|TUPLS14||Analyzing Accelerator Operation Data with Neural Networks||487|
Funding: Work is supported by DOE contract DE-AC02-76SF00515 (SLAC) and DOE contracts 2018-SLAC-100469 and 2018-SLAC-100469ASCR.
Accelerator operation history data are used to train neural networks in an attempt to understand the underly-ing causes of performance drifts. In the study, injection efficiency of SPEAR3  over two runs is modelled with a neural network (NN) to map the relationship of the injection efficiency with the injected beam trajectory and environment variables. The NN model can accurately predict the injection performance for the test data. With the model, we discovered that an environment parameter, the ground temperature, has a big impact to the injection performance. The ideal trajectory as a function of the ground temperature can be extracted from the model. The method has the potential for even larger scale application for the discovery of deep connections between machine performance and environment parameters.
|DOI •||reference for this paper ※ https://doi.org/10.18429/JACoW-NAPAC2019-TUPLS14|
|About •||paper received ※ 29 August 2019 paper accepted ※ 06 September 2019 issue date ※ 08 October 2019|
|Export •||reference for this paper using ※ BibTeX, ※ LaTeX, ※ Text/Word, ※ RIS, ※ EndNote (xml)|