Prediction of Partial Ring Current Index Using LSTM Neural Network
doi: 10.11728/cjss2022.05.210513061
-
Abstract: The local time dependence of the geomagnetic disturbances during magnetic storms indicates the necessity of forecasting the localized magnetic storm indices. For the first time, we construct prediction models for the SuperMAG partial ring current indices (SMR-LT), with the advance time increasing from 1 h to 12 h by Long Short-Term Memory (LSTM) neural network. Generally, the prediction performance decreases with the advance time and is better for the SMR-06 index than for the SMR-00, SMR-12, and SMR-18 index. For the predictions with 12 h ahead, the correlation coefficient is 0.738, 0.608, 0.665, and 0.613, respectively. To avoid the over-represented effect of massive data during geomagnetic quiet periods, only the data during magnetic storms are used to train and test our models, and the improvement in prediction metrics increases with the advance time. For example, for predicting the storm-time SMR-06 index with 12 h ahead, the correlation coefficient and the prediction efficiency increases from 0.674 to 0.691, and from 0.349 to 0.455, respectively. The evaluation of the model performance for forecasting the storm intensity shows that the relative error for intense storms is usually less than the relative error for moderate storms.
-
Table 2. Performance of SMR-LT indices forecast
p=1 h p=3 h p=6 h p=12 h Erms C Pe Erms C Pe Erms C Pe Erms C Pe SMR-00 Persistence 8.942 0.838 0.675 11.384 0.737 0.474 13.438 0.633 0.267 16.049 0.477 –0.046 Model 7.468 0.883 0.773 8.616 0.836 0.699 10.793 0.737 0.527 13.350 0.608 0.276 SMR-06 Persistence 5.957 0.902 0.804 8.012 0.822 0.645 9.804 0.734 0.468 12.172 0.590 0.180 Model 5.752 0.925 0.817 6.592 0.873 0.759 7.639 0.825 0.677 9.976 0.738 0.449 SMR-12 Persistence 8.054 0.893 0.786 11.886 0.767 0.534 13.898 0.682 0.363 16.899 0.529 0.059 Model 6.473 0.928 0.862 9.884 0.826 0.678 11.485 0.769 0.565 13.811 0.665 0.371 SMR-18 Persistence 9.065 0.895 0.790 13.517 0.767 0.534 16.311 0.661 0.322 19.851 0.498 –0.005 Model 7.316 0.934 0.864 10.692 0.844 0.708 14.022 0.743 0.499 16.836 0.613 0.277 Table 3. Performance of storm-time SMR-LT indices prediction
p=1 h p=3 h p=6 h p=12 h C Pe C Pe C Pe C Pe SMR-00 Persistence 0.838 0.675 0.726 0.452 0.597 0.194 0.352 –0.296 All-time 0.885 0.783 0.838 0.700 0.724 0.485 0.526 0.134 Storm 0.887 0.759 0.840 0.704 0.734 0.499 0.552 0.257 SMR-06 Persistence 0.896 0.793 0.804 0.608 0.686 0.372 0.457 –0.087 All-time 0.922 0.823 0.868 0.748 0.807 0.642 0.674 0.349 Storm 0.924 0.853 0.871 0.759 0.818 0.668 0.691 0.455 SMR-12 Persistence 0.895 0.789 0.763 0.525 0.647 0.293 0.406 –0.189 All-time 0.934 0.872 0.829 0.674 0.758 0.532 0.595 0.250 Storm 0.935 0.875 0.840 0.705 0.770 0.552 0.618 0.316 SMR-18 Persistence 0.894 0.788 0.756 0.513 0.613 0.226 0.361 –0.278 All-time 0.935 0.872 0.842 0.698 0.718 0.435 0.516 0.106 Storm 0.936 0.871 0.851 0.694 0.736 0.488 0.559 0.214 Table 4. Magnetic storm intensity distribution
Moderate storms Intense storms Total Training set 119 72 191 Validation set 53 13 66 Test set 52 9 61 Total 224 94 318 Table 5. Relative errors (mean±standard deviation) of storm intensity predictions
p / h Relative error / (%) SMR-00 SMR-06 SMR-12 SMR-18 52 moderate storms 1 –24.98±7.18 –10.99±7.73 –9.81±13.89 –12.49±12.10 3 –23.86±11.93 –13.73±13.84 –19.79±16.57 –24.42±14.52 6 –37.34±13.34 –16.88±14.66 –28.40±15.93 –32.85±13.80 9 intense storms 1 –11.80±9.16 –9.93±3.28 –8.76±9.73 –10.56±10.92 3 –13.04±13.28 –13.67±6.16 –23.19±14.57 –23.19±13.19 6 –31.10±13.58 –19.70±10.10 –34.66±12.02 –33.56±12.53 -
