Abstract: In recent years,the occurrence of large-scale blackout has been increasing with the increased interconnection upgrading of the regional power grids.It has great realistic significance for prediction of blackout risk.In this paper,a prediction model based on generalized extreme value（GEV）is applied to predicting power loss of large-scale blackouts in power grids.Firstly,according to an analysis of Chinese power grid blackout data from 1981 to 2011by the extreme value distribution suitability text,it is found that the power loss data is not suitable for extreme value analysis.Secondly,aprediction model based on generalized extreme value distribution is put forward after the relative value method for data processing to eliminate the influence of power grid scale variation.Finally,the blackout probability and power loss of Chinese power grids in the future are predicted by using the relative power loss data from 1997 to 2012,and then the predicted value of the corresponding blackout’s power loss in Chinese regional power grids is calculated.The feasibility of the proposed method is proven by actual blackout data.