Fast, Efficient and Flexible Particle Accelerator Optimisation Using Densely Connected and Invertible Neural Networks

Abstract

Particle accelerators are enabling tools for scientific exploration and discovery in various disciplines. However, finding optimised operation points for these complex machines is a challenging task 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 online and real-time multi-objective optimisation of particle accelerators.

Combining Bayesian Optimization and Neural Network to Optimize the Plasma Wakefield Acceleration

“…Strategy 2 is the approach discussed above, which uses framework (BO combined with NN) for optimization. The advantage of using BO combined with NN compared to the approach of BO and simulation software LCODE in terms of time is evaluated from Bellotti's method 23 . The time required for different processes such as using LCODE to calculate once, as well as the number of computer cores used by different processes and their corresponding values are shown in Table 4.…”

“…Variables in the optimization problem and the corresponding range of variation use of NN predictions can greatly reduce the computing resources23 . In this paper, we introduce this novel framework that combines BO and NN, by taking AWAKE Run 2 experiment at CERN as an example, to demonstrate the effectiveness of this method.…”

Computational models for high-power cyclotrons and FFAs

“…Cheap to evaluate surrogate models have gained a lot of interest lately. Statistical [6] or machine learning techniques are used [7]. These models can for example replace a computationally heavy model in a multi-objective optimization [8] or in the future be part of an on-line model.…”

High-power Fixed-Field Accelerators

“…Cheap-to-evaluate surrogate models have gained a lot of interest lately. Statistical [218] or machine learning techniques are used [219]. These models can for example replace a computationally expensive simulation in a multi-objective optimization [220][221][222] or become an online tuning tool.…”