Realization and comparison of several regression algorithms for electron energy spectrum reconstruction

摘要

To determine the electron energy spectra for medical accelerator effectively, we investigate a nonlinear programming model with several nonlinear regression algorithms, including Levenberg-Marquardt, Quasi Newton, Gradient, Conjugate Gradient, Newton, Principal-Axis and NMinimize algorithms. The local relaxation-bound method is also developed to increase the calculation accuracy. The testing results demonstrate that the above methods could reconstruct the electron energy spectra effectively. Especially, further with the local relaxation-bound method the Levenberg-Marquardt, Newton and NMinimize algorithms could precisely obtain both the electron energy spectra and the photon contamination. Further study shows that ignoring about 4% photon contamination would increase error greatly, and it also inaccurately makes the electron energy spectra 'drift' to the low energy.