Author(s)

Ziqing Wei, Ran Xiao, Gan Yang and Xiaoqiang Zhai

Corresponding Author

Ziqing Wei

ABSTRACT

Energy consumption in buildings takes a large part in the total energy consumptions in modern society. Energy consumption prediction with high precision is necessary for building energy conservation. Meanwhile, due to the difficulty of data acquisition, the influence of model dimensionality reduction on prediction accuracy needs further studies. In this paper, a particle swarm optimization - supported vector regression (PSO-SVR) algorithm is adopted to construct the prediction model. The energy consumption data of an office building in Shanghai from 2017 to 2018 is used to build and test the prediction model. Four different combinations of selected input variables are created and evaluated by the PSO-SVR algorithm to investigate the effect of different input selections. The results show that PSO-SVR algorithm can improve the model performance. The combination investigation shows that the features of indoor facilities influence the prediction performance greater than out door features and time-related features.

KEYWORDS

Building energy consumption, machine learning, hourly prediction, particle swarms optimization