New Framework Unifies Building Occupancy Modeling Research

Researchers from a collaborative team led by Geoffrey Bourgoin and Karim Beddiar have proposed a comprehensive analytical framework to standardize the fragmented field of building occupancy modeling. Published in Energy Informatics in April 2026, the study addresses the growing complexity of occupancy-related research, where heterogeneous terminology and analytical perspectives have hindered consistent comparison of results. The paper introduces a structured approach that organizes the literature according to six complementary dimensions: occupancy scale, information level, acquisition strategy, modeling approach, temporal dimension, and system constraints. This unified perspective aims to clarify the respective contributions and limitations associated with each dimension, providing a common conceptual basis for future research and practical application in building energy management.

The motivation for this work stems from the significant role of buildings in global energy consumption. In the early 2000s, building energy consumption in developed countries accounted for up to 40% of total energy use, a proportion that remains substantial today. In France, building use reached 43% of total energy consumption in 2024. Within the residential sector, Heating, Ventilation, Air-Conditioning and Lighting (HVAC-L) systems account for an average of 51% of total household energy consumption, with significant geographical disparities observed. In the United States and France, HVAC-L systems represent 57 to 65% of total consumption, whereas in China, they account for approximately 30%. These figures underscore the potential impact of optimizing building operations through better understanding of occupancy dynamics.

Occupancy information is defined in the study as any data describing the presence, estimation, localization, tracking, activity, comfort, behavior, and personal information of occupants that can be acquired, inferred, or modeled for building operation and control. Integrating such information enables Occupancy-Centric Control (OCC) strategies, which have shown strong potential to reduce lighting and HVAC energy consumption by up to 75% and 55%, respectively. The literature on this topic has evolved significantly over the past two decades, driven by technological advances in affordable sensing and data collection, as well as increasing interest in high-performance buildings. However, previous reviews often adopted heterogeneous categorizations centered on specific sensing technologies or modeling approaches, limiting cross-study comparison and the ability to identify consistent patterns across the field.

To conduct the review, the authors followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA 2020) guidelines to ensure transparency and reproducibility. The search was conducted in the SCOPUS database using the criteria ‘occupancy information AND building’ in titles, abstracts, and keywords, limited to English language and specific subject areas including engineering, computer science, and energy. From an initial pool of 1,670 papers, the team applied a multi-stage screening process. This included manual screening of titles and abstracts to exclude irrelevant topics such as robotics or traffic, followed by an automated prioritization step using a PUGH matrix. This matrix weighted criteria such as semantic proximity to occupancy synonyms, publication year, and paper type to identify a representative subset of studies for in-depth analysis.

After applying a threshold of 0.5 to the PUGH matrix scores, 116 unique papers were selected for retrieval. Sixteen of these were inaccessible, and a final manual verification was conducted to cross-check relevance. Ultimately, 93 papers were retained for inclusion in the systematic review. The proposed framework was then used as an analytical lens to review these papers from the perspective of occupancy scale. This approach allowed the researchers to identify consistent patterns, trade-offs, and limitations across different spatial levels of occupancy modeling, moving beyond simple classification to provide a structured synthesis of the field.

The framework defines five occupancy scales: city, building, multi-zones, zone, and occupant. City scale involves aggregated occupancy across multiple buildings or urban infrastructures, which is not the primary focus of this study. Building scale represents the building as a single entity, where occupancy information describes overall usage. Multi-zones scale covers groups of zones within a building, such as floors or areas sharing similar thermal behavior. Zone scale represents a single zone where indoor conditions are assumed to be homogeneous, such as a room or open-plan office. Occupant scale focuses on the individual occupant, capturing person-centered information such as activity and comfort that cannot be fully described by spatial aggregation alone. This hierarchical structure allows for a detailed analysis of how occupancy information is acquired and utilized at different levels of spatial resolution.

The analysis revealed that occupancy presence and estimation are the most widely investigated and practically exploitable information levels for energy-oriented applications, particularly at zone and multi-zones scales. At the building scale, presence and estimation account for 55% of the reviewed papers, with acquisition strategies relying on electricity consumption, environmental measurements, and network usage. These approaches provide coarse global indicators suitable for centralized building operation but lack the spatial resolution to capture heterogeneous occupancy distributions. In contrast, zone scale accounts for approximately two-thirds of all reviewed papers, reflecting the direct relevance of this level to HVAC control. At this scale, occupancy estimation dominates, accounting for about 60% of the papers, followed by presence at 15%. The high interest in zone scale is driven by its ability to offer a compromise between spatial resolution and system complexity, enabling more precise energy management.

