Abstract: Fault arc, as a common safety hazard that causes low-voltage electrical accidents, has characteristics such as randomness, concealment, and danger. The existing protection methods usually take corresponding measures after a fault occurs, lacking early warning before the fault occurs.This paper designs a fault arc hidden danger early warning model based on artificial intelligence algorithm, and learns the model by extracting the eigenvalues in each sample that have strong correlation with the fault arc as the fault discrimination index, so as to realize the accurate prediction of fault arc in advance and improve the Factor of safety of low-voltage users.The experimental verification results show that the model has high accuracy and can accurately predict whether Arc fault will occur in the future cycle. The comparative analysis with DT (decision tree), GRU, RNN and other mainstream classification prediction models shows the advantages of the model。