American Nuclear Society
Home

Home / Publications / Journals / Nuclear Technology / Volume 168 / Number 1 / Pages 178-181

Forecasting the Dose and Dose Rate from a Solar Particle Event Using Localized Weighted Regression

T. F. Nichols, L. W. Townsend, J. W. Hines

Nuclear Technology / Volume 168 / Number 1 / October 2009 / Pages 178-181

Dose/Dose Rate / Special Issue on the 11th International Conference on Radiation Shielding and the 15th Topical Meeting of the Radiation Protection and Shielding Division (Part 1) / Radiation Protection

The dose from solar particle events (SPEs) poses a serious threat to the health of astronauts. A method for forecasting the rate and total severity of such events would give time for the astronauts to take actions to mitigate the effects from an SPE. The danger posed from an SPE depends both on the total dose received and the temporal profile of the event. The temporal profile describes how quickly the dose will arrive. Previously developed methods used neural networks to predict the total dose from an event. Later, the ability to predict the temporal profiles was added to the neural network approach. Localized weighted regression (LWR) was then used to determine if better fits with less computer load could be accomplished. Previously, LWR was shown to be able to predict the total dose from an event. LWR is the model being used to forecast the dose and the temporal profile from an SPE. LWR is a nonparametric memory-based technique; it compares a new query to stored sets of exemplar data to make its predictions. It is able to forecast early in an SPE the dose and dose rate for the event. For many events the total dose is predicted within a factor of 2 within 20 min of the beginning of the event. SPEs that are within the training parameters have temporal predictions within a few hours of the start of the event. Using an LWR model, forecasts of the dose and dose rate can be made a few hours after the start of the event. The model is able to forecast most types of events within [approximately]10% accuracy. However, there are a few events that the model fails to forecast accurately.

 
Questions or comments about the site? Contact the ANS Webmaster.
advertisement