To improve the accuracy of tidal prediction further enhance the adaptability of the prediction model, and address a series of problems, including the difficulties of intelligent and adaptive extraction of low-frequency tidal components, weak ability to dynamically process tidal information, limitations of a single prediction model for overall tidal prediction, this paper proposes an improving tidal prediction model based on adaptive optimal variational modal decomposition of energy entropy and GRU recurrent neural networks. Firstly, the tidal data are normalized, and the VMD method is utilized for adaptive variational modal decomposition. Then, the optimal decomposition level is confirmed based on the energy entropy of the components. Finally, each component of the optimal decomposition is standardized and separately predicted and synthesized by GRU. The final prediction data are formed through reverse normalization. Through verification and analysis, compared with LSTM and BiLSTM models, the GRU model has better performance in terms of tidal prediction. The RMSE values are increased by 53% and 96. 8% , respectively. However, compared with a single GRU model, the proposed prediction model has RMSE increase 81. 3% again, and the accuracy improvement effect is more obvious. The method in this paper has high promotion and application value for tidal analysis and prediction.