To achieve better results in predicting the vibration trend of hydro-electric generating unit, a novel trend prediction model based on optimal variational mode decomposition（OVMD）and support vector regression（SVR）was proposed. Firstly, center frequency observation method and residual minimization criterion were employed to determine the parameters of OVMD, and the non-stationary vibration series were decomposed into a set of mode functions, after which the state matrix corresponding to each mode was obtained with phase space reconstruction. Then the inputs and outputs of SVR models were deduced. Each SVR model was trained and tested with grid search based on cross validation. Finally, prediction values of the original vibration series were calculated with the accumulation of outputs from all the SVR models. The successful application in predicting the vibration trend for a large mixed-flow hydro-electric generating unit, as well as comparative analysis with other methods, attests the effectiveness of the proposed model.