Subspace identification analysis of RFX and T2R reversed-field pinches

Input-output datasets from two magnetic confinement fusion (MCF) experiments of the reversed-field pinch (RFP) type are examined. The RFP datasets, which are samples of the distributed magnetic field dynamics, are naturally divided into many smaller batches due to the pulsed-plasma operation of the experiments. The two RFP experiments considered are (i) EXTRAP T2R (T2R) with 64 inputs and 64 outputs and (ii) RFX-mod (RFX) with 192 inputs and 192 outputs. Both T2R and RFX are magnetohydrodynamically unstable and operates under magnetic feedback with optional dither injection. Using subspace system identification techniques and randomised cross-validation (CV) methods to minimise the generalisation error, state-space orders of the empirical systems are suggested. These system orders are compared to "stabilisation diagrams" commonly used in experimental modal analysis practice. The relation of the CV system order to the decay of the singular values from the subspace method is observed. Both (i) stable vacuum diffusion and (ii) unstable plasma response datasets are analysed. Apparent simulation and prediction errors are quantified for both cases using a deviation-accounted-for index. These results are purely data-driven. A simple approach towards exploitation of the subspace techniques for finite-element model refinement and data confrontation is presented.