Application of a machine learning based algorithm to online optimization of the nonlinear beam dynamics of the Argonne Advanced Photon Source

Abstract

A machine learning-based optimization algorithm, the multigeneration Gaussian process optimizer, is used to optimize the nonlinear beam dynamics of the Advanced Photon Source storage ring. The dynamic aperture (DA) and the local momentum aperture (LMA) are first optimized separately with sextupole knobs. Solutions found with these optimizations are used to seed the initial population of seeds in two-objective optimizations that simultaneously optimize the DA and LMA, which lead to a distribution of solutions with different DA and LMA performances, from which a setting suitable for operation can be selected.