ORBIT CORRECTION STUDIES USING NEURAL NETWORKS

This paper reports the use of neural networks for orbit correction at the Australian Synchrotron Storage Ring. The proposed system uses two neural networks in an actor-critic scheme to model a long term cost function and compute appropriate corrections. The system is entirely based on the history of the beam position and the actuators, i.e. the corrector magnets, in the storage ring. This makes the system auto-tuneable, which has the advantage of avoiding the measure of a response matrix. The controller will automatically maintain an updated BPM corrector response matrix. In future if coupled with some form of orbit response analysis, the system will have the potential to track drifts or changes to the lattice functions in ”real time”. As a generic and robust orbit correction program it can be used during commissioning and in slow orbit feedback. In this study, we present positive initial results of the simulations of the storage ring in Matlab.