Modeling approaches across the reviewed literature show a strong preference for black-box models, which are data-driven and learn relationships from data without explicit physical representation. These models, including machine learning and deep learning techniques, are widely used due to their ability to achieve high accuracy and handle multimodal inputs. However, they often require large labeled datasets and may suffer from limited explainability and reduced generalization to new environments. Gray-box and knowledge-based models, which combine partial physical knowledge with data-driven components or rely on expert rules, are less frequently reported but offer advantages in interpretability and adaptability. The study highlights that while black-box models dominate, hybrid strategies that integrate physical constraints with data-driven performance represent a promising direction for improving robustness and real-world applicability.

Privacy emerges as a central constraint shaping both acquisition strategies and modeling choices. Building-centric acquisition methods, such as using electricity consumption or environmental data, are generally more compatible with privacy regulations but restrict the granularity of occupancy information that can be inferred. Occupant-centric approaches, such as using cameras or personal devices, enable richer modeling of activity and localization but raise significant ethical and legal challenges. Indirect acquisition strategies, such as wireless signal disturbances, attempt to balance information richness and privacy preservation, though they remain sensitive to assumptions regarding device carriage and signal stability. The authors emphasize that future developments should explicitly quantify the trade-offs between privacy preservation, information loss, and control performance, embedding privacy-by-design principles across occupancy scales to ensure long-term acceptability.

The study concludes that effective occupancy-centric control cannot rely on a single sensing or modeling strategy. Instead, real-world deployment requires adaptive combinations of occupancy scales, information levels, and acquisition strategies. Zone scale occupancy estimation appears particularly relevant for HVAC optimization, while building scale information can provide a robust baseline for centralized control. The authors recommend that future research focus on cross-scale integration, improved generalization of models across buildings and contexts, and rigorous evaluation of privacy-preserving strategies. By providing a unified framework, this work offers a structured path toward scalable, privacy-aware, and energy-efficient building control systems, addressing the fragmented nature of previous research and facilitating more consistent comparison and interpretation of occupancy modeling studies.

At the multi-zones scale, the reviewed literature reveals distinct practical implementations that balance spatial resolution with infrastructure demands. Rather than deploying dedicated hardware across every room, studies at this scale rely predominantly on indirect occupant-centric strategies such as network connectivity, which accounts for roughly 70% of reviewed multi-zones papers. Researchers demonstrate various spatial segmentation methods, including dividing floor plans based on network access point coverage or architectural layouts. For example, partitioning a building floor into ten distinct zones based on facade orientation and tracking computer logins demonstrated a 9% reduction in annual cooling loads. Other approaches track occupants through corridor-mounted cameras to monitor room entries while mitigating private zone surveillance, or utilize footstep-induced floor vibrations across open-plan offices, though simultaneous foot traffic complicates individual movement tracking.

Within individual zones, physical sensing characteristics impose measurable performance boundaries on data-driven models. Although carbon dioxide concentration correlates strongly with occupant presence, concentration shifts can lag behind occupant arrivals by up to 30 minutes, and models trained on single rooms experienced accuracy reductions exceeding 10% when transferred across different spaces. Nevertheless, integrating zone-level occupancy estimation into active control systems achieved dramatic experimental reductions in energy demand, including reported decreases of up to 65% for cooling loads and 80% for ventilation loads.

At the occupant scale, comparative evaluations demonstrate sharp trade-offs among sensing modalities. Categorizing individual activities using wearable wrist and ankle sensors achieved 98% accuracy, while smartphone accelerometers and microphones reached 95% accuracy and computer vision models achieved 80%.

Subject of Research: A systematic review of building occupancy modeling frameworks and technologies

Article Title: A comprehensive framework for modeling building occupancy: technologies, methods, and research challenges

Article References: Bourgoin, G., Micheneau, K., Boodi, A., & Beddiar, K. (2026). A comprehensive framework for modeling building occupancy: technologies, methods, and research challenges. Energy Informatics, 9(1), Article 54. https://doi.org/10.1186/s42162-026-00660-7

Image Credits: AI Generated

DOI: 10.1186/s42162-026-00660-7

Keywords: building occupancy, energy efficiency, systematic review, HVAC control, privacy constraints, machine learning, sensing technologies, building automation, comprehensive, framework, modeling, building