This article proposes a sector model early warning method based on ELM theory, with the background of big data processing cluster technology. A fault early warning model is established using naive Bayesian algorithm combined with time series similarity fault matching. The model adopts collected data such as relevant electrical quantities, switching values, event sequence information, power grid topology, and data obtained from fault recording devices during line faults. The naive Bayesian algorithm is used to mine the occurrence index of potential fault occurrence factors (i.e. fault factors), and then combined with time series similarity fault matching to perform fault warning on the line.