# Times Series Forecasting for Urban Building Energy Consumption Based on Graph Convolutional Network

**Authors:** Yuqing Hu, Xiaoyuan Cheng, Suhang Wang, Jianli Chen, Tianxiang Zhao, Enyan Dai  
**arXiv:** [2105.13399](https://arxiv.org/abs/2105.13399)  
**Date:** 2021-05-27  
**Summary page:** https://www.summarizepaper.com/en/paper/2105.13399v1/

## Key points

- The building industry accounts for over 40% of energy consumption in the United States
- Many cities are implementing ambitious energy-saving strategies by retrofitting existing buildings and constructing new sustainable communities
- Current urban building energy models have limitations in capturing inter-building interdependency due to their dynamic and non-linear characteristics
- A team of researchers proposed a novel data-driven UBEM that synthesizes solar-based building interdependency and spatial-temporal graph convolutional network (ST-GCN) algorithm
- The feasibility of the proposed model was tested by comparing its performance with other common time-series machine learning models, and the ST-GCN model outperformed all others
- The physical knowledge embedded in the model was well interpreted, highlighting how data-driven models integrated with engineering or physical knowledge can significantly improve urban building energy simulation.
- This study provides valuable insights into developing accurate UBEMs that can support designing energy-efficient communities for sustainable urban development.

## Summary

The world is rapidly urbanizing, and the building industry accounts for over 40% of energy consumption in the United States. To address this issue, many cities are implementing ambitious energy-saving strategies by retrofitting existing buildings and constructing new sustainable communities. However, current urban building energy models (UBEM) have limitations in capturing inter-building interdependency due to their dynamic and non-linear characteristics. This can lead to inaccurate urban energy modeling as these models either ignore or oversimplify building interdependencies. To address this research gap, a team of researchers proposed a novel data-driven UBEM that synthesizes solar-based building interdependency and spatial-temporal graph convolutional network (ST-GCN) algorithm. The study used a university campus located in downtown Atlanta as an example to predict hourly energy consumption. The feasibility of the proposed model was tested by comparing its performance with other common time-series machine learning models. The results showed that the ST-GCN model outperformed all others, indicating its potential for accurately predicting urban building energy consumption. Additionally, the physical knowledge embedded in the model was well interpreted, highlighting how data-driven models integrated with engineering or physical knowledge can significantly improve urban building energy simulation. Overall, this study provides valuable insights into developing accurate UBEMs that can support designing energy-efficient communities for sustainable urban development. By addressing limitations in current models through innovative approaches like ST-GCN algorithms, we can create more effective solutions for reducing energy consumption in buildings and promoting sustainable living practices globally.