Modeling Geomagnetospheric Disturbances with Sequential Bayesian Recurrent Neural Networks
Sequential Bayesian trained recurrent neural networks (RNNs) have not yet been considered for modeling the dynamics of magnetospheric plasma. We provide a discussion of the state-space modeling framework and an overview of sequential Bayesian estimation. Three nonlinear filters are then proposed for online RNN parameter estimation, which include the extended Kalman filter, the unscented Kalman filter, and the ensemble Kalman filter. The exogenous inputs to the RNNs consist of three parameters, \(b_z , b^2\), and \(b^2_y\) , where b, b z , and b y represent the magnitude, the southward and azimuthal components of the interplanetary magnetic field (IMF) respectively. The three models are compared to a model used in operational forecasts on a severe double storm that has so far been difficult to forecast. It is shown that some of the proposed models significantly outperform the current state of the art.
KeywordsGeomagnetic Storms Recurrent Neural Networks Filtering
Unable to display preview. Download preview PDF.
- 3.Dungey, J.W.: Interplanetary magnetic field and the auroral zones. Phys. Rev. Lett. 26, 47–48 (2000)Google Scholar
- 11.Julier, S., Uhlmann, J.: A New Extension of the Kalman Filter to Nonlinear Systems. In: Signal Processing, Sensor Fusion, and Target Recognition VI, vol. 3068, pp. 182–193 (1997)Google Scholar
- 14.Lundstedt, H., Gleisner, H., Wintoft, P.: Operational forecasts of the geomagnetic Dst index. Geophys. Res. Lett. 29, 34–1–34–4 (2002)Google Scholar
- 15.Lund Space Weather Center, http://www.lund.irf.se/rwc/dst/models/
- 16.Mirikitani, D.T., Nikolaev, N.: Dynamic Modeling with Ensemble Kalman Filter Trained Recurrent Neural Networks. In: ICMLA 2008, pp. 843–848 (2008)Google Scholar
- 17.Nikolaev, N., de Menezes, L.: Sequential Bayesian Kernel Modelling with Non-Gaussian Noise. Neural Networks 21(1), 36–47 (2008)Google Scholar