Application of a novel signal decomposition prediction model in minute sea level prediction. (15th September 2022)

• Record Type: Journal Article
• Authors: Song, Chao; Chen, Xiaohong; Xia, Wenjun; Ding, Xinjun; Xu, Chuang
• Is Part Of: Ocean engineering. Volume 260(2022)
• Publication Date: 2022-09-15
• Subjects: Signal decomposition prediction model -- Elman neural network -- Minute sea level -- Different time series lengths -- Robustness; Ocean engineering -- Periodicals; Ocean engineering; Periodicals; 620.4162
• DOI: 10.1016/j.oceaneng.2022.111961
• Languages: English
• ISSNs: 0029-8018

Abstract

Abstract: Although the prediction of hourly sea level and monthly mean sea level has been widely discussed, the prediction of minute sea level (MSL) has rarely been attempted under the background of human activities and climate change. The prediction of MSL is of great significance for real-time sea level measurements and tsunami early warnings. To improve the prediction accuracy of MSL, we propose several signal decomposition prediction models: the TVF-EMD-ENN model constructed using time varying filtering based empirical mode decomposition (TVF-EMD) and the Elman neural network (ENN); the WT-ENN model constructed using wavelet transform (WT) and ENN; and the CEEMD-ENN model constructed using complementary ensemble empirical mode decomposition (CEEMD) and ENN. These models first decompose the MSL into several subcomponents using different signal decomposition methods (i.e., TVF-EMD, WT, and CEEMD), and then the subcomponents are predicted by ENN. Lastly, the predicted values from the subcomponents are compiled to obtain the predicted MSL values. We applied these models to stations in six different countries, and the results showed that the TVF-EMD-ENN model was the most robust and had the best predictive performance. The prediction performances of WT-ENN and CEEMD-ENN models were slightly inferior to TVF-EMD-ENN. When the MSL sequence length was shortened, the TVF-EMD-ENN still performed best and achieved relatively robust prediction performance. The prediction performance of the ENN model was unstable with large differences among stations. Our study emphasizes the superiority of signal decomposition prediction models in solving MSL series predictions, and provides a new method and an important reference for solving non-stationary MSL prediction problems. Highlights: We first applied TVF-EMD-ENN model to minute sea level (MSL) prediction. We compared the prediction performances of decomposition prediction models constructed with different decomposition methods in MSL. We assessed the prediction performance of different decomposition prediction models when applied to sequences of different lengths. Our research emphasized the superiority of the signal decomposition prediction model in solving MSL predictions.