A high toroidal eddy current induced in a vacuum vessel during plasma-current quench, Ip quench, results in errors in determining the vertical position of the plasma-current center, ZJ, calculated from standard linear regression sensor algorithms. These deviations result in a vertical displacement event (VDE) that must be avoided because of the expected severe damage on the first wall in tokamak fusion reactors like the International Thermonuclear Experimental Reactor (ITER). On the other hand, high ZJ calculation accuracy must be maintained at steady state to obtain reasonable plasma performance. Thus, real-time sensor algorithms for the calculation of ZJ applicable to the two cases of steady state and slow Ip quench are investigated. When a statistical method is applied to the ZJ calculation, its deviation from the actual ZJ cannot be completely reduced at the same time for both cases. On the contrary, a neural network demonstrates high accuracy in the calculation of ZJ for both cases, which enables real-time feedback control of ZJ during slow Ip quench, avoids VDE, and keeps reasonable plasma performance during steady state.