An adaptive variational modal decomposition(AVMD) combined with kernel extreme learning machine(KELM) is used to predict the vibration of the sluice in the discharge process, which is used to assist decision-making and early warning. Firstly, the decomposition modal number of AVMD is determined based on mutual information criterion to overcome the disadvantage of blindly selecting decomposition parameters of variatronal modal decomposition (VMD). AVMD is used to decompose on-line monitoring vibration sequence of sluice into several IMFs, which is used as input and output of KELM model.Secondly, each component is predicted separately by KELM, the hidden layer does not need to be artificially set and the output weights are calculated using the least squares method. Finally, the prediction results of IMFs corresponding to each measurement point are reconstructed as the final predicted value. Combined with the on-line monitoring data of Yucao Sluice under the condition of free discharge, the AVMD-KELM and KELM models and SVM models are used to predict the vibration trend, and the forecast results are compared and analyzed. The results show that the predicted results of AVMD-KELM modal are closer to the measured values, the calculation speed is faster, the accuracy is higher, and the error is smaller. The method can effectively predict the vibration trend of sluice structure.