Harnessing the power of gradient-based simulations for multi-objective optimization in particle accelerators

Particle accelerator operation requires simultaneous optimization of multiple objectives. Multi-objective optimization (MOO) is particularly challenging due to trade-offs between the objectives. Evolutionary algorithms, such as genetic algorithms (GAs), have been leveraged for many optimization prob...

Full description

Saved in:
Bibliographic Details
Main Authors: Kishansingh Rajput, Malachi Schram, Auralee Edelen, Jonathan Colen, Armen Kasparian, Ryan Roussel, Adam Carpenter, He Zhang, Jay Benesch
Format: Article
Language:English
Published: IOP Publishing 2025-01-01
Series:Machine Learning: Science and Technology
Subjects:
Online Access:https://doi.org/10.1088/2632-2153/adc221
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1849726010612776960
author Kishansingh Rajput
Malachi Schram
Auralee Edelen
Jonathan Colen
Armen Kasparian
Ryan Roussel
Adam Carpenter
He Zhang
Jay Benesch
author_facet Kishansingh Rajput
Malachi Schram
Auralee Edelen
Jonathan Colen
Armen Kasparian
Ryan Roussel
Adam Carpenter
He Zhang
Jay Benesch
author_sort Kishansingh Rajput
collection DOAJ
description Particle accelerator operation requires simultaneous optimization of multiple objectives. Multi-objective optimization (MOO) is particularly challenging due to trade-offs between the objectives. Evolutionary algorithms, such as genetic algorithms (GAs), have been leveraged for many optimization problems, however, they do not apply to complex control problems by design. This paper demonstrates the power of differentiability for solving MOO problems in particle accelerators using a deep differentiable reinforcement learning (DDRL) algorithm. We compare the DDRL algorithm with model-free reinforcement learning (MFRL), GA, and Bayesian optimization (BO) for simultaneous optimization of heat load and trip rates in the continuous electron beam accelerator facility. The underlying problem enforces strict constraints on both individual states and actions as well as cumulative (global) constraints on energy requirements of the beam. Using historical accelerator data, we develop a physics-based surrogate model which is differentiable and allows for back-propagation of gradients. The results are evaluated in the form of a Pareto-front with two objectives. We show that the DDRL outperforms MFRL, BO, and GA on high dimensional problems.
format Article
id doaj-art-e80606d10b8948ffb74c8d249de2d5ef
institution DOAJ
issn 2632-2153
language English
publishDate 2025-01-01
publisher IOP Publishing
record_format Article
series Machine Learning: Science and Technology
spelling doaj-art-e80606d10b8948ffb74c8d249de2d5ef2025-08-20T03:10:20ZengIOP PublishingMachine Learning: Science and Technology2632-21532025-01-016202501810.1088/2632-2153/adc221Harnessing the power of gradient-based simulations for multi-objective optimization in particle acceleratorsKishansingh Rajput0https://orcid.org/0000-0002-4430-9937Malachi Schram1https://orcid.org/0000-0002-3475-2871Auralee Edelen2Jonathan Colen3https://orcid.org/0000-0003-4162-0276Armen Kasparian4Ryan Roussel5Adam Carpenter6He Zhang7Jay Benesch8Thomas Jefferson National Accelerator Facility , Newport News, VA 23606, United States of America; Department of Computer Science, University of Houston , Houston, TX 77204, United States of AmericaThomas Jefferson National Accelerator Facility , Newport News, VA 23606, United States of America; Department of Computer Science, Old Dominion University , Norfolk, VA 23529, United States of AmericaSLAC National Laboratory , Menlo Park, CA 94025, United States of AmericaJoint Institute on Advanced Computing for Environmental Studies, Old Dominion University , Norfolk, VA 23539, United States of America; Hampton Roads Biomedical Research Consortium , Portsmouth, VA 23703, United States of AmericaThomas Jefferson National Accelerator Facility , Newport News, VA 23606, United States of AmericaSLAC