Article Figures & Data
Figures
- FIGURE 1
Architecture of the FFNN
- FIGURE 2
LSTM architecture (Fang, et al., 2020)
- FIGURE 3
Temporal deep learning LSTM NN architecture for training
- FIGURE 4
Temporal deep learning LSTM NN architecture for prediction
- FIGURE 5
Test RMSE with different LSTM units
- FIGURE 6
Flowchart of data processing
- FIGURE 7
Kp-indices on May 12-18 and October 9-15
- FIGURE 8
The regional coverage
- FIGURE 9
Hourly MAXAE, MAE, and RMSE for October 9 and May 12, 2021: ionosphere prediction
- FIGURE 10
October 9, 2021: prediction error plots
- FIGURE 11
May 12, 2021: prediction error plots
- FIGURE 12
October 9, 2021: comparison between the broadcasting ionosphere and CODE rapid products
- FIGURE 13
May 12, 2021: a comparison between the broadcasting ionosphere and CODE rapid products
- FIGURE 14
Three-Station (mtcu, rkd1, and ptaa) VTEC comparison of model prediction, rapid IONEX, and IGS broadcast spherical harmonics for the October 9 and May 12 datasets
- FIGURE 15
One-week prediction comparisons: with and without daily updates
- FIGURE 16
One-week comparisons (May 12–18 and October 9–15) between predictions made by Ionex, FFNN, and LSTM models
Tables
Parameters Value Activation Function Relu Max Epochs 250 Number of LSTM Layers 2 or 1 1st LSTM Layer Hidden Units 400 2nd LSTM Layer Hidden Units 200 3rd LSTM Layer Hidden Units 200 Number of Dense Layers 1 Learning Rate 0.005 Optimizer Adam Dropout 0.2 Loss function MSE Dataset Area Longitude Area Latitude Start Time End Time With Kp-index 1 −130°~ -95° 40° ~ 52.5° Jul 11, 2021 Oct 9, 2021 No 2 −130°~ -95° 40° ~ 52.5° Jul 11, 2021 Oct 9, 2021 Yes 3 −130°~ -95° 40° ~ 52.5° Feb 11, 2021 May 12, 2021 No 4 −130° ~ -95° 40° ~ 52.5° Feb 11, 2021 May 12, 2021 Yes Case Dataset Input Dimension Neural Network 1 1 2160 * 48 48–400–48 2 1 2160 * 48 48–400–200–48 3 2 2160 * 49 49–400–49 4 2 2160 * 49 49–400–200–49 5 3 2160 * 48 48–400–48 6 3 2160 * 48 48–400–200–48 7 4 2160 * 49 49–400–49 8 4 2160 * 49 49–400–200–49 9 1 2160 * 48 48–400–200–200–48 10 2 2160 * 49 49–400–200–200–49 11 3 2160 * 48 48–400–200–200–48 12 4 2160 * 49 49–400–200–200–49 Test Results Neural Network MAXAE (TECU) MAE (TECU) RMSE (TECU) 1 48–400–48 2.90 0.74 0.96 2 48–400–200–48 3.01 0.60 0.79 3 49–400–49 3.64 0.93 1.22 4 49–400–200–49 3.66 1.03 1.30 5 48–400–48 5.32 1.27 1.84 6 48–400–200–48 5.40 1.30 1.66 7 49–400–49 5.99 1.32 1.66 8 49–400–200–49 7.10 1.77 2.20 9 48–400–200–200–48 4.37 0.78 1.17 10 49–400–200–200–49 6.88 1.36 2.12 11 48–400–200–200–48 5.16 1.97 2.37 12 49–400–200–200–49 6.44 1.43 1.85 Dates RMSE (TECU) Dates RMSE (TECU) May 12 1.66 Oct 9 0.79 May 13 1.96 Oct 10 1.37 May 14 1.21 Oct 11 1.22 May 15 0.63 Oct 12 1.20 May 16 0.53 Oct 13 1.76 May 17 0.57 Oct 14 1.18 May 18 0.92 Oct 15 1.42 Period Station FFNN LSTM May 12–18 mtcu 4.17 0.75 rkd1 4.02 0.72 ptaa 3.86 0.74 October 9–15 mtcu 2.60 1.02 rkd1 2.53 1.02 ptaa 2.60 1.02 Station Rapid Ionosphere ML Predicted Ionosphere IGS RT-GIM mtcu 0.743 0.805 0.938 ptaa 0.854 0.842 0.984 rkd1 0.846 0.883 1.014 Station Rapid Ionosphere ML Predicted Ionosphere IGS RT-GIM mtcu 0.664 0.665 0.696 ptaa 0.946 0.941 0.927 rkd1 0.857 0.871 0.859
Jump to section
Related Articles
Cited By...
- No citing articles found.