| Wave height is an important parameter in the study of the ocean.By predicting the wave height,the situation of sea wind and wave can be obtained in advance,which can help Marine navigation,coastal residents and Marine operations and reduce the loss caused by accidental disasters.Therefore,high precision prediction of wave height is of great practical significance for ocean exploration.The traditional numerical model in the early period required huge computational work and complex derivation.Classical time series model is not good for non-stationary time series prediction.The deep learnbased method is gradually applied to ocean wave height prediction,but the prediction effect of a single model is limited.In order to further improve the prediction effect,researchers began to add decomposition algorithm and optimization algorithm,and use the deep learning timing prediction hybrid model for prediction to improve the accuracy of prediction.On the basis of existing research and data from NOAA sites,this paper proposes two improved models to improve the prediction accuracy.The main work is summarized as follows:First of all,the research background and significance of sea wave height prediction are expounded.The development and research process of sea wave height prediction from the early numerical model to the current deep learning model are described.The evaluation indexes used in the model are also explained.The algorithm used in the model is introduced.Secondly,at present,many researchers directly use the deep learning model to analyze and forecast the original sea wave height.Due to its own structural limitations,the network model has its own limitations and has certain errors with the real value.This paper analyzes and mines the error sequence,takes the long and short term memory network as the prediction framework,and on this basis,reduces the dimension and integrates the time series data.By forecasting the time series data of wave height,this research can further improve the prediction accuracy of wave height.For the method of direct fusion of decomposed modal components by existing researchers,Spearman is used in this paper to analyze the correlation between error modal component and error sequence,and the error modal component is classified according to the strength of correlation,and LSTM is used to predict the error modal component,so as to generate the error value of future point position.Finally,the predictive value is modified by the fusion of correlation and weight.On this basis,the decomposed modal components are classified by clustering method.After the dimension of the modal components is reduced,the correlation analysis is used to analyze the correlation between the fused modal components and the historical error order.Finally,weighted error compensation is carried out to correct the predicted value of LSTM’s future point wave height according to the correlation coefficient.Finally,the LSTM wave height prediction model with error correction proposed in this paper is analyzed by ablation experiment,and the optimal model of the retained mode component is analyzed,and compared with other models,which shows that the proposed model is effective.The sea wave height prediction model using the nearest neighbor propagation algorithm is compared with other models.It is proved that the improved model has obvious advantages in the accuracy and prediction effect of the sea wave height prediction.Finally,the two models are compared,and the experiment shows that the effect of usingcorrelation strength for classification is slightly better than that of the nearest neighbor propagation algorithm model for clustering.However,the model complexity and calculation amount are higher when the correlation strength classification is used,while the model complexity and data dimension are simplified when the clustering dimension reduction is used,and the accuracy is still good,which is suitable for the deployment and application of lightweight and small platforms. |
PDF Full Text Request