Irregular wave prediction of hybrid prediction model based on adaptive variable weight

Because ocean surface waves are inherently random and nonlinear, accurate forecasting—crucial for safe maritime transportation and offshore operations—benefits from statistical analysis. This study proposes an adaptive, variable weight deep learning model for instantaneous irregular wave prediction to address the challenges posed by wave nonlinearity. We generated irregular waves using the JONSWAP spectrum and applied outlier correction and standardization to construct training, validation, and test sets. Because hyperparameters strongly affect predictive performance, we tuned them via Bayesian optimization. Furthermore, our hybrid approach integrates Neural Basis Expansion Analysis for Time Series (N-BEATS) and Neural Hierarchical Interpolation for Time Series (N-HITS), with weights optimized by a novel Smart Guidance Optimization (SGO) algorithm. Compared with baseline models, the proposed hybrid model achieves higher predictive accuracy, sustains performance over extended prediction length, and reduces training time relative to TCN and CNN–LSTM baselines. The study also outlines directions for future optimization to guide further research in wave forecasting.