Wave Height Prediction in Ocean Observation Based on Improved MVO-GRU-Att Deep Learning Model

Jiuling Dong¹Zehui Li¹Yuanshuo Zheng²Jingtang Luo³Min Zhang¹Xiaolong Yang^1*

• ¹University of Science and Technology Beijing, Beijing, China
• ²Hainan Normal University, Haikou, Hainan Province, China
• ³State Grid Sichuan Economic Research Institute, Chengdu, Sichuan Province, China

Wave height prediction (WHP) is a crucial component of ocean observations. However, due to the strong spatiotemporal correlations and multi-source heterogeneity of ocean observing data, traditional models face inherent limitations in feature extraction and multidimensional information processing, making high-precision prediction a challenging task. To address this issue, this paper proposes a gated recurrent unit (GRU)-based model enhanced with an attention mechanism and multi-verse optimization (MVO), referred to as MVO-GRU-Att, to improve the accuracy and generalization capability of WHP. First, a spatiotemporal feature enhancement strategy is implemented by extracting representative spatiotemporal sequence samples from multi-site buoy sensor data, enabling the model to effectively capture the dynamic characteristics of ocean wave variations. Next, an attention-enhanced GRU optimization is introduced by integrating an attention mechanism (Att) into the GRU network, enhancing the model's ability to focus on critical features and improving its long-sequence modeling capabilities. Finally, hyperparameter optimization with MVO is applied, where the MVO algorithm is utilized to fine-tune GRU hyperparameters, thereby improving the model's generalization performance and adaptability to diverse ocean environments, ultimately enhancing prediction accuracy. To validate the effectiveness of the proposed model, we trained and evaluated it on data from multiple observation stations and conducted error analysis by comparing the predicted results with actual measurements. Experimental results demonstrate that, compared to five traditional methods, the MVO-GRU-Att model achieves a significant reduction in average error metrics (RMSE, MAE, MAPE) by 3.63%-38.01%, while increasing the similarity index R to 98.59%. These findings highlight the model's strong generalization ability across different marine environments and underscore its potential in advancing marine environmental monitoring and early warning systems, providing a more robust and adaptable approach for high-precision wave height prediction.