A Survey on Semantic Modeling for Building Energy Management

arXiv:2404.11716 · cs.AI · Submitted 2024-04-17 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: AI Radio. Generated commentary on the latest Artificial Intelligence papers.

Tom: Next we'll be talking about the paper "A Survey on Semantic Modeling for Building Energy Management".

Jane: The paper was written by Miracle E. Aniakor, Vinicius V. Cogo and Pedro M. Ferreira from LASIGE and Faculdade de Ciências, Universidade de Lisboa, Portugal and DI, Faculdade de Ciências, Universidade de Lisboa, Portugal.

Tom: Stay tuned as we take you through the paper and discuss its implications.

Paper discussion segment 2: Tom: We've got the context of why this survey exists, so now let's talk about what the authors actually found in "A Survey on Semantic Modeling for Building Energy Management."

Jane: They looked at a massive amount of work—sixty-one different semantic models and analyzing over twenty use cases.

Tom: It seems like the big takeaway is that current semantic models are really good at describing physical things, like the structure of a building or its sensors.

Lu: But, Lu sees that’s often where the limitations start; it's great knowing what a sensor is, but that' not enough to figure out how to optimize energy use based on occupant behavior.

Meng: That gap between physical assets and operational goals is exactly what they found—the engineering reality is that systems are good at mapping hardware, but bad at making the decision-making logic.

Lalam: I think it’s a crucial observation because we need the digital twin to understand the 'why' behind the data, not just the 'what'.

Tom: So, how do they bridge that gap? It seems like they rely heavily on reusing existing ontologies or extending them.

Jane: Yes, instead of building one massive perfect model, they are using integration and specialization to patch together these diverse pieces of solutions.

Lu: I find the concept of 'reuse' particularly compelling; it suggests that we shouldn't reinvent the wheel for every specific BEM task if a core concept already exists in another ontology.

Meng: Pragmatically, reusing an existing model is much faster than building one that has to cover both hardware and software logic from scratch.

Lalam: It’s about finding the shared conceptual DNA between various systems, recognizing that we need many specialized pieces to create a whole picture of context.

Paper discussion segment 3: Tom: We've established that models are great for physical assets but not fully equipped for abstract operational concepts, so let's talk about how the authors measured this in "A Survey on Semantic Modeling for Building Energy Management."

Jane: They developed a framework to measure this called Ontology Evidence Completeness, or OEC.

Tom: OEC is a really neat way of saying whether or not explicitly tracing the task-relevant operational concepts—like key performance indicators—to the ontology classes actually used in the study.

Lu: I think it’s revolutionary because most academic papers just claim they used an ontology, but OEC forces them to show the actual class-level mapping evidence.

Meng: For an engineer, this is gold because it tells us where a model might be missing something critical—it's not just a label; it's proof of coverage.

Lalam: Lalam thinks this method is essential for ensuring that when we build AI applications, we are trusting the data representation at a granular level.

Tom: The paper also quantifies how much more work is needed using two metrics: Ontology Instantiation Rate, or OIR, and Necessity to Extend, or NTE.

Jane: OIR tells us how broad the use was relative to the total classes available in the ontology, while NTE measures the percentage of required concepts that had to be created because they weren't there.

Lu: I like how those numbers reflect a clear cost-benefit analysis; if you need high NTE, it means significant manual intervention or a new ontology is needed.

Meng: From a deployment perspective, if an application has high NTE, it suggests that the core model is too narrow for certain tasks and needs to adapt to the real-world requirements.

Lalam: It shows us where we have semantic gaps versus where we have robust frameworks that can handle complex operational demands.

Paper discussion segment 4: Tom: We’ve looked at the measurements, so let’s discuss what this paper suggests for future directions in "A Survey on Semantic Modeling for Building Energy Management."

Jane: The authors aren't proposing a new ontology, but they are pointing out how we can make better use of existing ones.

Tom: They highlighted that applying these models requires a lot of clever integration—reusing one model while extending another—to cover the full scope of BEM tasks.

Lu: I think the future is less about finding one perfect, monolithic ontology and more about sophisticated alignment tools that allow modular pieces to talk seamlessly to each other.

Meng: We need tools that can handle this kind of integration automatically, so that's where engineering efforts should be focusing—on seamless semantic stitching.

Lalam: Lalam thinks the future is about creating a cohesive narrative across all the components, ensuring that the entire system behaves as one intelligent entity, not just a collection of separate parts.

Tom: So, instead of just looking at how many classes an ontology has, we need to look at how those models are combined to address specific gaps.

Jane: The paper suggests moving away from isolated solutions toward a more coordinated approach that leverages the strengths of various existing models across different domains.

Conclusion: Tom: We’ve covered a lot of ground today, seeing exactly where "A Survey on Semantic Modeling for Building Energy Management" stands in its findings.

Jane: It really shows us that while semantic modeling is absolutely vital for creating smart building systems, it isn't a silver bullet; it requires careful management and consistent application.

