Beam Size Prediction and Control using Neural Network*

Experimental results from beam lines are sensitive to the size of the beam itself. However, predicting and controlling the beam size still prove to be a challenge. Even with the feed-forward strategy using recorded control data (aka look-up tables), the beam size variance on vertical direction is still ~2 μm. Herein, we provide a machine learning based approach to predict the beam size using neural network. We perform a systematic study to optimize the prediction result using different neural network architectures and regularizers. Based on the model, we propose a neural network based beam size stablization strategy by tuning a certain experimental parameter (dispersion wave parameter). The variance of beam size on vertical direction is reduced to ~0.3 μm in the online test.