Finding a Comfortable Temperature through Machine Learning
In the pursuit of creating comfortable workspaces and optimizing energy efficiency, large buildings often struggle to regulate temperature adequately. Heating, ventilation, and air conditioning (HVAC) systems are tasked with finding the right balance, but they sometimes fall short. However, researchers from Carnegie Mellon University are exploring how machine learning (ML) models can play a significant role in predicting people’s thermal perceptions and improving the efficiency of HVAC systems.
By leveraging multidimensional association rule mining (M-ARM), the team of researchers aims to address the biases and uncertainties associated with human-perception data and inaccurate temperature predictions. In a recent study featured in Building and Environment, they employed M-ARM to analyze and correct biases present in human responses to temperature across seven ML models.
The study focused on identifying the comfort zone for most people in a building by considering conflicting information provided by occupants when answering multiple questions about their thermal comfort. The goal was to eliminate miscalibration issues and uncover potential subjective data biases. The researchers found that their method significantly improved the accuracy of predicting how individuals feel about the temperature.
Pingbo Tang, the associate professor of civil and environmental engineering who led the study, emphasized the potential of this work in saving energy and enhancing occupant comfort in large buildings. Tang explained that current defective datasets may contribute to excessive energy consumption, and the perception of comfort encompasses multiple factors beyond just temperature, such as humidity and clothing choices.
The research considered various impact factors, including the dataset size, classifier types, and calibration methods. By doing so, the team successfully improved prediction reliability and reduced errors in the existing models used.
These findings have important implications for the advancement of ML-based strategies, with the ultimate goal of achieving more accurate predictions of thermal perception. By gaining better insights into how occupants truly perceive temperature, building managers and HVAC systems can take more effective control measures to enhance comfort and reduce energy consumption.
In summary, the combination of ML models and M-ARM offers a promising approach to finding a comfortable temperature in large buildings. By addressing biases in human responses to temperature and improving prediction accuracy, this research holds the potential to revolutionize energy efficiency while ensuring occupant comfort. As Tang aptly stated, This work is about using the question-answering behavior of the person when they are facing a few related thermal comfort questions to adjust self-conflicts and estimate reality.
Overall, these advancements in ML-based strategies bring us closer to a future where buildings can efficiently and precisely create a comfortable environment for everyone while minimizing energy waste.
