This book systematically discusses the algorithms and principles for achieving stable and optimal beam (or products of the beam) parameters in particle accelerators. A four-layer beam control strategy is introduced to structure the subsystems related to beam controls, such as beam device control, beam feedback, and beam optimization. This book focuses on the global control and optimization layers. As a basis of global control, the beam feedback system regulates the beam parameters against disturbances and stabilizes them around the setpoints. The global optimization algorithms, such as the robust conjugate direction search algorithm, genetic algorithm, and particle swarm optimization algorithm, are at the top layer, determining the feedback setpoints for optimal beam qualities.
In addition, the authors also introduce the applications of machine learning for beam controls. Selected machine learning algorithms, such as supervised learning based on artificial neural networks and Gaussian processes, and reinforcement learning, are discussed. They are applied to configure feedback loops, accelerate global optimizations, and directly synthesize optimal controllers. Authors also demonstrate the effectiveness of these algorithms using either simulation or tests at the SwissFEL. With this book, the readers gain systematic knowledge of intelligent beam controls and learn the layered architecture guiding the design of practical beam control systems.
Author(s): Zheqiao Geng, Stefan Simrock
Series: Particle Acceleration and Detection
Publisher: Springer
Year: 2023
Language: English
Pages: 163
City: Cham
Preface
Contents
Abbreviations
1 Introduction
1.1 Overview of Beam Controls
1.1.1 Beam Control Tasks
1.1.2 Beam Control Methods
1.2 Beam Control System
1.2.1 Hierarchy of Beam Control System
1.2.2 Beam Device Layer
1.2.3 Instrumentation Layer
1.2.4 Global Control Layer
1.2.5 Global Optimization Layer
1.2.6 Role of Machine Learning
1.2.7 SwissFEL Two-Bunch Operation
References
2 Beam Feedback Control
2.1 Beam Feedback Control Overview
2.2 Beam Feedback Control Analysis
2.2.1 Plant Characteristics
2.2.2 Static and Dynamical Controllers
2.2.3 Local and Global Control Loops
2.3 Beam Response Matrix
2.3.1 Response Matrix Identification
2.3.2 Singular Value Decomposition
2.3.3 Response Matrix Uncertainties
2.4 Static Linear Feedback Controller Design
2.4.1 Difficulties in Response Matrix Inversion
2.4.2 Matrix Inversion with SVD
2.4.3 Matrix Inversion with Least-Square Method
2.4.4 Robust Control Design
2.4.5 SwissFEL Bunch2 Feedback Control
2.5 Further Reading and Outlook
References
3 Beam Optimization
3.1 Beam Optimization Overview
3.1.1 Optimization Problems in Beam Controls
3.1.2 Formulation of Optimization Problems
3.1.3 Noise in Online Optimization Problems
3.2 Optimization Algorithms
3.2.1 Overview of Optimization Algorithms
3.2.2 Test Function
3.2.3 Spontaneous Correlation Optimization
3.2.4 Random Walk Optimization
3.2.5 Robust Conjugate Direction Search
3.2.6 Genetic Algorithm
3.2.7 Particle Swarm Optimization
3.2.8 Comparison of Optimization Algorithms
3.3 Beam Optimization Examples and Tools
3.3.1 Practical Considerations
3.3.2 FEL Optimization with SCO
3.3.3 Operating Point Changing
3.3.4 Optimization Software Tools
3.4 Further Reading and Outlook
References
4 Machine Learning for Beam Controls
4.1 Introduction to Machine Learning
4.1.1 Machine Learning Algorithms
4.1.2 Machine Learning Models
4.1.3 Machine Learning Workflow
4.1.4 Machine Learning Processes
4.2 Accelerator Modeling with Machine Learning
4.2.1 Neural Network Regression Model
4.2.2 Gaussian Process Regression Model
4.3 Applications of Machine Learning Models in Beam Controls
4.3.1 Surrogate Models of Beam Responses
4.3.2 Response Matrix Estimation with Neural Network Surrogate Models
4.3.3 Beam Optimization with Neural Network Surrogate Models
4.3.4 Feedforward Control with Neural Network Surrogate Models
4.3.5 Beam Optimization with GP Surrogate Models
4.4 Feedback Control with Reinforcement Learning
4.4.1 Introduction to Reinforcement Learning
4.4.2 Feedback Controller Design with Natural Actor-Critic Algorithm
4.4.3 Example: RF Cavity Controller Design
4.4.4 Example: Static Feedback Controller Design
4.5 Further Reading and Outlook
References
Index
