Graph attention spatio-temporal graph neural network based human–light interaction modeling for indoor lighting environments

Abstract

In indoor light environment design, the dynamic interaction between human behavior and lighting systems has an important impact on environmental perception results. However, existing methods are difficult to describe its spatial structure and temporal evolution characteristics at the same time. To this end, this paper constructs a human-light spatio-temporal relationship modeling method based on graph attention spatio-temporal graph neural network, which uniformly represents human body nodes and lamp nodes as spatio-temporal graph structures, and introduces heterogeneous graph attention and adaptive temporal attention mechanisms for joint modeling. The study conducted experimental verification on real indoor scene data for 7 consecutive days and a total of 302,400 time steps. The results show that the proposed model achieves MAE of 0.096 ± 0.010, MSE of 0.017 ± 0.004, RMSE of 0.130 ± 0.015, and R² of 0.943 ± 0.018 in the environment perception prediction task, which is significantly better than various comparison models. At the same time, the environmental perception results show stable and continuous evolution characteristics in the space and time dimensions. Research shows that modeling of human-light spatiotemporal relationships based on graph attention can effectively improve the accuracy and interpretability of indoor light environment perception, and provides important methodological support for intelligent light environment design and human factor-driven control.