A Novel Framework for Significant Wave Height Prediction based on Adaptive Feature Extraction Time-Frequency Network (2505.06688v1)

Abstract: Precise forecasting of significant wave height (Hs) is essential for the development and utilization of wave energy. The challenges in predicting Hs arise from its non-linear and non-stationary characteristics. The combination of decomposition preprocessing and machine learning models have demonstrated significant effectiveness in Hs prediction by extracting data features. However, decomposing the unknown data in the test set can lead to data leakage issues. To simultaneously achieve data feature extraction and prevent data leakage, a novel Adaptive Feature Extraction Time-Frequency Network (AFE-TFNet) is proposed to improve prediction accuracy and stability. It is encoder-decoder rolling framework. The encoder consists of two stages: feature extraction and feature fusion. In the feature extraction stage, global and local frequency domain features are extracted by combining Wavelet Transform (WT) and Fourier Transform (FT), and multi-scale frequency analysis is performed using Inception blocks. In the feature fusion stage, time-domain and frequency-domain features are integrated through dominant harmonic sequence energy weighting (DHSEW). The decoder employed an advanced long short-term memory (LSTM) model. Hourly measured wind speed (Ws), dominant wave period (DPD), average wave period (APD) and Hs from three stations are used as the dataset, and the four metrics are employed to evaluate the forecasting performance. Results show that AFE-TFNet significantly outperforms benchmark methods in terms of prediction accuracy. Feature extraction can significantly improve the prediction accuracy. DHSEW has substantially increased the accuracy of medium-term to long-term forecasting. The prediction accuracy of AFE-TFNet does not demonstrate significant variability with changes of rolling time window size. Overall, AFE-TFNet shows strong potential for handling complex signal forecasting.

Adaptive Feature Extraction Time-Frequency Network for Significant Wave Height Prediction

The paper "A Novel Framework for Significant Wave Height Prediction based on Adaptive Feature Extraction Time-Frequency Network" introduces the Adaptive Feature Extraction Time-Frequency Network (AFE-TFNet), an encoder-decoder rolling framework designed for improved prediction accuracy and stability in forecasting significant wave height (Hs). This framework is constructed to tackle the inherent challenges of Hs prediction due to its non-linear and non-stationary characteristics by simultaneously extracting data features while preventing data leakage.

The encoder of AFE-TFNet comprises two stages: feature extraction and feature fusion. During the feature extraction stage, the model extracts both global and local frequency-domain features by leveraging the combination of Wavelet Transform (WT) and Fourier Transform (FT), utilizing multi-scale frequency analysis with Inception blocks. This dual approach effectively captures both the comprehensive frequency characteristics of ocean signals and the intricate local variations. In the feature fusion stage, the integration of time-domain and frequency-domain features is dynamically managed using Dominant Harmonic Sequence Energy Weighting (DHSEW), which adjusts weights based on the periodicity of the input time series, enhancing the model's adaptation to varying wave signals. The decoder employs an advanced Long Short-Term Memory (LSTM) network to process and predict wave heights from the fused features.

The empirical evaluation involved datasets comprising hourly measured wind speed, dominant wave period, average wave period, and significant wave height from multiple buoy stations, ensuring a diversified test across different geographical contexts. Results decisively indicate that AFE-TFNet exhibits substantial improvements in prediction accuracy over traditional statistical and machine learning methods, particularly for medium-term to long-term forecasts. Furthermore, its prediction accuracy exhibits marginal variability with changes in rolling time window size, affirming the robustness of the model against different temporal granularity.

Key numerical outcomes demonstrate AFE-TFNet's high accuracy across several datasets, significantly outperforming other benchmark models such as NaiveDrift, XGBoost, CatBoost, LightGBM, LSTM, and MWNet. Specifically, the paper reports an average reduction in RMSE and MAE of over 20% and 21%, respectively, compared to the closest competing models, along with improved correlation coefficients.

The implications of AFE-TFNet are twofold. Practically, its adoption could enhance operational precision in marine energy harnessing, ensuring optimized deployment of wave energy converters and improved safety protocols in maritime operations. Theoretically, the model's architectural innovations could stimulate further research into hybrid models combining deep learning with advanced signal processing techniques, particularly in non-linear dynamic environments. Moreover, the paper prompts continued exploration of rolling models in complex time-series forecasting, suggesting the potential for broad applicability across related domains, such as wind speed and other environmental phenomena.

In terms of future directions, the integration of innovative multiscale analysis techniques and other deep learning architectures may serve to fine-tune the predictive capability of AFE-TFNet further. Additionally, the deployment of AFE-TFNet in real-world applications will necessitate comprehensive evaluations concerning the influence of diverse environmental variables on prediction outcomes, underscoring its adaptability and robustness in different scenarios.