A reliable prediction of energy consumption is crucial for a reasonable building energy management. Considering the uncertain principles of annual electricity consumption with limited datasets, a modified grey interval prediction model abbreviated as BOGIM(1,1) is proposed in this paper. Firstly, the changing patterns of annual series were detected, in order to lower the uncertainty. Afterwards, the predicted intervals were obtained with modified BOGIM(1,1), in which various weakening and enhancing buffer operators were added simulate different future operation scenarios. Finally, the adaptability of this model is summarized based on recognized patterns and predicted accuracy. Specifically, 92 office buildings in Beijing of China were adopted to test the BOGIM(1,1) model. Results show that this proposed model outperforms the traditional GM(1,1) by improving the prediction accuracy for almost 90% of the buildings up to 18.45%, and it is more applicable for target-oriented energy policies.
Keywords annual electricity consumption, interval prediction, office buildings, grey model, buffer operator