JACoW logo

Joint Accelerator Conferences Website

The Joint Accelerator Conferences Website (JACoW) is an international collaboration that publishes the proceedings of accelerator conferences held around the world.


BiBTeX citation export for TUCPL01: Adding Machine Learning to the Analysis and Optimization Toolsets at the Light Source BESSY II

@InProceedings{veraramirez:icalepcs2019-tucpl01,
  author       = {L. Vera Ramirez and G. Hartmann and T. Mertens and R. Müller and J. Viefhaus},
  title        = {{Adding Machine Learning to the Analysis and Optimization Toolsets at the Light Source BESSY II}},
  booktitle    = {Proc. ICALEPCS'19},
  pages        = {754--760},
  paper        = {TUCPL01},
  language     = {english},
  keywords     = {injection, booster, controls, network, operation},
  venue        = {New York, NY, USA},
  series       = {International Conference on Accelerator and Large Experimental Physics Control Systems},
  number       = {17},
  publisher    = {JACoW Publishing, Geneva, Switzerland},
  month        = {08},
  year         = {2020},
  issn         = {2226-0358},
  isbn         = {978-3-95450-209-7},
  doi          = {10.18429/JACoW-ICALEPCS2019-TUCPL01},
  url          = {https://jacow.org/icalepcs2019/papers/tucpl01.pdf},
  note         = {https://doi.org/10.18429/JACoW-ICALEPCS2019-TUCPL01},
  abstract     = {The Helmholtz Association has initiated the implementation of the Data Management and Analysis concept across its centers in Germany. At Helmholtz-Zentrum Berlin, both the beamline and the machine (accelerator) groups have started working towards setting up the infrastructure and tools to introduce modern analysis, optimization, automation and AI techniques for improving the performance of the (large scale) user facility and its experimental setups. This paper focuses on our first steps with Machine Learning techniques over the past months at BESSY II as well as organizational topics and collaborations. The presented results correspond to two complementary scenarios. The first one is based on supervised ML models trained with real accelerator data, whose target are real-time predictions for several measurements (lifetime, efficiency, beam loss, …); some of these techniques are also used for additional tasks such as outlier detection or feature importance analysis. The second scenario includes first prototypes towards self-tuning of machine parameters in different optimization cases (injection efficiency, orbit correction, …) with Deep Reinforcement Learning agents.},
}