Frontiers in data analysis methods: From causality detection to data driven experimental design

Abstract

On the route to the commercial reactor, the experiments in magnetical confinement nuclear fusion have become increasingly complex and they tend to produce huge amounts of data. New analysis tools have therefore become indispensable, to fully exploit the information generated by the most relevant devices, which are nowadays very expensive to both build and operate. The paper presents a series of innovative tools to cover the main aspects of any scientific investigation. Causality detection techniques can help identify the right causes of phenomena and can become very useful in the optimisation of synchronisation experiments, such as the pacing of sawteeth instabilities with ion cyclotron radiofrequency heating modulation. Data driven theory is meant to go beyond traditional machine learning tools, to provide interpretable and physically meaningful models. The application to very severe problems for the tokamak configuration, such as disruptions, could help not only in understanding the physics but also in extrapolating the solutions to the next generation of devices. A specific methodology has also been developed to support the design of new experiments, proving that the same progress in the derivation of empirical models could be achieved with a significantly reduced number of discharges.