Clustering Analyses of Tokamak Discharge Experiments From a Macroscopic Perspective

The vast amounts of diagnostic data generated by tokamaks are fundamental resources for fusion researchers when conducting experimental analyses. Current research in data analysis has been focused on specific research domains or measurement results of particular subsystems. A macrolevel analysis of an experimental operational status from the perspective of an experimental operation manager has yet to be undertaken. This article presents a discharge data analysis framework based on clustering techniques. Building on this framework, an improved means for measuring the effectiveness of density clustering is proposed for datasets without clear cluster boundaries. Meanwhile, the clustering results are further optimized through an algorithm for fine-tuning the clusters embedded with each other in shape. The research group deployed this analytical framework in an experimental network to conduct an analysis of various plasma parameters, including plasma current and plasma stored energy, normalized beta and plasma internal inductance, as well as plasma electron density and poloidal beta before and after the major upgrade of the Experimental Advanced Superconducting Tokamak (EAST). The experimental results show that the overall plasma performance of the EAST has been improved. However, plasma disruptions become more likely to occur at lower plasma current and lower plasma stored energy, and the probability of disruption significantly increases under high-current discharge conditions.