Performance drift over long periods of operation due to changes in machines settings or the environment has been a longstanding problem for particle accelerators. Algorithms which are capable of tuning machine settings while keeping the performance within a desired threshold can be used to compensate for such drifts. We have developed a modified version of the Multi-Generation Gaussian Process Optimizer (MG-GPO) which is capable of tuning accelerator settings during user operation. The modified algorithm uses Gaussian Process regression to predict the performance of potential trial settings and removes ones with a high probability of giving too poor of a performance before selection for evaluation on the machine. The modified MG-GPO has been tested on analytic functions and applied to the SPEAR3 kicker-bump matching problem as a proof of concept. It is expected that the modified MG-GPO will be applied to maintain optimal trajectory of the beam injected into the SPEAR3 storage ring.
Yeung et al. (Thu,) studied this question.