Research On Spatiotemporal Prediction Of Sea Surface Temperature Based On Deep Learning And Series Decomposition

Sea surface temperature is a key hydrological parameter and basic information of the sea environment,which has an important impact on the climate and ecological environment,so accurate prediction of sea surface temperature has been an active research topic.This thesis takes a sub-region of the East China Sea as the research object,and uses the historical grid data of sea surface temperature and related meteorological variables to conduct spatiotemporal prediction research on the sea surface temperature.Combining the time series decomposition algorithm STL decomposition,channel attention mechanism SE module and spatiotemporal prediction model Conv GRU,this thesis constructs a hybrid model STL-SE-Conv GRU which is suitable for sea surface temperature spatiotemporal prediction.Firstly,the time series corresponding to each grid of sea surface temperature is decomposed by STL to obtain three components:trend,season and residual,and then each component is spliced with meteorological variables and input into the SE module to obtain new feature outputs,which are sent to the Conv GRU model for prediction,and finally the prediction results of each component are summed to obtain the final prediction results.To test the effect of the STL-SE-Conv GRU model,this thesis compares it with various models.By comparing with the ablation model,the effectiveness of STL module and SE module is verified.By comprehensively comparing STL decomposition with a variety of different series decomposition algorithms,it is found that STL decomposition is more suitable for the problem of sea surface temperature series decomposition.By comparing with the single model,it is found that the hybrid model is better than the single model,and the rationality of using Conv GRU as the basic model is proved.In addition,by comparing with other models,it is found that the model constructed in this thesis not only has higher overall prediction accuracy,but also has better spatial stability,that is,the difference in prediction effect of different grid is small.