基于VMD-SSA-LSTM的海面温度预报OA北大核心

Sea surface temperature(SST)is a key parameter in the ocean-atmosphere interaction,which is crucial for tropical cyclone track and intensity prediction and global climate change research.However,the traditional long short-term memory(LSTM)neural networks provide SST forecasts with low accuracy under extreme low and high tem-peratures.In order to improve the performance of SST forecasting,this paper proposes a combined variational modal decomposition(VMD)-sparrow search algorithm(SSA)-LSTM model.This model first uses VMD to decompose the nonsmooth and nonlinear SST time series into multiple eigenmodal functions with central frequencies.Subsequently,it adopts SSA to optimize the structure and parameters of LSTM to improve the model's learning ability and generalization ability.The model was experimentally validated using SST data of the South China Sea for 82 years taken from ERA5 reanalysis.The results showed that compared with the traditional LSTM model,the VMD-SSA-LSTM model considera-bly improves the forecast accuracy under extreme temperature conditions:the average root mean squared error of the proposed model is lower by 67%,the mean absolute percentage error is lower by 65.8%,and the mean absolute error is lower by 65.6%.This result demonstrated the great advantage of the combined model in dealing with complex nonlinear climate variables.This study provides a new approach for building high-precision intelligent forecasting models for SST and provides theoretical support for monitoring and early warning of extreme climate events.