Automatic fitting of SOLPS-ITER predictions to experimental data using Bayesian optimization

Matching the predictions of SOLPS-ITER to experimental data provides a means of investigating the role various physical processes play in determining the edge and divertor plasma state, and is a key component of divertor physics studies. This is a time-expensive process, as the modeler must manually inspect the results of each SOLPS simulation and adjust the input parameters to obtain a better match. We present a method for automatic fitting of SOLPS predictions to experimental data based on Bayesian optimisation - a technique designed to minimize the number of model evaluations required to achieve a match with the data. By automating the fitting process, SOLPS can be run continuously (including multiple runs in parallel) to greatly reduce the time required to obtain a good match with experiment. This approach was validated using synthetic edge Thomson-scattering data generated from MAST-U simulations, where parameters defining the anomalous transport coefficient profiles were optimized in order to match SOLPS predictions to the synthetic Thomson-scattering data. We present results of these synthetic tests in addition to results of applying the method to actual MAST-U data. Our method is not specific to MAST-U, and can be applied to any tokamak simulated in SOLPS-ITER.