The peak of short-circuit current is of great importance for selective protection of low-voltage distribution system and reliable breaking of circuit breaks. However, it is lack of intensive study now. Based on the early fault detection technology and the parameters analyzed by simulation, this paper concludes the main factors which influence the peak value by using grey correlation degree. Furthermore, the extreme learning machine (ELM) is used to forecast the peak value of fault current. Simulation results show that the grey correlation degree can identify the main factors of the short-circuit current effectively. And it can also reduce the dimensions of characteristic variable of short-circuit current. Finally, the short-circuit current prediction method based on the early fault detection and extreme learning machine shows strong robustness and high precision, which can lay the foundation for the realization of low-voltage selective protection technology.