Wave prediction is crucial for ensuring the safety and disaster mitigation of coastal areas, helping to support marine economic activities. Currently, many deep learning models, such as the temporal convolutional network (TCN), have been applied to wave prediction. In this study, a prediction model based on improved TCN-Attention (ITCN-A) is proposed. This model incorporates improvements in two aspects. Firstly, to address the difficulty of calibrating hyperparameters in traditional TCN models, a whale optimization algorithm (WOA) has been introduced to achieve global optimization of hyperparameters. Secondly, we integrate dynamic ReLU to implement an adaptive activation function. The improved TCN is then combined with the attention mechanism to further enhance the extraction of long-term features of wave height. We conducted experiments using data from three buoy stations with varying water depths and geographical locations, covering prediction lead times ranging from 1 h to 24 h. The results demonstrate that the proposed integrated model reduces the $RMSE$ of prediction by 12.1% and $MAE$ by an 18.6% compared with the long short-term memory (LSTM) model. Consequently, this model effectively improves the accuracy of wave height predictions at different stations, verifying its effectiveness and general applicability.

This study proposes a novel causality-structured matrix long short-term memory (C-mLSTM) model for significant wave height (SWH) forecasting. The framework incorporates a two-stage causal feature selection methodology using cointegration testing and Granger causality testing to identify long-term stable causal relationships among variables. These relationships are embedded within the C-mLSTM architecture, enabling the model to effectively capture both temporal dependencies and causal information within the data. Furthermore, the model integrates Bayesian optimization (BO) and twin delayed deep deterministic policy gradient (TD3) algorithms for synergistic optimization. This combined TD3-BO approach achieves an 11.11% improvement in the mean absolute percentage error (MAPE) on average compared to the base model without optimization. For 1–24 h SWH forecasts, the proposed TD3-BO-C-mLSTM outperforms the benchmark models TD3-BO-LSTM and TD3-BO-mLSTM in prediction accuracy. Finally, a Shapley additive explanations (SHAP) analysis was conducted on the input features of the BO-C-mLSTM model, which reveals interpretability patterns consistent with the two-stage causal feature selection methodology. This research demonstrates that integrating causal modeling with optimization strategies significantly enhances time-series forecasting performance.