Research On Long-term Significant Wave Height Prediction Based On Marine Multi-Element Correlation

The propagation of ocean waves has a significant impact on various human activities such as sea navigation,maritime military operations,fishing,and offshore construction work.With the increasing daily life and shipping activities in coastal waters,accurate and long-term forecasting of the significant wave height(SWH)has become crucial to ensure the smooth implementation of these activities and to avoid or reduce personnel and economic losses.To date,research on effective wave height prediction methods has mainly been divided into numerical models,traditional time series prediction models,machine learning models,and deep learning models.Numerical models are based on the physical characteristics of wave motion and simulate wave propagation processes by combining wind field data,which involves high computational complexity.Traditional time series prediction models require time series to be stationary or weakly stationary.However,when facing complex and diverse oceanic element time series prediction tasks,their predictive performance is inadequate.Machine learning and deep learning models implement effective wave height prediction based on data-driven methods and have significantly improved predictive performance compared to numerical and traditional time series prediction models.However,existing prediction methods based on deep learning have not fully considered the impact of multiple oceanic elements on SWH,and their ability to capture long-term correlations is insufficient,resulting in low accuracy in long-term SWH prediction.To address these issues,this study proposes two models to solve the aforementioned problems,which mainly involve the following aspects of work:(1)In view of the limitations of current methods for predicting SWH,such as their inability to account for the impact of multiple oceanic elements,failure to capture longterm correlations in time series,and poor long-term prediction accuracy,this paper presents a novel SWH prediction model,called Multi-elements Local and Global Correlation for Wave height Prediction(MLG-SWH).First,using multiple factors such as significance wave height,wind speed as input,a Local-Global Embedding(LGE)module is designed to embed local correlation and periodic information of ocean multielements.Then,an encoder-decoder structure is used to extract the features of ocean wave height,where a casual dilated convolution and self-attention module is designed to effectively capture the global long-term correlation of ocean multi-element sequences and the generative reasoning prediction in the decoder is adopted to avoid errors accumulated in the single-step iterative prediction.Finally,the data of two stations with different characteristics of SWH variation in the North Atlantic are selected for experimental evaluations.Compared with classical time-series forecasting models and mainstream deep learning methods,the MLG-SWH model reaches the lowest mean square error(MSE)and mean absolute error(MAE)in 24 and 48 hours SWH forecasting,having a greater advantage in long-term time-series prediction.(2)In order to more effectively learn the correlation and periodic change patterns among the features of ocean multiple elements and to improve the accuracy of prediction models,GFA-SWH,a SWH prediction model combining the graph structure and the frequency domain attention mechanism is proposed.First of all,a graph representing the complex relationships between multiple marine elements is constructed.It takes different marine elements as "nodes" and the relationships between different marine elements as "edges",which are automatically extracted by using a LSTM network with self-attention mechanism to express the potential time-series dependency.The time-series graph was then fed into a frequency-domain attention module for learning about features.This module performs the following operations: converting the time series of ocean multielement structure from the time domain to the frequency domain using Graph Fourier Transform(GFT);learning complex periodic patterns of the ocean multi-element sequence using the frequency domain attention mechanism;extracting features using Graph convolution;and finally transforming the features from frequency domain to time domain using Inverse Fourier Transform.By stacking multiple the frequency-domain attention modules,a residual network structure is built to effectively extract the potential complex time patterns in the ocean multi-element sequence.Lastly,the Conv LSTM model is used to further capture the spatio-temporal correlation of the sequence,and the multi-step prediction results of SWH are obtained with the full-connection layer.Compared to MLG-SWH and graph-based deep learning methods,GFA-SWH model has the lowest error evaluation index for 24 and 48 hours prediction.Compared to MLGSWH,the prediction time is reduced by 15 seconds and the prediction efficiency is improved by 35%.