An Approach for Assessing Occupancy and Indoor CO₂ Concentrations Based on Alternative Environmental Metrics Through Machine Learning

There has been an evolving discussion regarding indoor air quality (IAQ) challenges in buildings through the outbreak of the most recent pandemic and other respiratory illnesses. Proper management of indoor environmental factors is crucial for maintaining a healthy indoor environment while also optimizing energy usage. One of the key factors affecting indoor air quality is CO2 concentration, where controlling its levels within acceptable ranges over time is essential to maintain occupants’ cognitive performance. The use of multiple CO2 sensors for more precise IAQ monitoring or installation of variable speed air systems that adjust the ventilation needs at any given time has limitations due to their relatively high cost and are missing in most existing buildings. Therefore, there is indeed a need for a simpler, yet sufficiently accurate and affordable approach to assess actual CO2 levels to keep them below the acceptable CO2 thresholds. In this study, we propose a machine learning method as an alternative approach to provide better control over CO2 concentration based on assessing the importance of other indoor environmental factors in real-time predictions. The obtained results reveal a strong correlation between CO2 concentrations and indoor temperature variations as a possible proxy, combined with fan coil unit (FCU) speeds and absolute humidity (AH) in a case study with a DOAS system.