Abstract: In view of the unreasonable data iteration and the problem that the new effective information can not be fully utilized of the grey prediction model in the short circuit current peak prediction of power system,a grey BP neural network dynamic prediction model was proposed to adapt for power system.By introducing the dynamic data iteration model,the traditional gray model was improved with the minimum relative error as the target parameter.The short circuit fault model of power system was built by Matlab/Simulink for simulation analysis,and the current data of short circuit of power system under different initial phase angles were obtained. The improved grey model was trained by the short circuit current,the fault initial angle,the prediction result of the grey model and its relative residuals as the input of training BP neural network to obtain the final prediction model of the short circuit current peak.Verification experiments showed that the model could achieve fast and accurate prediction of short circuit current peaks,and was suitable for complex systems with few original sample points,significant nonlinear features and strong randomness.