Ensemble-Empirical-Mode-Decomposition (EEMD) on SWH prediction: The effect of decomposed IMFs, continuous prediction duration, and data-driven models

• Published in: Ocean engineering Vol. 324; p. 120755
• Main Authors: Guo, Yuanye, Si, Jicang, Wang, Yulian, Hanif, Farhan, Li, Shuang, Wu, Mengwei, Xu, Minyi, Mi, Jianchun
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 30.04.2025
• Subjects: Continuous prediction; Ensemble Empirical Mode Decomposition(EEMD); IMF decomposition; Machine learning; Significant wave height(SWH) prediction; Significant wave height(SWH) prediction; Ensemble Empirical Mode Decomposition(EEMD); Continuous prediction; Machine learning; IMF decomposition
• ISSN: 0029-8018
• DOI: 10.1016/j.oceaneng.2025.120755

This paper systematically investigates the impact of key factors on predicting significant wave height (SWH) using Ensemble Empirical Mode Decomposition (EEMD), specifically examining the number of decomposed Intrinsic Mode Functions (IMFs), the duration of continuous predictions, and the selection of data-driven models. Buoy data from Santa Monica Bay, California, is utilized. The findings reveal a significant improvement in prediction accuracy when EEMD is applied before using Support Vector Regression (SVR) and Artificial Neural Network (ANN) model, while improvements with Long Short-Term Memory (LSTM) networks are less pronounced. The study also determines the optimal number of intrinsic mode functions (IMFs) in EEMD decomposition necessary to balance predicting accuracy against computational cost, with RMSE values varying significantly based on the number of IMFs. Furthermore, the analysis indicates that increasing the length of continuous prediction steps leads to significant error accumulation, with the ANN model showing the slowest rate of error increase among the models tested. The results highlight the importance of optimizing model configurations to enhance predictive accuracy while managing computational demands. •This study integrates EEMD with predictive models, analyzing the impact of IMF quantity and prediction length.•EEMD with SVR, ANN, and LSTM enhances SWH prediction accuracy, offering better stability than single models.•Optimal IMF number improves accuracy, while excessive decomposition causes noise, overfitting, and higher costs.•Prediction errors accumulate rapidly after the tenth step, requiring additional variables for long-term stability.