 |
| Titel |
Prediction of Sym-H index by NARX neural network from IMF and solar wind data |
| VerfasserIn |
L. Cai, S.-Y. Ma, R.-S. Liu, K. Schlegel, Y.-L. Zhou, H. Luehr |
| Konferenz |
EGU General Assembly 2009
|
| Medientyp |
Artikel
|
| Sprache |
Englisch
|
| Digitales Dokument |
PDF |
| Erschienen |
In: GRA - Volume 11 (2009) |
| Datensatznummer |
250026055
|
|
|
|
|
|
| Zusammenfassung |
| Similar to Dst, the Sym-H index is also an indicator of magnetic storm intensity, but
having distinct advantage of higher time-resolution. In this study an artificial neural
network (ANN) of Nonlinear Auto Regressive with eXogenous inputs (NARX) has
been developed to predict for the first time Sym-H index from solar wind and IMF
parameters.
In total 73 great storm events during 1998 to 2006 are used, out of which 67 are selected
to train the network and the other 6 samples including 2 super-storms for test. The newly
developed NARX model shows much better capability than usual BP and Elman network
in Sym-H prediction. When using IMF Bz, By and total B with a history length
of 90 minutes along with solar wind proton density Np and velocity Vsw as the
original external inputs of the ANN to predict Sym-H index one hour later, the
cross-correlation between NARX network predicted and Kyoto observed Sym-H is 0.95 for
the 6 test storms as a whole, even as high as 0.95 and 0.98 respectively for the two
super-storms. This excellent performance of the NARX model can mainly be attributed to a
feedback from the output neuron with a suitable length of about 120 min. to the
external input. It is such a feedback that makes the ring current status properly
brought into effect in the prediction of storm-time Sym-H index by our NARX
network.
Furthermore, different parameter combinations with different history length (70 to 120
min.) for IMF and solar wind data as external inputs are examined along with different hidden
neuron number. It is found that the NARX network with 10 hidden units and with 100 min.
length of Bz, Np and Vsw as external inputs provides the best results in Sym-H prediction.
Besides, efforts have also been made to predict Sym-H longer time ahead, showing that the
NARX network can predict Sym-H index 180 min. ahead with correlation coefficient of 0.94
between predicted and observed Sym-H and RMSE less than 19 nT for the 6 test samples. |
|
|
|
|
|
|