Ontology-driven LLM framework for knowledge graph in smart buildings

Abstract

The rapid data growth in smart building environments requires advanced tools to integrate, interpret, and utilize this information effectively. Smart buildings generate vast and heterogeneous data streams, including sensor readings, occupancy metrics, and environmental conditions, which are critical for optimizing energy efficiency, enhancing occupant comfort, and enabling predictive maintenance. However, the lack of a structured approach to automatically connect and contextualize these data sources limits the insights that can be derived. To address these challenges, this paper presents a framework for assessing data in a knowledge graph by automatically retrieving entities and establishing relationships from diverse data sources, incorporating metadata standards and time-series data relevant to smart buildings. A standard ontology from the domain is used to drive the experiment, enabling the automatic construction of a semantic graph-based model for a real-world smart building environment. The framework s objective is to ensure information is comprehensible as a preliminary step to intelligent decision-making in data-driven smart buildings, enabling applications like fault detection, performance measurement, and energy auditing. The proposed approach explores the potential of large language models (LLMs) to automate data integration, reducing reliance on experts. This paper addresses existing literature gaps on metadata mapping and lays the groundwork for future advancements in digital twin technologies for smart building applications.