[1] DUNGEY J W. Interplanetary magnetic field and the Auroral zones[J]. Physical Review Letters, 1961, 6(2): 47-48 doi: 10.1103/PhysRevLett.6.47 [2] GONZALEZ W D, JOSELYN J A, KAMIDE Y, et al. What is a geomagnetic storm[J]. Journal of Geophysical Research: Space Physics, 1994, 99(A4): 5771-5792 doi: 10.1029/93JA02867 [3] BURTON R K, MCPHERRON R L, RUSSELL C T. An empirical relationship between interplanetary conditions and Dst[J]. Journal of Geophysical Research, 1975, 80(31): 4204-4214 doi: 10.1029/JA080i031p04204 [4] DAGLIS I A, THORNE R M, BAUMJOHANN W, et al. The terrestrial ring current: origin, formation, and decay[J]. Reviews of Geophysics, 1999, 37(4): 407-438 doi: 10.1029/1999RG900009 [5] KAMIDE Y, BAUMJOHANN W, DAGLIS I A, et al. Current understanding of magnetic storms: storm-substorm relationships[J]. Journal of Geophysical Research: Space Physics, 1998, 103(A8): 17705-17728 doi: 10.1029/98JA01426 [6] WANG C B, CHAO J K, LIN C H. Influence of the solar wind dynamic pressure on the decay and injection of the ring current[J]. Journal of Geophysical Research: Space Physics, 2003, 108(A9): 1341 doi: 10.1029/2003JA009851 [7] FENRICH F R, LUHMANN J G. Geomagnetic response to magnetic clouds of different polarity[J]. Geophysical Research Letters, 1998, 25(15): 2999-3002 doi: 10.1029/98GL51180 [8] O’BRIEN T P, MCPHERRON R L. Forecasting the ring current index Dst in real time[J]. Journal of Atmospheric and Solar-Terrestrial Physics, 2000, 62(14): 1295-1299 doi: 10.1016/S1364-6826(00)00072-9 [9] LUNDSTEDT H, WINTOFT P. Prediction of geomagnetic storms from solar wind data with the use of a neural network[J]. Annales Geophysicae, 1994, 12(1): 19-24 doi: 10.1007/s00585-994-0019-2 [10] LUNDSTEDT H, GLEISNER H, WINTOFT P. Operational forecasts of the geomagnetic Dst index[J]. Geophysical Research Letters, 2002, 29(24): 2181 doi: 10.1029/2002GL016151 [11] WU J G, LUNDSTEDT H. Prediction of geomagnetic storms from solar wind data using Elman Recurrent Neural Networks[J]. Geophysical Research Letters, 1996, 23(4): 319-322 doi: 10.1029/96GL00259 [12] WEI H L, ZHU D Q, BILLINGS S A, et al. Forecasting the geomagnetic activity of the Dst index using multiscale radial basis function networks[J]. Advances in Space Research, 2007, 40(12): 1863-1870 doi: 10.1016/j.asr.2007.02.080 [13] LAZZÚS J A, LÓPEZ-CARABALLO C H, ROJAS P, et al. Forecasting of DST index from auroral electrojet indices using time-delay neural network + particle swarm optimization[J]. Journal of Physics:Conference Series, 2016, 720: 012001 doi: 10.1088/1742-6596/720/1/012001 [14] GRUET M A, CHANDORKAR M, SICARD A, et al. Multiple-hour-ahead forecast of the Dst index using a combination of long short-term memory neural network and Gaussian process[J]. Space Weather, 2018, 16(11): 1882-1896 doi: 10.1029/2018SW001898 [15] BOYNTON R J, BALIKHIN M A, BILLINGS S A, et al. Data derived NARMAX Dst model[J]. Annales Geophysicae, 2011, 29(6): 965-971 doi: 10.5194/angeo-29-965-2011 [16] LI H, WANG C, TU C, et al. Machine learning approach for solar wind categorization[J]. Earth and Space Science, 2020, 7(5): e2019EA000997 doi: 10.1029/2019EA000997 [17] LU J Y, PENG Y X, WANG M, et al. Support Vector Machine combined with distance correlation learning for Dst forecasting during intense geomagnetic storms[J]. Planetary and Space Science, 2016, 120: 48-55 doi: 10.1016/j.pss.2015.11.004 [18] LI H, WANG C, KAN J R. Contribution of the partial ring current to the SYMH index during magnetic storms[J]. Journal of Geophysical Research: Space Physics, 2011, 116(A11): A11222 doi: 10.1029/2011JA016886 [19] NEWELL P T, GJERLOEV J W. SuperMAG-based partial ring current indices[J]. Journal of Geophysical Research: Space Physics, 2012, 117(A5): A05215 doi: 10.1029/2012JA017586 [20] BENGIO Y, SIMARD P, FRASCONI P. Learning long-term dependencies with gradient descent is difficult[J]. IEEE Transactions on Neural Networks, 1994, 5(2): 157-166 doi: 10.1109/72.279181 [21] HOCHREITER S, SCHMIDHUBER J. Long short-term memory[J]. Neural Computation, 1997, 9(8): 1735-1780 doi: 10.1162/neco.1997.9.8.1735 [22] LETHY A, EL-ERAKI M A, SAMY A, et al. Prediction of the Dst index and analysis of its dependence on solar wind parameters using neural network[J]. Space Weather, 2018, 16(9): 1277-1290 doi: 10.1029/2018SW001863 [23] LAZZÚS J A, VEGA P, ROJAS P, et al. Forecasting the Dst index using a swarm-optimized neural network[J]. Space Weather, 2017, 15(8): 1068-1089 doi: 10.1002/2017SW001608