Machine Learning Calculation of Feature Importance between Coupled Parameters for Fusion Experiment Design

A sheared flow stabilized Z pinch has been shown to successfully mitigate MHD instability modes on the ZaP-FuZE experiment, paving the way for scientific breakeven on the next generation machine Fuze-Q. Machine learning algorithms that can identify the relationship between the input parameters and desirable outcomes such as neutron yield and pinch current will fill a critical role in achieving this milestone. One impediment in this analysis is the high dimensionality and coupled relationship between input parameters. We use Sparse Principal Component Analysis to reduce the dimensionality of our input parameters as in ( Godino et. al., 2021 ) to reduce our input dimensionality and enable meaningful analysis via a random forest. The prediction results are intended to advise future design decisions and act as an input to a feedback control system. This work lays the foundation for continued study of a data-oriented approach to optimization and control problems arising from the complexity of the parameter space and coupling of underlying physical processes.