The development of a real time forecasting tool for the Dst Index: A system identification approach

A system identification approach is used to develop a real time Dst index model. System identification methodologies automatically deduce mathematical models of dynamical systems with unknown physics from measured data. For this study, we employ a Nonlinear AutoRegressive Moving Average eXogenous (NARMAX) technique. NARMAX models are built up term by term in a way that reveals the contribution of each term, which makes the models physically interpretable. This is advantageous over most other nonlinear data driven modelling methods since we do not only want an accurate forecast but also wish to understand the physics of the underlying system. The main challenge of the NARMAX approach is to determine the final model structure from an initial full model consisting of many model terms some of which are potentially redundant. Different NARMAX algorithms are trialed for deducing the Dst Index model structure, using the same inputs and training data. The performance of the models deduced by these techniques are compared and then a final model is selected and implemented in real time to provide online forecasts