This work presents an investigation of plasma shape reconstruction using a newly developed code based on the Cauchy-condition surface (CCS) method. The code is benchmarked against equilibrium configurations designed for the EXL-50U device and successfully reconstructs limiter, double-null, and lower single-null divertor configurations with a shape deviation of less than 1 cm. The impact of key parameters on the reconstruction results is analyzed. It has been found that the number of CCS nodes significantly influences the accuracy of the last closed flux surface (LCFS) reconstruction. As increases, the shape difference decreases and converges. Parameters defining the CCS shape and location, which must be specified prior to simulation, are found to have minimal effect on LCFS accuracy. The relationship between the noise level of the input magnetic signals and the truncation parameters in the singular value decomposition process is also analyzed, revealing their critical role in ensuring reconstruction stability and accuracy. To validate its application, an EXL-50U experimental discharge is used to track plasma shape changes, with the reconstructed LCFS showing good agreement with charge-coupled device images. A comparison with EFIT reconstructions reveals notable discrepancies, particularly during the electron cyclotron heating phase, and the possible causes of these differences are discussed.
A machine-learning-based tool for last closed-flux surface reconstruction on tokamaks