VMD-LSTM-FEDformer降水预测融合模型研究OA

In recent years,the frequency of extreme rainfall events has increased,making it particularly urgent to improve the accuracy of precipitation prediction to support crop growth.Leveraging the advantages of Variational Mode Decomposition(VMD)in time series decomposition,and the complementary strengths of Long Short-Term Memory(LSTM)networks—which excel in handling local temporal information,and FEDformer,which is adept at processing global dependencies and possesses frequency domain characteristics,this paper proposes a precipitation prediction method based on a combined VMD-LSTM-FEDformer model.Five meteorological stations with different geographical features in Henan Province were selected for prediction analysis.The results indicated that the prediction errors of the combined model at all meteorological stations were within 5 mm,demonstrating the strong robustness of the model.In comparative experiments,the Mean Absolute Error(MAE),Root Mean Square Error(RMSE),and Coefficient of Determination(R2)of the VMD-LSTM-FEDformer combined model were 9.339 8 mm,12.703 5 mm,and 0.964 4,respectively.These metrics outperformed those of other models,proving that the proposed model possesses excellent predictive capability and practical application value.