In the actual process of forecasting, the credibility of the result of traditional urban space load density prediction method depends on a number of sample data. But, in the actual, collecting a complete feasible data is quite difficult. Therefore this paper puts forward a way which combines the group decision-making method of mixed language information and the BP neural network to forecast the city power load density. This way uses group decision-making method of mixed language information to get the score value of the economy, population, geographic environment in all urban district, then by using BP neural network to train the score and the load density, after that utilizing the net to predict the load density of pending district. The result shows that not only the computation process can get rid of the problem which need large collection of specific indicators quantitative data but also the result is very good.