Ionospheric scintillation can lead to increased error rates in communication systems and decreased accuracy in GNSS positioning.Due to the sporadic nature of ionospheric scintillation,predicting scintillation events is challenging.To improve the prediction accuracy of ionospheric scintillation,this study proposes a hybrid prediction model that combines multiple methods and utilizes the scintillation label value(S4label)as an auxiliary input.The model integrates a deep learning framework based on the decompose-ensemble approach to perform the prediction.First,the original data is decomposed into multiple sub-sig-nals using the CEEMDAN algorithm.These sub-signals are then reconstructed into three components using the K-Means algorithm and the sample entropy criterion:high-frequency,low-frequency,and trend signals.The high-frequency signal is further decomposed using the VMD method,and a self-attention LSTM model is employed to progressively predict the high-frequency and low-frequency signals.Experimental results demonstrate that the proposed hybrid model significantly improves the prediction accuracy compared to the traditional LSTM model.During geomagnetically quiet periods,the model achieves remarkable improve-ments in prediction performance,with R2,RMSE,MAE and MAPE metrics showing accuracy enhance-ments of 32.2％,58.7％,51.2％and 44.7％,respectively.Therefore,the proposed model provides more ac-curate predictions of ionospheric scintillation events and holds significant value for forecasting research in this field.
维普
万方数据