Advanced graph embedding for intelligent heating, ventilation, and air conditioning optimization: An ensemble learning-based recommender system

Abstract

This study introduces a robust and scalable software architecture designed for real-time data ingestion, processing, and user interaction within a smart building setting. Utilizing advanced graph embedding techniques combined with ensemble learning models, we developed a recommender system tailored for Heating, Ventilation, and Air Conditioning (HVAC) optimization in Shenzhen Qianhai Smart Community. We employed a mixed-methods approach, including the generation of synthetic multivariate time series data, data preprocessing, statistical correlation analysis, and the implementation of GraphSAGE, Graph Attention Networks (GAT), and Node2Vec for graph embedding. The ensemble learning framework integrated Decision Trees, Random Forests, Gradient Boosting Machines (XGBoost), and Neural Networks to enhance prediction accuracy. Our findings demonstrate that the proposed architecture maintains high performance under increased loads, with key sensor correlations effectively managed to optimize HVAC operations. The recommender system achieved a 51 % reduction in energy consumption of chilled water pumps and a 15 % increase in occupant satisfaction by providing personalized HVAC settings. These results highlight the significance of integrated system designs and data-driven strategies in developing intelligent building management solutions. The study contributes actionable insights into system scalability and user-centric environmental controls, paving the way for future research in real-world implementations and advanced analytical techniques.