Divertor power load predictions based on machine learning

Abstract

Machine learning based data-driven approaches to thermal load prediction on the divertor targets of ASDEX upgrade (AUG) are presented. After selecting time averaged data from almost six years of operation of AUG and applying basic physics-motivated cuts to the data we find that we are able to train machine learning models to predict a scalar quantifying the steady state thermal loads on the outer divertor target given scalar operational parameters. With both random forest and neural network based models we manage to achieve decent agreement between the model predictions and the observed values from experiments. Furthermore, we investigate the dependencies of the models and observe that the models manage to extract trends expected from previous physics analyses.