It has lower accuracy in predicting mid-long term electricity consumption of power system by using NGM (1,1, k) gray model. For that a method of adding a correction value to revise the model coefficients is proposed. The least sum of square error between the fitting values and the actual values is used to solve the correction value, the raw data is processed by using the buffer operator, and the improved gray electricity consumption forecasting model is established. The improved model is used to forecast the mid-long term electricity consumption in two regions and the forecasting results show that it effectively enhances the prediction accuracy.