ECO-FOCUS: Integrating computer vision and machine learning for personalized comfort prediction and adaptive HVAC control in office buildings

The building sector accounts for a considerable portion of global energy consumption, especially in heating, ventilation, and air conditioning (HVAC) systems. Traditional, static HVAC operations often fail to adapt to evolving indoor conditions or individual comfort preferences, resulting in energy waste and occupant discomfort in office spaces. This study proposes ECO-FOCUS (energy-conscious operation framework for occupant-centric control using sensors), a formal framework that integrates real-time sensing, thermal performance modeling, and decision-making systems. The prototype of ECO-FOCUS consists of several key components. A cost-effective sensor network collects real-time indoor data, combining wireless cameras and environmental sensors. These data feed into a computer vision model that accurately detects occupants and estimates clothing insulation, both of which are critical factors in assessing thermal comfort. A hybrid prediction model that integrates the traditional predicted mean vote model with machine learning is employed to continuously predict personalized comfort setpoints based on long-term occupant feedback. These setpoints are dynamically applied via a zonal HVAC control strategy, supported by an EnergyPlus co-simulation platform for real-time performance analysis. A case study conducted in a university office demonstrates a 29.3 % reduction in room-level cooling energy usage, with zone-level savings ranging from 1.5 % to 97 %, depending on occupancy patterns. These energy reductions were achieved along with improved thermal comfort, especially in zones with occupants who share similar comfort preferences. The findings highlight ECO-FOCUS as an adaptive and flexible solution for the sustainable HVAC operations in office buildings.

© 2025 Elsevier Ltd. All rights are reserved, including those for text and data mining, AI training, and similar technologies.