Utilizing VMD and BiGRU to predict the short-term motion of buoys

Buoy, as the primary means for observing ocean elements, the motion mechanism of ocean buoys is mainly influenced by the ocean dynamic environment. Changes in dynamic environments can affect the reliability of platforms and instruments, leading to measurement errors and impacting the operational safety and data quality of ocean observation buoys. Therefore, it is necessary to accurately predict the buoy's motion. Due to the high-frequency, non-stationarity, and volatility of short-term buoy motion, these randomness and fluctuations pose significant challenges to motion prediction. To address this issue, this study innovatively proposes a buoy motion prediction method based on multimodal decomposition. It employs Variational Mode Decomposition (VMD) to decompose the buoy motion signals into multiple frequency scales, reducing the impact of nonlinear signals on prediction. Bidirectional Gated Recurrent Unit (BiGRU) is introduced at each frequency scale to capture the inherent patterns of buoy motion, enhancing data feature extraction capabilities and improving prediction accuracy and robustness. Through extensive experiments and comparisons with different methods, the results show that the VMD-BiGRU model can effectively predict the buoy motion changes in the next 7 s, and outperforms traditional models in buoy motion prediction accuracy with smaller errors. Taking the last step prediction of heave, pitch, and roll in dataset 1 as an example: without VMD, the mean absolute errors are 0.0355 m, 1.1895°, and 1.3870°; after introducing VMD, the mean absolute errors are 0.0135 m, 0.3518°, and 0.4504°, representing improvements of 62%, 70%, and 67.5% respectively. Therefore, this study demonstrates the effectiveness of the proposed VMD-BiGRU model for buoy motion prediction, and the model has potential application prospects in the field of ocean engineering. • Utilizes VMD to decompose buoy attitude data: trend, low-freq & high-freq fluctuation terms. • BiGRU is then employed to learn and capture the temporal feature dependencies of each mode after decomposition. • Carry out data reconstruction to achieve accurate prediction of buoy attitude changes in the next half wave period. • Short prediction time ensures real-time application needs, delivering prompt & efficient results.