Content area
This research introduces a modular data management strategy for IoT-based building monitoring systems to enhance the evaluation of energy performance in future smart districts. By developing a flexible semantic data modeling framework, we integrate Industry Foundation Classes (IFC) data from the planning phase and sensor networks into Resource Description Framework (RDF) graphs using the Brick Schema ontology and a plant identification key scheme for data points, creating a modular knowledge graph. Additionally, we embed sensor metadata as a vector index, enabling a Large Language Model (LLM)-assisted graph query system with contextual awareness of the measurement infrastructure. Furthermore, we employed an Agentic Graph Retrieval-Augmented Generation (Agentic GRAG) technique powered by LLMs to facilitate natural language interaction and automate data processing. Testing on an energy data assessment platform testbed demonstrated improved operational efficiency through natural language queries. Our results highlight the effectiveness of the proposed data management approach, showing that the choice of LLM and adaptive prompting significantly affects system performance. In addition, incorporating relevant examples and prior chat history enhanced system responsiveness. This approach advances data analysis and decision-making by enabling efficient querying of knowledge graphs. In contrast, the knowledge graph’s modularity ensures scalable and adaptable data-modeling pipelines for building operators.