AI
Survey identifies critical gaps in semantic models for smart building energy management
A comprehensive review of 60 ontologies and 20+ real-world BEM applications reveals that semantic models struggle to capture abstract operational concepts despite strong support for physical infrastructure.
As buildings account for substantial global energy consumption and carbon emissions, the push to make them smarter and more efficient has intensified. While IoT sensors now flood buildings with operational data, the lack of standardized semantic frameworks continues to hamper the development of truly autonomous, context-aware energy management systems that can work across different building types and vendors.
The semantic interoperability challenge#
A new survey from arXiv on semantic modeling for Building Energy Management (BEM) examines how ontologies—machine-readable structured representations of building data and systems—can bridge this gap. The research analyzed 60 different semantic models and studied how they were applied in over 20 real-world BEM projects, uncovering significant blind spots.
The core problem lies in semantic interoperability: heterogeneous data formats, varied device descriptions, and inconsistent ways of representing building context make it difficult to build reusable systems. Ontologies, which provide standardized vocabularies and relationships for how building components interact, are meant to solve this. Yet their actual deployment reveals constraints.
What ontologies do well—and where they fall short#
Current semantic models perform reasonably well at capturing tangible, physical aspects of buildings. They consistently represent the building's structural layout, mechanical and electrical systems, sensors, and raw operational measurements from those devices.
However, the analysis introduces a concept called Ontology Evidence Completeness (OEC)—a measure of whether researchers explicitly map real operational concepts to the ontology classes meant to represent them. Using this lens, the survey found persistent gaps:
- Underrepresented: Key performance indicators, performance assessments, service definitions, control logic, optimization tasks, and computational workflows
- Well-covered: Physical structure, technical systems, sensing hardware, and observable data
This means that while we can model what a building's systems are doing, we struggle to semantically represent why we're measuring things, what goals we're optimizing for, or how the system should respond to different conditions.
