode functions(IMF)and residual quantities(R)were generated, and then the IMF series and R were fitted and predicted by the correlation vector machine(RVM). Finally, the IMF series and R were added by equal weight to get the predicted value of the seepage flow. The number and length of the training set, the choice of the prediction set number and the treatment of the jump points were also discussed. The results of an engineering application case show that the EEMD-RVM model has high prediction accuracy, and is significantly higher than RVM model and GA-BP model, which verifies the feasibility of the model.