Abstract: Vibration signal of hydropower unit is an important content of health state evaluation and deterioration early warning. An accurate prediction of unit vibration change trend can improve the safety and reliability of unit operation. In view of the problem that single model is difficult to obtain the optimal prediction results,a CNN-BiGRU combination model vibration prediction method is proposed. Firstly,the convolutional neural network（CNN）is used to extract the local features of the data,and then the CNN-BiGRU combined prediction model is constructed in parallel with the bidirectional gated cyclic unit（BiGRU）network. The model aims to improve the prediction accuracy and universality by combining CNN’s ability to adaptively extract local information with BiGRU’s time series prediction advantages. Finally,the prediction of axial vibration peak value of a domestic hydropower station unit is studied. The experimental results show that the proposed combined model can effectively predict the variation trend of unit vibration,and provide a new idea for vibration prediction of hydropower units.