AI-Driven Indoor Environmental Monitoring for Smart Buildings Using Graph Neural Networks

Air pollution has been identified as one of the world’s leading environmental health risks, impacting millions globally. Ensuring protection against its harmful effects has become a pri- mary concern, especially in indoor environments where people spend the majority of their time. This thesis presents a novel framework focused on the estimation and spatial interpolation of PM2.5 concentrations within indoor spaces, enhancing the capability to monitor and mitigate air pollution in real time. Our approach combines recurrent and feed-forward deep learning models, specifically leveraging a Graph Neural Network (GNN) model that captures spatiotem- poral dependencies and integrates contextual features unique to indoor environments. The model is engineered to process sensor data from different indoor locations while considering complex airflow obstacles, including walls, HVAC vents, air purifiers, and their operational statuses. Furthermore, this framework can be extended to monitor additional environmental param- eters within indoor spaces, such as PM10, CO2, temperature, and noise, broadening its utility in creating healthier indoor environments and saving costs for facility managers. Finally, we propose a scalable cloud-based deployment strategy to support efficient, real-time monitoring and enhance integration across multiple client sites.