A Method For Predicting Sea Temperature In The East China Sea Based On Deep Learnin

Sea Surface Temperature(SST)is an important parameter for oceanography.The East China Sea contains abundant oil and fishery resources and is not only an important sea trade route but also a crucial marine strategic location in China.Therefore,accurate prediction of SST in the East China Sea is of significant importance.The nonlinearity,volatility,and randomness of the SST sequence pose great challenges for accurate prediction.Deep learning has been widely applied to SST prediction in recent years due to its powerful analytical ability.However,existing deep models not only suffer from limited prediction accuracy,low operating efficiency,and high model complexity but also cannot overcome the impact of noise on prediction results.Therefore,this paper introduces the Variational Mode Decomposition(VMD),the Broad Learning System(BLS),and the Error Correction method to address these problems.The main contents and innovations of this paper are as follows:(1)To reduce the influence of noise,the VMD method is used to decompose the SST sequence into sub-sequences of different frequencies,thereby reducing the prediction errors caused by non-stationarity of the data.The Convolutional Neural Network and Long-Short Term Memory Network(CNN-LSTM)model is used for multi-resolution analysis to extract the temporal features of the sequence.Finally,the predictions for each component are output through the fully connected layer(Dense)and then summed up to obtain the final prediction.A SST prediction model based on VMD-CNN-LSTM is proposed.The model is validated in the East China Sea,and the Mean Absolute Error and Root Mean Square Error values of the model are 0.1261 and 0.1607,respectively.The prediction accuracy is significantly better than that of other models,validating the effectiveness of VMD.(2)To improve the operating speed and mitigate the prediction lag in deep learning,transfer learning is used to combine LSTM and BLS.LSTM is used as the feature mapping node of BLS,and BLS is used as the predictor instead of the traditional Dense in LSTM.A SST prediction model based on VMD-LSTM-BLS is proposed.The model is validated in the East China Sea,and the MAE and RMSE values of the model are 0.0727 and 0.0957,respectively.The prediction accuracy is further improved,demonstrating the effectiveness of BLS.(3)To improve the error amplification caused by modal accumulation and further improve the prediction accuracy,the error sequence was learned and predicted,and the error prediction result is fed back to the prediction value to correct the model error.A SST prediction method based on post-processing Error Correction is proposed,and two Error Correction model structures ar proposed for different correction targets.One is the prediction value Error Correction model structure,and the other is the component Error Correction model structure.The corrected model is validated in the East China Sea,and the MAE and RMSE values of the corrected model are 0.0184 and 0.0245,respectively.The fitting effect of the model at the extreme values is improved,demonstrating the effectiveness of the Error Correction.