Day-ahead Prediction Method of Hourly Building Energy Consumption in Transition Season

Abstract

Reliable energy consumption prediction methods play a key role in optimizing the air-conditioning system operation and energy management in public buildings. In order to predict the building energy consumption in transition season and improve prediction accuracy, this paper proposes and introduces a day-ahead prediction model based on key feature search. The proposed indirect key feature search is carried out by using the similarity relation between forecast daily features and historical factors. The proposed model is applied in an office building with the scope to manage the day-ahead prediction of hourly total term. Results show that the key feature search can improve the accuracy by 14.5% of forecast days in spring and 4.9% in Autumn. However, the traditional method is still work to select the training set for the energy consumption prediction in summer. In addition, the proposed search method is most useful for improving the application of predictive models in energy management platforms.