Microbiome-Aware HVAC Control in Green Buildings Using Reinforcement Learning

Most of us spend the vast majority of our lives inside buildings. We sleep, work, and gather in spaces where the air is managed by mechanical systems designed to keep us comfortable and safe. For decades, the standard for judging the quality of this indoor air has focused on chemical pollutants and energy use. Certifications for green buildings, for instance, measure how well a structure filters out carbon dioxide, dust, and volatile chemicals, while also tracking how much electricity it consumes to stay cool or warm. These measures are vital for safety, but they tell only part of the story. They overlook the invisible, living world that thrives within our walls: the microbiome. This is the complex community of bacteria, fungi, and other microscopic organisms that circulate in the air we breathe. Just as the soil in a forest has a specific balance of life that keeps the ecosystem healthy, the air inside a building hosts its own microbial community, which can influence human health, from our immune systems to our respiratory well-being.

Until now, the technology that controls our indoor air has largely ignored this biological layer. Heating and cooling systems are tuned to manage temperature and humidity based on chemical sensors, not on the health of the microbial ecosystem. A new study by Pradip Thorat introduces a different way of thinking, proposing that we can and should manage the air in our buildings to support a stable, healthy community of microbes. The research does not rely on physical experiments in a real building but instead uses a sophisticated computer simulation to test a new kind of artificial intelligence. This AI acts as a building manager, constantly adjusting the ventilation and airflow to see if it can create a better environment for these microscopic life forms while still saving energy. The goal is to move beyond simply keeping the air clean of chemicals to actively nurturing a balanced biological environment.

The researchers built a digital model of a three-zone office building, complete with an office, a conference room, and a lobby. They programmed this virtual building to mimic how air moves, how humidity changes, and how microbial populations grow and shift in response to those conditions. To measure the success of their system, they created a new metric called the Microbiome Stability Index. Think of this index as a score that reflects how balanced and diverse the microbial community is, while penalizing situations where harmful pathogens take over or when the population fluctuates wildly. The higher the score, the more stable and healthy the air is considered to be. The team then trained a reinforcement learning agent, a type of artificial intelligence that learns by trial and error, to control the building's ventilation. The AI's job was to adjust how much fresh air was brought in, balancing the desire for a high stability score against the cost of the energy required to move that air.

The results of these simulations revealed a surprising truth about how we manage our indoor air. The AI-controlled system managed to achieve a higher stability score than the traditional method, which relies on a fixed amount of ventilation running constantly. While the standard approach kept the ventilation at a steady rate of two air changes per hour, the AI learned to be much more dynamic. It reduced the average ventilation to just 1.30 air changes per hour, yet it still improved the stability of the microbial community by 1.2 percent. This finding challenges the common assumption that more ventilation is always better for air quality. The AI learned that by carefully timing when to bring in fresh air and when to conserve energy, it could maintain a healthier biological environment with less effort. The system was particularly effective at keeping the air stable during most of the simulation, only ramping up ventilation aggressively when the stability score began to drop.

One of the most striking discoveries in the study concerned humidity. Current building standards generally recommend keeping relative humidity between 40 and 60 percent to prevent mold growth and dust mites. However, the simulation showed that the stability of the microbial community actually increased as humidity rose, reaching its peak at 80 percent. This suggests that the current guidelines, which prioritize suppressing mold, might be inadvertently limiting the diversity and stability of the broader microbial ecosystem. The data indicated a strong, positive relationship between higher humidity and a more stable microbiome, a finding that contradicts the traditional focus on keeping air dry. The researchers noted that while 80 percent humidity might seem high for preventing mold, it appears to be the sweet spot for maintaining a balanced microbial community in this specific model.

The study also highlighted that not all buildings or climates are created equal when it comes to managing this biological balance. The simulations showed that buildings in coastal and humid-tropical climates were naturally better at sustaining a stable microbiome, requiring less energy to maintain high scores. In contrast, arid desert and cold-dry climates imposed a steep penalty, making it much harder and more energy-intensive to achieve the same level of stability. This suggests that a single, universal rule for managing indoor air quality may not work everywhere. The AI also demonstrated that different rooms within the same building require different amounts of energy to reach the same level of stability. For example, the office space in the simulation needed twice as much energy as the smaller conference room to achieve the same stability score, simply because of differences in volume and airflow dynamics. This points to a future where building controls are highly customized, adjusting to the specific needs of each room and the local climate.

While the results are promising, the researchers are careful to note that these findings come from a computer simulation, not a physical building. The model made several simplifications, such as assuming a constant temperature and treating the air in each room as a perfectly mixed substance, which does not account for the complex micro-environments around desks or windows. The formula used to calculate the stability score also relied on fixed weights that might need to be adjusted for different types of buildings or seasons. Despite these limitations, the study provides a crucial proof of concept. It demonstrates that it is possible to use artificial intelligence to manage the biological quality of indoor air, balancing health and energy efficiency in a way that traditional systems cannot. The work suggests that the next generation of green buildings might not just be about saving energy or filtering chemicals, but about actively cultivating a healthy, living atmosphere for the people inside.