Traditional ventilation strategies often fail to maintain both optimal CO2 levels and indoor thermal comfort. CO2 levels exceeding 1200 ppm can impair cognitive function, reduce productivity, and negatively impact mental health, posing pressing public health concerns. However, many remain unaware of these risks and often overlook the hazards of indoor CO2 toxicity. In the U.S., natural ventilation is inadequate in 88% of low-income housing. Existing automated window systems provide a solution, but they are highly expensive, require retrofitting, and need more maintenance than traditional windows.

The real-data collection conducted in a dorm room revealed concerning indoor environmental conditions, particularly in terms of CO2 concentration. Throughout the observation period, the CO2 levels consistently exceeded recommended thresholds, indicating poor air quality and potential health risks for the occupants. The data highlights the urgent need for effective ventilation solutions to improve indoor air quality and ensure a healthier environment for residents.

Physics-based air volume exchange and energy consumption approach models the relationship between air flow dynamics and required heating energy within indoor environments. By integrating this model with thermal energy equations, the energy required to maintain or restore indoor comfort conditions during air exchanges is quantified. This framework not only provides insights into the machine learning baseline  but also enables comparisons with subsequent computational simulations and real-world experimental data.

The machine learning model leverages occupant behavior predictions to optimize window operation based on user activities and needs. By analyzing both indoor and outdoor environmental conditions, the system anticipates human activity and occupancy patterns to determine the ideal window opening angles and durations for enhanced indoor comfort. Unlike fully automated window systems, this solution focuses on providing actionable insights to occupants, empowering them to manage their indoor environment effectively. This approach minimizes costs by eliminating the need for expensive hardware or retrofitting while maintaining high flexibility and adaptability for various building types.

The machine learning control system demonstrated notable improvements in both indoor environmental quality and energy efficiency. The results revealed a significant reduction in CO2 levels, reflecting the system's capability to enhance air quality through optimized ventilation strategies. While the system achieved slightly higher indoor temperatures, this difference was minimal due to the testing occurring during the shoulder season, when outdoor temperatures were moderate. Despite this, the system still exhibited reduced energy consumption compared to traditional fixed temperature set-point controls, showcasing its ability to efficiently balance comfort and resource conservation.