Wu, Lin and An, Yi and Qin, Pan and Hu, Huosheng (2025) Prediction of significant wave height based on feature decomposition and enhancement. Expert Systems with Applications, 277. p. 127255. DOI https://doi.org/10.1016/j.eswa.2025.127255 (In Press)
Wu, Lin and An, Yi and Qin, Pan and Hu, Huosheng (2025) Prediction of significant wave height based on feature decomposition and enhancement. Expert Systems with Applications, 277. p. 127255. DOI https://doi.org/10.1016/j.eswa.2025.127255 (In Press)
Wu, Lin and An, Yi and Qin, Pan and Hu, Huosheng (2025) Prediction of significant wave height based on feature decomposition and enhancement. Expert Systems with Applications, 277. p. 127255. DOI https://doi.org/10.1016/j.eswa.2025.127255 (In Press)
Abstract
Predicting significant wave height (SWH) are crucial for maritime activities, including offshore operations, ship navigation, and meteorological forecasting. However, the complexity, non-stationarity, and distribution shifts of SWH result in relatively low prediction accuracy. Additionally, the inadequate use of local information in many prediction models further hinders accuracy improvements. To solve these problems, this paper proposes a novel multimodal feature enhancement transformer (MFET) method for SWH prediction. The method primarily consists of a signal decomposition module, an encoder stack, and a decoder stack. The signal decomposition module uses the sparrow search algorithm-variational mode decomposition (SSA-VMD) method to optimally decompose SWH signals. The decomposed signals are combined with wave features to form 3D data, which is then input into the encoder and decoder stacks for prediction. Each stack contains six encoders and six decoders respectively. Each encoder comprises a squeeze-and-excitation (SE) attention module, a multi-head convolutional attention (MHCA) module, and a multi-layer perceptron (MLP) module, while each decoder includes a multi-head self-attention (MHSA) module, a cross-attention (CA) module, and an MLP module. The SE attention mechanism dynamically adjusts the influence of each channel by selectively enhancing or suppressing their contributions. A parallel convolution layer is proposed in MHCA to effectively capture local wave feature within each channel. Furthermore, the reversible instance normalization (RevIN) method is used to eliminate distribution shifts. The MFET improves prediction accuracy by optimally decomposing the SWH signal, dynamically enhancing channel information, and extracting local features in parallel. Experimental results show that MFET achieves MSE of 0.0062, 0.0019, and 0.0073, along with R² of 98.51%, 98.93%, and 95.09% on the three datasets. Code is available at this repository: https://github.com/wulin777/SWH-Prediction
Item Type: | Article |
---|---|
Uncontrolled Keywords: | Prediction of significant wave height; Sparrow search algorithm-variational mode decomposition; Squeeze-excitation attention; Parallel convolution |
Subjects: | Z Bibliography. Library Science. Information Resources > ZR Rights Retention |
Divisions: | Faculty of Science and Health > Computer Science and Electronic Engineering, School of |
SWORD Depositor: | Unnamed user with email elements@essex.ac.uk |
Depositing User: | Unnamed user with email elements@essex.ac.uk |
Date Deposited: | 01 Apr 2025 14:55 |
Last Modified: | 01 Apr 2025 15:30 |
URI: | http://repository.essex.ac.uk/id/eprint/40575 |
Available files
Filename: JESA-V277-2025-127255.pdf
Licence: Creative Commons: Attribution 4.0