FZJ, Jülich, Germany
Successful operation of a particle accelerator requires accompanying model calculations. The model helps in understanding the machine and predicts the impact of a change in the settings (e.g. current of magnetic elements). For the COoler SYnchrotron (COSY) at Research Center Jülich the accelerator simulation software MAD-X is used to model the machine. The model parameters are steadily being improved based on various manual adjustments and analytical studies, however are hardly optimized all at once. This can be improved with machine learning methods. The model is used to predict measurable quantities, like Orbit Response Matrix (ORM) or betatron tunes. Several observables for different particle energies have been measured recently and the corresponding machine settings are available. We describe the effort to improve the agreement between measured and calculated ORMs and hence improve the agreement between model and (real) machine and report on the optimization using a multivariate algorithm (e.g. genetic algorithm). This facilitates the setup of COSY and will allow to perform high precision experiments e.g. a measurement of an electric dipole moment of deuterons at COSY.
