摘要：The Space Weather Forecast Testbed (SWFT) is developed by a team of space weather scientists and mathematicians at the University of Southern California (USC) and Jet Propulsion Laboratory (JPL) to foster the creation of models for space weather forecast by exploration of existing historic data using techniques of machine learning. As an example to demonstrate the potential power of SWFT, we present in this paper a multilinear regression-based forecast model for solar wind. Solar wind is one of the key drivers for numerous physics-based models for space weather including thermosphere and ionosphere models. Many attempts have been made to produce forecasts for the solar wind. SWFT provides a unified framework for forecast model formulation, training, and performance assessment. In particular, the preparation of training and validation data by SWFT takes into account the realistic constraints on data latency and forecast lead time. In developing a solar wind forecast model, SWFT allows fast exploration of many metaparameters such as the list of predictive variables and their time history used in constructing a model. We present the impact of metaparameter selection, as well as performance relative to existing solar wind forecast models.
