Building energy consumption prediction for campus accommodation buildings based on spatial temporal graph convolution networks

The Net Zero Building (NZB) strategy has been regarded as the fundamental pathway to achieve sustainable cities and communities and to amend climate change. This necessitates an accurate understanding of the energy consumption of buildings which serves as a baseline reference for any energy planning or building retrofit. Artificial Intelligence (AI) approaches have been well documented in predicting energy consumption of buildings with credits of accuracy and efficiency. However, most of the studies focused on a single building which ignored or overlooked the interdependencies of buildings, especially for those buildings with the same group of users such as educational campuses where students and university staffs usually share the facilities and infrastructures across buildings. Predicting energy consumption independently significantly affects the accuracy of AI models. To fill this research gap, this study proposes a spatial-temporal graph convolutional network (STGCN) algorithm to predict the hourly energy consumption of campus buildings in northern England. To evaluate the feasibility of the STGCN algorithm, several popular AI algorithms were also employed for comparison. The results indicated that STGCN can significantly improve the prediction performance that conventional machine learning algorithms.