Tom: The survey confirmed that existing models are strong with physical reality but weak on the abstract operational intelligence needed for decision-making.

Lu: And I think we’ve seen how this suggests that the path forward is heavily reliant on modularity and intelligent alignment of existing structures.

Meng: It definitely shows us where the practical engineering challenge lies—we can't just deploy a single standard; we need to integrate multiple, specialized models.

Lalam: Lalam feels that by defining these gaps, this paper has laid the groundwork for a cultural shift toward genuine contextual understanding in architecture and technology.

Tom: We’ve seen how this work addresses the need for interoperability, using OIR and NTE to measure success.

Jane: It's a comprehensive look at why simply stating an ontology wasn' not enough to understand the depth of semantic coverage required for BEM.

Tom: So, as we wrap up our discussion on "A Survey on Semantic Modeling for Building Energy Management," we hope this provides listeners with a roadmap for how to think about building data in a way that truly empowers smarter systems.

Lu: I'm excited to see the possibilities when AI can finally access those clearly defined semantic relationships.

Meng: I'm ready to start looking at the integration tools that make these complex models work together practically.

Lalam: I hope this helps us move towards a more coherent and intelligent future environment for all of us.

LASIGE · Faculdade de Ciências, Universidade de Lisboa, Portugal · DI, Faculdade de Ciências, Universidade de Lisboa, Portugal

cs.AI

Submitted: 2024-04-17

Updated: 2026-09-04

Comments: 49 pages, 7 figures, 5 tables

Code: https://github.com/Miracle-labmirx/prisma_oriented_screening_pipeline

Project page: https://lambdamusic.github.io/Ontospy

License: http://creativecommons.org/licenses/by/4.0/

Importance score: 89/100

The gist: Building Energy Management (BEM) is critical for reducing carbon emissions and energy consumption in the building sector, yet its development is hindered by "heterogeneous data models" and "semantic

Key concepts

Ontology Evidence Completeness (OEC)
A framework used to measure whether task-relevant operational concepts, such as key performance indicators, are explicitly traced to the ontology classes used in a study. It requires researchers to provide actual class-level mapping evidence.
Ontology Instantiation Rate (OIR) and Necessity to Extend (NTE)
OIR measures how broadly an ontology was utilized relative to all available classes. NTE measures the percentage of required concepts that had to be created because they were not present in the existing models.
Semantic Modeling Gap
The identified limitation where systems are proficient at mapping physical hardware, such as sensors, but lack the necessary logical framework to optimize energy use based on abstract operational goals or occupant behavior.

Terminology

Summary

Building Energy Management (BEM) is critical for reducing carbon emissions and energy consumption in the building sector, yet its development is hindered by heterogeneous data models and semantic interoperability issues arising from IoT technologies. This survey addresses these limitations by examining semantic modeling for BEM during the operational phase, providing a comprehensive analysis of existing ontological efforts to clarify their capabilities and identify necessary directions for creating more generalized, autonomous, and context-aware BEM systems.

How it works: Review Methodology

The survey employs a hybrid methodology combining a PRISMA-oriented database search with complementary source identification. This process yields an ontology-relevant corpus of 150 references, including 109 peer-reviewed studies and 41 technical/contextual support sources. The selection criteria focus on studies that make an explicit ontology-based contribution to building operational tasks or BEM applications, excluding non-graph-based representations like UML or relational schemas. This rigorous approach allows the the authors to move beyond merely noting ontology availability and instead examine which operational concepts are documented at the class level, which remain unsupported, and what modeling strategies are used to fill these gaps.

How it works: Key Findings and Quantitative Analysis

The analysis of sixty-one semantic models reveals that current models more consistently represent physical building structure than abstract or dynamic operational concepts. The study introduces two quantitative metrics to evaluate this coverage: Ontology Instantiation Rates (OIR) and Necessity to Extend (NTE). These metrics quantify the breadth of ontology use at the parent-class level and the proportion of required concepts that needed newly created classes, respectively.

The survey identifies several recurring patterns in how semantic gaps are addressed across different BEM use cases:

  • Single-ontology scenarios: Often rely on an existing model's scope but require external inheritance or application ontology extension to cover missing concepts.

  • Two- and Multiple-ontology scenarios: Frequently employ integration, where complementary models are combined to extend the descriptive capacity of a single model when it cannot adequately capture all relevant building concepts.

How it works: Limitations in Semantic Coverage

A significant limitation identified is that existing semantic models do not provide complete coverage for abstract operational concepts. These include:

  • Key performance indicators (KPI) and their calculation logic.

  • Service execution and control logic.

  • Computational workflows and optimization tasks.

The analysis shows that while core models like SAREF, Brick, and SSN/SOSA provide foundational support for physical entities (e.g., devices, sensors), the semantic coverage in BEM use cases remains uneven regarding these abstract operational processes.