National Laboratory , Menlo Park, CA 94025, United States of AmericaThomas Jefferson National Accelerator Facility , Newport News, VA 23606, United States of AmericaThomas Jefferson National Accelerator Facility , Newport News, VA 23606, United States of AmericaThomas Jefferson National Accelerator Facility , Newport News, VA 23606, United States of AmericaParticle accelerator operation requires simultaneous optimization of multiple objectives. Multi-objective optimization (MOO) is particularly challenging due to trade-offs between the objectives. Evolutionary algorithms, such as genetic algorithms (GAs), have been leveraged for many optimization problems, however, they do not apply to complex control problems by design. This paper demonstrates the power of differentiability for solving MOO problems in particle accelerators using a deep differentiable reinforcement learning (DDRL) algorithm. We compare the DDRL algorithm with model-free reinforcement learning (MFRL), GA, and Bayesian optimization (BO) for simultaneous optimization of heat load and trip rates in the continuous electron beam accelerator facility. The underlying problem enforces strict constraints on both individual states and actions as well as cumulative (global) constraints on energy requirements of the beam. Using historical accelerator data, we develop a physics-based surrogate model which is differentiable and allows for back-propagation of gradients. The results are evaluated in the form of a Pareto-front with two objectives. We show that the DDRL outperforms MFRL, BO, and GA on high dimensional problems.https://doi.org/10.1088/2632-2153/adc221reinforcement learningBayesian optimizationgenetic algorithmmulti objectiveMORLMOGA
spellingShingle Kishansingh Rajput
Malachi Schram
Auralee Edelen
Jonathan Colen
Armen Kasparian
Ryan Roussel
Adam Carpenter
He Zhang
Jay Benesch
Harnessing the power of gradient-based simulations for multi-objective optimization in particle accelerators
Machine Learning: Science and Technology
reinforcement learning
Bayesian optimization
genetic algorithm
multi objective
MORL
MOGA
title Harnessing the power of gradient-based simulations for multi-objective optimization in particle accelerators
title_full Harnessing the power of gradient-based simulations for multi-objective optimization in particle accelerators
title_fullStr Harnessing the power of gradient-based simulations for multi-objective optimization in particle accelerators
title_full_unstemmed Harnessing the power of gradient-based simulations for multi-objective optimization in particle accelerators
title_short Harnessing the power of gradient-based simulations for multi-objective optimization in particle accelerators
title_sort harnessing the power of gradient based simulations for multi objective optimization in particle accelerators
topic reinforcement learning
Bayesian optimization
genetic algorithm
multi objective
MORL
MOGA
url https://doi.org/10.1088/2632-2153/adc221
work_keys_str_mv AT kishansinghrajput harnessingthepowerofgradientbasedsimulationsformultiobjectiveoptimizationinparticleaccelerators
AT malachischram harnessingthepowerofgradientbasedsimulationsformultiobjectiveoptimizationinparticleaccelerators
AT auraleeedelen harnessingthepowerofgradientbasedsimulationsformultiobjectiveoptimizationinparticleaccelerators
AT jonathancolen harnessingthepowerofgradientbasedsimulationsformultiobjectiveoptimizationinparticleaccelerators
AT armenkasparian harnessingthepowerofgradientbasedsimulationsformultiobjectiveoptimizationinparticleaccelerators
AT ryanroussel harnessingthepowerofgradientbasedsimulationsformultiobjectiveoptimizationinparticleaccelerators
AT adamcarpenter harnessingthepowerofgradientbasedsimulationsformultiobjectiveoptimizationinparticleaccelerators
AT hezhang harnessingthepowerofgradientbasedsimulationsformultiobjectiveoptimizationinparticleaccelerators
AT jaybenesch harnessingthepowerofgradientbasedsimulationsformultiobjectiveoptimizationinparticleaccelerators