AI Framework Automates Building Management System Data Mapping

Building Management Systems (BMS) are crucial for optimizing energy use and operations in modern buildings. However, the lack of standardization in BMS data across different manufacturers creates significant challenges for integration and data utilization. The Brick schema offers a standardized ontology, but mapping BMS points to its 936 classes is complex due to the sheer number of classes, limited LLM domain knowledge, and the extensive manual effort required for verification. To address these issues, researchers have developed Brick-DICL, a two-stage dynamic in-context learning framework. This framework leverages a metadata-RAG component to retrieve relevant examples, thereby enhancing the LLM's understanding of the domain. It also includes a class-RAG component that helps narrow down the vast number of potential Brick classes. Furthermore, Brick-DICL incorporates a multi-LLM filtering mechanism. This system compares predictions from multiple models and flags classifications with low confidence for human review, which significantly reduces the need for manual verification. The framework is applicable across various BMS manufacturers and metadata formats, demonstrating improved classification accuracy and efficiency in digital building onboarding.