How it works: Future Directions

To address these limitations, the survey suggests several paths for future research and development in BEM semantic modeling:

  1. Systematic Comparison: Evaluating established models like SAREF and Brick using transparent concept-to-class mappings.

  2. Coordinated Development: Moving away from isolated new ontologies toward improved ontology alignment and extension mechanisms to reduce fragmentation.

  3. Integrating AI Agents: Leveraging advancements in natural language processing (NLP) and AI agents, where semantic layers define not only what data is accessible but also what they can interpret, explain, and act upon.

The conclusion emphasizes that realizing the full potential of semantic modeling for BEM requires moving beyond merely describing physical assets to capturing the computational and decision-support processes that are essential for developing truly interoperable, generalizable, and context-aware building systems.

Improvements for AI systems

Based on a rigorous analysis of this survey, I have identified several critical improvements needed to elevate current BEM AI systems from simple data processing tools to truly autonomous, context-aware decision-support agents. The focus is on bridging the gap between physical asset representation and abstract operational logic.

Improvement: Implement an internal OEC assessment module within the AI's knowledge graph processing pipeline. This module must systematically trace task-relevant operational concepts (e.g., thermal comfort state, energy efficiency KPI) to the specific ontology classes used in any given data stream or model integration attempt.

Improved Capability: The AI system can self-diagnose its semantic limitations, identifying precisely which required operational concepts are not explicitly mapped to existing ontology classes. Instead of failing silently when encountering novel conditions, the the AI will flag a Semantic Gap, allowing it to trigger a targeted search for necessary ontology reuse or initiate an application-specific extension workflow.

Improvement: Replace simple time-series data ingestion with a multi-stage semantic mapping layer that formally links raw sensor readings (e.g., 25 C) not just to a value, but to an ontological entity (e.g., sosa:Observation of saraf:TemperatureSensor on saraf:Device). This requires structuring data using foundational models like SSN/SOSA for sensors and SAREF for device/service contexts.

Improved Capability: The AI can perform complex, multi-variable reasoning across heterogeneous systems. For instance, it can correlate a temperature rise (sosa:Observation) with a specific HVAC operation (saraf:Service activation) and determine the resulting impact on occupant comfort (e.g., oncom:ThermalComfortState), rather than merely recording two data points.

Improvement: Move beyond simply using ontologies for physical assets; explicitly model the functional relationships between performance metrics, evaluation processes, and responsive actions (the KPIs-Assessments-Services layer). This requires integrating specialized ontologies like EM-KPI and BOP into the decision-making architecture.

Improved Capability: The AI can execute sophisticated optimization tasks. Instead of running a heuristic algorithm, it can evaluate its proposed solution against predefined KPI ontology classes (e.g., Max Energy Consumption) and simulate the resulting assessment (bop:Assessment) before triggering the necessary control action (saraf:Service execution), ensuring that the decision is fully justifiable by its semantic framework.

Improvement: Implement a dynamic ontology orchestration engine capable of selecting, integrating, and specializing multiple ontologies (Multi-Ontology Strategy) based on the specific operational task requirements (e.g., combining Brick for physical location/equipment with SARGON2 for grid interaction). This engine must automatically apply strategies like external inheritance or application-specific extension when a concept is missing in the core ontology.

Improved Capability: The AI can handle complex, cross-domain BEM scenarios (e.g., Optimal cooling under fluctuating renewable energy prices). It will dynamically pull necessary concepts from various specialized ontologies, combine them into a unified semantic model for the task, and execute the required control logic without requiring pre-programmed rules for multiple distinct scenarios.

Abstract

Building Energy Management (BEM) is central to reducing energy use and CO2 emissions in the building sector. Although IoT technologies now provide extensive operational data, heterogeneous data models, device descriptions, and contextual representations continue to limit semantic interoperability, limiting the development of generalisable, autonomous, context-aware BEM applications. Ontologies address this challenge by providing structured, machine-interpretable representations of building data, systems, and operational context. This survey examines semantic modelling for BEM during the building operational phase. It reviews 60 semantic models and analyses more than 20 ontology-based BEM use cases. It further quantifies Ontology Instantiation Rates (OIR) and missing concepts across those use cases. To support evidence-based assessment of ontology use, we introduce the notion of Ontology Evidence Completeness (OEC), a measure of whether studies explicitly map operational concepts to the ontology classes used to represent them. Findings show that current semantic models more consistently represent physical building structure, technical systems, sensing devices, and observable operational data than abstract and dynamic operational concepts. Concepts such as key performance indicators, assessments, services, control logic, optimisation tasks, and computational workflows remain less consistently covered. Applied BEM studies therefore frequently depend on ontology reuse, integration, specialisation, external inheritance, or application-specific extension to address coverage and interoperability gaps across BEM. By synthesising these patterns, this survey clarifies the capabilities of existing semantic models and identifies directions for more interoperable, generalisable, and context-aware BEM systems.

Related papers