Research On The Prediction Of Sea Ice Area Based On Deep Learning

In the context of the era of big data,the difficulty of acquiring remote sensing satellite data is continually decreasing.This data opening provides great convenience in the detection and identification of large areas and difficult to measure environments.The dynamic change of sea ice is a key factor affecting the albedo of the ocean,especially the polar region,and it is also an important part of the global heat exchange system.A large number of studies have shown that there is a significant interaction between ocean and climatic characteristics,so long-term and short-term sea ice prediction is a subject worthy of study.In recent years,relying on more and more data,continuous optimization of hardware equipment and endless algorithm models,deep learning methods have developed vigorously and gradually penetrated into practical applications in various industries.Based on the deep learning method,this paper first analyzes the prediction availability of the sea ice density dataset in different situations,and based on the evaluation results,develops the Long Short-Term Memory(LSTM)in the field of sea ice area prediction.Related research.Finally,based on the relevant knowledge of neural network,the model is improved and optimized accordingly,and a deep learning method suitable for multivariate sea ice area time series prediction is proposed.The main research work of the thesis is as follows:(1)Compare and analyze the three sea ice density data sets of NSIDC,SICCI and BLM,and evaluate the applicability of different data sets according to the latitude,density and segmentation of the Arctic waterway segments,in order to achieve different prediction goals(such as long-term prediction or specific(Regional prediction)Arctic sea ice area change trend discovery provides data selection basis.(2)A deep-learning model training using LSTM and several variants has obtained a prediction model of Arctic sea ice area.The prediction result of this model is much higher than that of the traditional time series model,which also proves that the deep neural network model has advantages in the prediction of sea ice area.(3)The sea ice area time series with different time lengths are used as training data sets,and the multi-step prediction method of the LSTMs model is used to achieve short-term and long-term predictions of sea ice area,respectively,and then give the predictions of several models under different prediction lengths.Accuracy and applicable situation.(4)Taking various factors such as sea ice density,sea surface temperature,and sea surface wind speed as data sources,two prediction models of sea ice area were further discussed.One is based on the data features extracted by CNN,and the LSTM model is used to fit historical information;the other is to introduce attention mechanism to extract significantly fine-grained features,which is convenient for LSTM to capture time dependence more effectively.The experimental results show thatthe prediction results of the two models generated by the improved network training are significantly improved than the prediction results of the original network.