Speaker
Description
The widely adopted IPB98(y,2) confinement scaling law exhibits an approximately quadratic dependence of the energy confinement time in ELMy H‑mode tokamak plasmas on the device major radius. In contrast, the more recent ITPA20 confinement scaling indicates a substantially weaker size dependence. In this study, optimisation and classification techniques are applied to the ITPA global H‑mode confinement database to investigate the origin of this discrepancy. We demonstrate that the reduced major‑radius exponent predominantly arises from a distinct subset of discharges that is well localised in dimensionless parameter space. When this subset is excluded, the remaining dataset yields a confinement scaling that is broadly consistent with IPB98(y,2), and, crucially, recovers a significantly stronger major‑radius dependence. The analysis indicates that the anomalously weak size scaling cannot be attributed solely to any
single device, to the presence of metallic plasma‑facing components, or to predictor multicollinearity. Instead, the subsets associated with weak and strong size scaling are most clearly differentiated by key dimensionless parameters, including the normalised plasma pressure, collisionality, normalised gyroradius, and safety factor. Based on the resulting scalings, confinement time predictions are provided for expected operating points of ITER and SPARC.