Speaker
Description
In accelerator controls, there are several situations in which the optimization space is not just a scalar or vector, but a function of such over time that needs to be optimized as a whole. Traditional optimization methods rely on reformulating the problem to an optimization over a scalar representing an offset or a multiplicative factor over a reference function. By translating the function over an appropriate representation, it is possible to optimize over the coefficients space. Optimizing directly in the function space, rather than through a finite-dimensional parametric representation, allows us to more closely approximate the natural optimal control function, without the constraints imposed by a predefined parameter space. This promises better performance and greater flexibility in capturing the true optimal behavior. In this contribution we apply Bayesian functional optimization with multi-output Gaussian process modeling to constrain the resulting function. This optimization is applied to the Low Energy Ion Ring (LEIR) at CERN, demonstrating how to improve the tune function and the electron gun voltage profile.
| In which format do you inted to submit your paper? | LaTeX |
|---|