The OPAL Discussion Forum

["Adelmann Andreas (PSI)" <andreas.adelmann AT psi.ch>]

Please join us for the MSc presentation of Zacharias Mélissa on: Performance of different machine learning techniques for forecasting of particle accelerator interlocks

Abstract:  Two machine learning algorithms were applied to decrease beam time loss in the High Intensity Proton Accelerator complex (HIPA) by forecasting interlock events. The random forest and a convolutional neural network (CNN) trained on recurrence plots constructed from multivariate time series were trained on HIPA data from September to December 2019. The Random Forest model reached a beam time loss of 19.4 ± 0.5 seconds per interlock, compared to the baseline loss for no intervention of 25 seconds per interlock. The CNN reached a beam time loss of 20.5 ± 0.7 seconds per interlock. Preliminary testing on live predictions using the random forest model didn’t result in timely interlock predictions. Simulated live testing on the CNN indicated possible data leakage from the interlock time stamps labeling. Measures have been taken to eliminate this leakage and subsequent tests of the CNN model are promising but in need of further tuning. The number of relevant features used in both models could be reduced from the original 311 features to about 50 for the Random Forest model and 11 for the CNN model using feature analysis. Correlation analysis of the selected features in the different sets suggests relations that may warrant further investigation.

Data: 18. May

Time: 16.00

Zoom:  ID: 470-582-4086 Password: AdA

Cheers A

------
Dr. sc. math. Andreas (Andy) Adelmann
Head a.i. Labor for Scientific Computing and Modelling 
Paul Scherrer Institut OHSA/ CH-5232 Villigen PSI
Phone Office: xx41 56 310 42 33 Fax: xx41 56 310 31 91
Zoom ID: 470-582-4086 Password: AdA
-------------------------------------------------------
Friday: ETH HPK G 28   +41 44 633 3076
============================================
The more exotic, the more abstract the knowledge, 
the more profound will be its consequences.
Leon Lederman 
============================================