In order to effectively forecast large area power grid line trip blackout caused by mountain fire, spatial-temporal characteristics of fire hazards in Guangdong power grid mountains are studied by using the data of fire spots number, the area of overfire and the tripping frequency of main network from 2008 to 2017. A model of optimized risk early warning system based on partial least square(PLS) is proposed. It is found that the east and north of Guangdong are sensitive to mountain fires, and the occurrence of mountain fires in the overhead line coverage area has an obvious increasing trend. In addition, based on historical meteorological data, vegetation data and topographic data, the PLS method is used to select the main influencing factors of mountain fire prediction, and an optimized risk early warning model of mountain fire in power grid is proposed based on the principle of applied statistics. The validity and practicability of proposed method are verified by the case test of mountain fire trip in Guangdong power grid.