Solar Particle Event Dose Forecasting Using Regression Techniques

Doses from solar particle events can be a serious threat to the wellbeing of crews traveling through space. Therefore predicting the time that such event will take place, forecasting the dose buildup over time, and the total dose from such event is needed to enable crews to take actions to mitigate the effects by entering a shielded area designed for their protection. Earlier work developed methods that used neural networks and Bayesian methods to forecast the total dose and dose versus time profile from an event. Subsequently, Locally Weighted Regression (LWR) and Kernel Regression (KR) techniques have been investigated to forecast the total dose. In this work, Kernel Regression methods are used to train and dose forecasting software using the dose rate and total accumulated dose. After training, the software predicts the dose buildup over time and the total dose for the test event. In the current research we have divided all of the events in our database into eight groups and use KR to train each group separately. We then test them to determine if the percentage differences between the dose forecast predictions for the test events and the actual event data, for each event in the group, are less than a 15% target value within 4 hours of the onset of the event. Results for the current dose forecasting system are presented.