Abstract: The key and difficult point of fast phase controlled breaking of short circuit current is to solve the inherent contradiction between the rapidity and accuracy of fault identification and zero prediction. Therefore, we investigated and proposed a fast prediction method of short-circuit current zero crossing based on the long short term memory (LSTM) algorithm. A phase control device test platform was built, and the current prediction ability of the LSTM algorithm was verified by the RTDS test and artificial short-circuit fault recording test. Moreover, we studied and discussed the influences of the number of nodes in the hidden layer of LSTM network, the length of the sampling window, the initial phase angle of the fault, the amplitude of the power frequency component, the DC attenuation time constant and the signal to noise ratio on the zero prediction error. The simulation and test results show that the fault identification time is 0.3 ms, the zero-prediction sampling time is 3 ms, and the zero-prediction error is ±0.5 ms. Under the condition that the prediction accuracy is equivalent to that of traditional methods, LSTM method can be adopted to significantly shorten the prediction time and to improve the prediction rapidity. The results can provide a theoretical basis and technical support for the fast breaking of system faults.