Research On Significant Wave Height Prediction Method Of Ocean Wave Based On Depth Learning

Marine meteorological forecast is of great significance in contemporary marine related projects.Among many marine meteorological parameters,the significant wave height is an indispensable parameter to evaluate the wave energy resources and the critical marine meteorological conditions of marine activities.The prediction of significant wave height is also the key issue of marine meteorological prediction.However,the current significant wave height depth learning prediction algorithm has some problems,such as insufficient prediction accuracy and weak universality for different prediction steps.In this paper,the depth learning algorithm is combined with the wavelet transform method commonly used in the field of signal processing to improve the prediction accuracy of significant wave height.It mainly includes the following aspects:(1)Analyze and preprocess the data set of significant wave height.Firstly,the 2017-2019 data of 46087 wave buoy station provided by NOAA global buoy data center in the United States are cleaned,and the autocorrelation,numerical distribution and seasonal distribution of wave height are statistically analyzed.According to the statistical analysis results,the data of wave related variables are preprocessed by data standardization and wavelet transform algorithm,so as to facilitate the subsequent extraction of related features by neural network.(2)Aiming at the problem that the feature extraction algorithm used in the field of signal processing will lead to a large increase in input sequences,this paper combines the channel attention mechanism,spatial attention mechanism and residual shrinkage mechanism to improve the residual block structure of the ordinary convolution residual network,and proposes an attention residual shrinkage block structure(ARS),The prediction accuracy of significant wave height is compared when attention mechanism and residual shrinkage mechanism are used alone and combined.Experiments show that the use of ARS can significantly improve the prediction results of significant wave height under a long time step.(3)Aiming at the problem that the deep convolution residual neural network is difficult to use the data features extracted from the shallow convolution layer,referring to the Yolo V3multi-scale feature fusion algorithm used in the field of target detection to improve the recognition rate of small targets,this paper proposes a multi-layer feature fusion structure for the time series prediction task,and combines this structure with the attention mechanism module,residual shrinkage module and ARS module respectively,Improve the ability of the network to use the features extracted by each convolution layer.The experimental results show that the multi-layer wave feature fusion model can effectively improve the prediction accuracy of significant wave height.In this study,the improved convolution residual network combines the attention mechanism and residual shrinkage structure to improve the feature extraction ability of the network.By referring to the idea of multi-scale feature fusion in the target detection task,the multi-layer feature fusion algorithm is introduced to improve the utilization ability of the features extracted from each layer of convolution network and significantly improve the prediction accuracy of significant wave height.Based on the improved model,the web platform of significant wave height prediction is deployed to realize the rapid preview and statistical function of significant wave height prediction results.