Magnetic storm has always been one of the key issues of solar-terrestrial space physics for the past half century. The Dst index is the longitudinally averaged magnetic field depression at low latitudes. It is the primary measure of the magnitude of magnetic storms, and provides a convenient way to monitor the magnetospheric ring current. By using artificial neural network and considering the effects of the period of the geomagnetic activity, this paper brought out a method of forecasting
Dst index, an hour in advance. The inputs of the network include time, season, Dst index and the first difference and the second difference of
Dst index, the mean value of 27 days ago at t time, and the output is the observed
Dst index data at next time. The trained net then can forecast
Dst index 1 h ahead. Some examples are presented by using the
Dst index data in 1985, 1986, 1990, 1991, respectively. The results indicate that the predicted
Dst index has good agreement with observed data and the corresponding root mean errors of the model comparing with the measurement were 4.00 nT, 3.72 nT, 5.35 nT and 6.82 nT, respectively. In addition, the error analysis indicates that the predicted root-mean-square error of Dst index is smaller in low solar activities than in high solar activities.