Speaker
Description
In the light of recent developments to employ high-Q superconducting cavities to reduce the energy consumption of particle accelerators, the problem of minimizing cavity detuning becomes highly relevant. Meanwhile, radio frequency cavities are known for their non-stationary behavior, so finding a proper modeling approach is crucial for any model-based detuning control. In this contribution, we present a data-driven modeling framework that combines two complementary perspectives on dynamical systems: the Koopman operator approach, which captures global patterns in cavity behavior, and Kalman-inspired ideas, which enable local, adaptive adjustments in response to changing cavity conditions. Following this, our architecture 1) separates long-term structure from local variability and then 2) blends them using a tunable weighting mechanism. We demonstrate the effectiveness of this approach on both synthetic cavity data and real-world cavity measurements. The results show a potential to make the detuning control more robust to the non-stationary cavity behavior.
