The structure method for the background value in grey forecasting model has an important influence on the forecasting precision. This paper defines non-equidistant GM(1,1) power model and makes a series of researches about it. Based on the error analysis of the GM(1,1) power model, the discrete function with non-homogeneous exponential law is used to fit the accumulated sequence, which can propose a structure method of the background value in GM(1,1) power model. The example indicates that the precision of fitting and forecasting of the non-equidistant GM(1,1) power model with optimized background value is significantly improved, which is valuable for both theory and practice.