Data Analysis And Prediction Of Ocean State Parameters Based On Deep Learning

The accurate prediction of ocean state parameters is essential for humans to conduct maritime navigation,maritime military activities,fishery fishing,and maritime construction operations and other related activities.The daily production activities and shipping operations in the Beibu Gulf are increasing.Therefore,timely and accurate wave prediction can effectively guarantee the smooth progress of these activities,and also avoid or reduce the loss of people and economy.Deep learning is currently a popular research field at home and abroad.In this paper,deep learning technology is used as the research method to analyze and predict the ocean parameter data of the Beibu Gulf,and through reasonable construction and training network model,the feasibility of the application of deep learning in the field of Marine science is explored,laying a foundation for future research.The main contents of this paper are summarized as follows:1?The effective wave height in beibu gulf Marine status parameters,flow velocity and flow to the processing and analysis,statistical characteristic value of the corresponding parameter analysis and chart analysis,the results show that the beibu gulf Marine data more obvious seasonal change appears significant wave height,the spread of beibu gulf current direction mainly concentrated in the northeast and southwest,its velocity and flow direction change is sharp,and the change law is not obvious.2?Based on the data collected by single-station buoys in the Beibu Gulf,an effective wave height prediction method based on the LSTM-Res Net(Long Short-Term Memory-Residual Neural Network)model is proposed,First,the wave height data is divided into four models for training in spring,summer,autumn and winter;Then compare the numerical prediction results of the LSTMRes Net model with the numerical calculation results of the LSTM(Long Short-Term Memory)network,the BP(Back Propagation)network and the Res Net(Residual Neural Network)network model.And analyze them in the short-term wave height prediction;The prediction results are comprehensively evaluated using 4 statistical indicators,and the minimum average absolute errors of the 1-hour and 6-hour predictions are 0.0730 meters and 0.1011 meters,respectively.The results show that LSTM-Res Net can achieve a stable prediction effect,with a higher prediction accuracy in 1 hour and a better prediction effect in 6 hours.Finally,the 12-hour,18-hour and 24-hour predictions for the four seasons show that the LSTM-Res Net model has strong long-term predicting capabilities.3?In order to improve the prediction accuracy of the flow velocity and direction of ocean currents,this paper proposes a prediction model based on the combination of Attention Mechanism and GRU(Residual Neural Network)network.The model extracts valuable information from the input through the gated loop unit,and uses the Attention mechanism to highlight the input information that plays a key role in prediction.In addition,a prediction model of ocean current components was established based on the data.In order to verify the effectiveness of the method,it was compared with other five prediction models.The simulation results show that compared with the harmonic analysis method,the Attention-LSTM model,the LSTM model,the GRU model and the BP model,the Attention-GRU model established in this paper has increased the prediction accuracy of the Beibu Gulf flow velocity by 41.9%,6.1%,23.1%,18.2% and 32.9% respectively;the prediction accuracy of the Beibu Gulf flow direction increased by 14.9%,3.5%,6.1%,3.4%and 9.1%respectively.