Energy Management of Smart Buildings Using Graph Neural Network

Document Type : Original Article

Authors

Tarbiat Modares University

Abstract

Buildings account for a significant portion of global energy consumption, with HVAC, lighting, water pumping, and other building-related energy requirements contributing to a considerable fraction of energy consumption. As the demand for energy continues to escalate and concerns about primary energy sources grow, effective energy management in buildings has become a crucial challenge. To tackle this issue, machine learning algorithms have been widely used for predicting building energy consumption. However, these algorithms often overlook the interdependence and impact of different building zones. In this article, we propose a graph convolutional network (GCN) approach for predicting energy consumption in smart buildings. The proposed method effectively models the energy consumption pattern in different zones of a building and considers the influence of neighboring zones. We evaluated the GCN model using the CU-BEMS dataset, which includes energy consumption data from various zones of a 7-story building. The experimental results demonstrate that the proposed GCN method can predict the energy consumption of a specific floor with 5 zones, for only 0.6 of mean squared error (MSE)

Keywords

TABRIZ JOURNAL OF ELECTRICAL ENGINEERING

Articles in Press, Corrected Proof
Available Online from 11 May 2024

Files

Share

How to cite

Statistics