BuildSTG: A Multi-building Energy Load Forecasting Method using Spatio-Temporal Graph Neural Network

TL;DR

BuildSTG introduces a spatio-temporal graph neural network for multi-building energy load forecasting, effectively capturing spatial dependencies and providing interpretability, outperforming traditional methods on real-world data.

Contribution

The paper presents a novel multi-building energy prediction method using graph neural networks that incorporate building similarities and environmental factors for improved accuracy.

Findings

Outperforms baseline models like XGBoost and GRU in accuracy.

Effectively captures spatial dependencies among buildings.

Provides interpretable insights into building similarities.

Abstract

Due to the extensive availability of operation data, data-driven methods show strong capabilities in predicting building energy loads. Buildings with similar features often share energy patterns, reflected by spatial dependencies in their operational data, which conventional prediction methods struggle to capture. To overcome this, we propose a multi-building prediction approach using spatio-temporal graph neural networks, comprising graph representation, graph learning, and interpretation. First, a graph is built based on building characteristics and environmental factors. Next, a multi-level graph convolutional architecture with attention is developed for energy prediction. Lastly, a method interpreting the optimized graph structure is introduced. Experiments on the Building Data Genome Project 2 dataset confirm superior performance over baselines such as XGBoost, SVR, FCNN, GRU, and Naive, highlighting the method's robustness, generalization, and interpretability in capturing meaningful building similarities and spatial relationships.