A reconstruction-based VMD-GWO-LSTM model for significant wave height prediction

Irregular sea states pose significant risks to maritime navigation and human safety, highlighting the need for accurate wave height prediction models. To address the limitations of conventional methods, this study proposes a reconstruction-based VMD–GWO–LSTM model for predicting significant wave height (SWH) in the South China Sea. The Grey Wolf Optimization (GWO) algorithm is first employed to adaptively determine the parameters of Variational Mode Decomposition (VMD). The decomposed intrinsic mode functions (IMFs) are then evaluated using envelope entropy and sample entropy and reconstructed into three representative feature sequences. These reconstructed features are used as inputs to a Long Short-Term Memory (LSTM) network, whose hyperparameters are further optimized via GWO. The methodological contribution lies in the coordinated integration of entropy-guided IMF reconstruction, sequential GWO-based parameter optimization, and LSTM forecasting. Experimental results on the investigated South China Sea ERA5-derived dataset show that the proposed model achieves an R2 of 0.9566 and reduces MAE and RMSE by 7.6% and 7.3%, respectively, relative to the selected GWO–LSTM benchmark. These reductions indicate improved overall error control and may provide more accurate information for subsequent sea-state assessment. The present results should be understood as case-study evidence for the investigated dataset, while broader validation is required to determine the applicability of the framework under other wave climates and operational conditions.