In the energy demand forecast, s ince national energy demand contains complex elements, it is virtually impossible to mak e an accurate forecast of energy demand with one single fo recasting model. The combined forecasting model applied in this paper is constructed through comparing the predictive effects brought about by grey GM (1,1) model, regression - type BP neural network (BPNN), triple exponential smoothing (TES) and linear regr ession model (LRM) on China’s energy demand. Variable Weighted Combined Forecasting method is used to raise forecast accuracy. China’s energy demand forecast is practiced based on synchronous GDP, fixed assets investment , power generation capacity and population size. The calculation results show that GM(1,1) model and regression - type BP neural network forecast method’s forecast accuracy is higher than others, and that the forecast model combined by thes e two forecast method’s forecast data are more accurate and robust