Fast, efficient and flexible particle accelerator optimisation using densely connected and invertible neural networks

TL;DR

This paper introduces deep, invertible neural network surrogate models that significantly accelerate the multi-objective optimization of particle accelerators, enabling faster and more cost-effective operation tuning.

Contribution

The paper presents novel invertible neural network models for surrogate modeling, drastically reducing optimization time and cost for particle accelerators compared to traditional physics-based methods.

Findings

Reduced time-to-solution by up to 640 times

Decreased computational cost by up to 98%

Applicable to different types of particle accelerators

Abstract

Particle accelerators are enabling tools for scientific exploration and discovery in various disciplines. Finding optimized operation points for these complex machines is a challenging task, however, due to the large number of parameters involved and the underlying non-linear dynamics. Here, we introduce two families of data-driven surrogate models, based on deep and invertible neural networks, that can replace the expensive physics computer models. These models are employed in multi-objective optimisations to find Pareto optimal operation points for two fundamentally different types of particle accelerators. Our approach reduces the time-to-solution for a multi-objective accelerator optimisation up to a factor of 640 and the computational cost up to 98%. The framework established here should pave the way for future on-line and real-time multi-objective optimisation of particle accelerators.