| Résumé | The integration of Artificial Intelligence (AI) into Building Automation Systems (BAS) has potential to enhance operational efficiency, adaptability, and predictive capabilities. Traditional BAS rely on static rule-based automation, limiting their responsiveness to dynamic conditions. This paper proposes a modular AI-driven framework that utilizes machine learning techniques on a commercial BAS dataset to optimize energy usage, improve occupant comfort, and could be easily adapted for demand response applications. Our framework, designed for seamless integration with existing infrastructure, enhances traditional BMS through generative AI agents employing tool calling and Retrieval-Augmented Generation (RAG), offering robust insights and advanced data exploration capabilities for facility managers. By combining the proven reliability of traditional BAS with the advanced capabilities of modern AI techniques, our modular system significantly advances building performance and sustainability. We evaluate cloud, Semi-Local, and fully local AI deployments, highlighting trade-offs in accuracy, latency, and privacy. Results show that while cloud models offer high accuracy, Semi-Local architectures provide a balance between performance and security. Advances in lightweight LLMs and edge computing indicate growing feasibility for fully local AI solutions. This framework demonstrates the potential of AI-enhanced BAS, offering a scalable approach to smart building management. |
|---|
