C45 -- The Virtual Building Manager - A Large Language Model Powered Intelligent Building Management System

Artificial intelligence (AI) technologies have the potential to radically transform the way buildings are managed and operated. As Building Management Systems (BMS) are expected to have more intelligent functions and enhanced human-machine interaction, there is an emerging need for a new framework designed to effectively address and manage the increasing system complexity and demand for energy conservation and decarbonization. This study shows that large language models (LLMs) could provide the foundation for this next generation of intelligent BMS, leveraging interactive AI agents to probe the BMS for the right information, and provide bi-directional information exchange between personnel and BMS (human-in-the-loop). In this study, a novel interactive AI agent is designed to act as a virtual extension of a building manager. This AI agent has the ability to comprehend natural language instructions from human operators and provide relevant information to assist in final decision-making and operations. Additionally, it can process real-time sensory data from building installation systems, such as HVAC, enabling the activation of various functions and the application of data-driven methodologies. This framework provides a more intuitive approach to intelligent building management. To validate the practical feasibility and effectiveness of this approach, we have implemented and showcased the AI agent specifically for the use case of Fault Detection and Diagnostics (FDD) in HVAC systems – a prevalent challenge in predictive maintenance and optimal operations, with the goal of identifying and diagnosing faults using raw sensor data to reduce energy waste and carbon emissions. The expected results underscore the potential of large language model empowered AI agents in BMS, offering a disruptive perspective on data utilization and interactive management technologies.