Multiobjective Bayesian optimization for online accelerator tuning

Particle accelerators require constant tuning during operation to meet beam quality, total charge and particle energy requirements for use in a wide variety of physics, chemistry biology experiments. Maximizing the performance an accelerator facility often necessitates multiobjective optimization, where operators must balance trade-offs between multiple competing objectives simultaneously, using limited, temporally expensive observations. Usually, optimization problems are solved off-line, prior actual operation, with advanced line simulations parallelized methods (NSGA-II, swarm optimization). Unfortunately, it is not feasible these online since measurements can only be done serial fashion, large number converge useful solution. Here, we introduce Bayesian scheme, which finds full Pareto front problem efficiently serialized manner thus critical step towards practical accelerators. This method uses set Gaussian process surrogate models, along acquisition function, reduce observations needed by at least order magnitude over current methods. We demonstrate how this modified specifically solve challenges posed includes addition constraints, objective preferences costs related changing parameters.