Predict the Amplitude of Recurrent Geomagnetic Disturbance Using NARX Neural Network

A nonlinear autoregressive network with exogenous inputs (NARX) model is developed to predict the amplitude of recurrent geomagnetic disturbance one day ahead. The Ap index are used to measure the geomagnetic disturbance amplitudes. The external inputs are source surface features abtained from solar observations. The prediction results of Ap index from CR2181 to CR2190 show that the prediction accuracy of the NARX model for Ap index is higher than that of the 27-day persistent method. In addition, for the geomagnetic disturbed days when the Ap index are greater than 10 and 15, the NARX model also have better probability of detection and false alarm rate than the 27-day persistent method.