Forecasting Dst Index With Artificial Neural Network

In this work, fully connected neural network, a new kind of artificial network, has been introduced to construct the model for Dst index forecasting, Through studying the mechanism that the geomagnetosphere was affected by the condition of interplanetary media, the geomagnetic disturbance index Dst was found to have close, and complex relationship with both the solar wind parameters and IMF features. By employing the measured parameters from ACE spacecraft, these parameters were the solar wind velocity, the density of solar wind plasma and the southward component of IMF. The most recent measured Dst was also figured to correlate to the Dst several hours ahead. To construct the relationship between interplanetary measured parameters and Dst index, fully connected neural network was introduced. This neural network could demonstrate the complex relationship through building up the internal connection between separate neurons in hidden layer. After a training process with historical data, the forecast model was built during which the neural network will adjust the internal connect weight between units automatically according to the input parameters. The storm time data of 1998 and 1999 was selected in the training process of model construction. The data set during the geomagnetic storm in July 24-29 was used to test the model and the error of the test data was 14.3%.