Machine learning for orders of magnitude speedup in multiobjective optimization of particle accelerator systems

Abstract

High-fidelity physics simulations are powerful tools in the design and optimization of charged particle accelerators. However, the computational burden of these simulations often limits their use in practice for design optimization and experiment planning. It also precludes their use as on-line models tied directly to accelerator operation. We introduce an approach based on machine learning to create nonlinear, fastexecuting surrogate models that are informed by a sparse sampling of the physics simulation. The models are Oð10 6 Þ-Oð10 7 Þ times more computationally efficient to execute. We also demonstrate that these models can be reliably used with multiobjective optimization to obtain orders-of-magnitude speedup in initial design studies and experiment planning. For example, we required 132 times fewer simulation evaluations to obtain an equivalent solution for our main test case, and initial studies suggest that between 330-550 times fewer simulation evaluations are needed when using an iterative retraining process. Our approach enables new ways for high-fidelity particle accelerator simulations to be used, at comparatively little computational cost.

Advanced Control Methods for Particle Accelerators (ACM4PA) 2019

“…ML-based tools, such as neural networks (NN), can be trained to automatically tune and control large complex systems such as particle accelerators [37][38][39][40]. ML tools are being developed to provide surrogate models to create diagnostics [41]. A NN model has been designed to predict the resonant frequency of the radio frequency quadrupole (RFQ) in the PIP-II Injector Experiment (PXIE), to be used in a model predictive control scheme [42].…”

Model-independent tuning for maximizing free electron laser pulse energy

“…ML-based tools, such as neural networks (NN), can be trained to automatically tune and control large complex systems such as particle accelerators [45][46][47][48]. ML tools are being developed to provide fast and accurate surrogate models to create diagnostics that enable feedback control and tuning of accelerators [49]. A NN model has been designed to predict the resonant frequency of the radio frequency quadrupole (RFQ) in the PIP-II Injector Experiment (PXIE), to be used in a model predictive control scheme [50].…”

Neural network-based multiobjective optimization algorithm for nonlinear beam dynamics

“…And, a combination of MOGA and MOPSO has more potential of avoiding local optima than using the MOGA or MOPSO alone [12]. Moreover, accelerator scientists are now exploring different machine learning enhanced MOGAs [14][15][16][17][18]. The basic consideration is that the dataset fX n ; Y n g continuously produced and accumulated by the MOGA can be used as the training data of machine learning, so as to reveal some hidden properties of the data which, in turn, helps to speed up the convergence and/or increase the diversity among solutions.…”

“…The new competitive offspring individuals are then generated within the elite variable range and used to replace the original data in the population, which can result in faster convergence. Another approach is to use the data from MOGA to train a machine learning surrogate model (e.g., [16,17]), which can predict the objective values in fractions of a second, much faster than the actual evaluator. To take advantage of such a surrogate model, one can use it to replace the actual MOGA evaluator as in Ref.…”

“…To take advantage of such a surrogate model, one can use it to replace the actual MOGA evaluator as in Ref. [16]. Such an approach can greatly reduce the optimization time.…”