The rapid and accurate prediction of the zero point of short-circuit current holds importance for the restriction and elimination of generator faults. In this paper, a rapid zero-prediction algorithm for short-circuit current based on the Extreme Learning Machine (ELM) is proposed. Firstly, the principle of the ELM network along with its training and zero-prediction procedures are introduced. Subsequently, a three-phase short-circuit simulation model of the generator is established to validate the zero-prediction capability of the ELM network. The results indicate that when the sampling window length for zero-crossing point prediction is 2 ms, the prediction error of the first zero-crossing point under different initial fault phases is no more than 0.3 ms, and the prediction error of the second zero-crossing point is no more than 0.5 ms. In comparison with traditional algorithms, the ELM network significantly reduces the required prediction time and enables a faster zero-prediction. When contrasted with other artificial intelligence algorithms, the ELM network demonstrates advantages in training speed and the complexity of the network structure.