Optimization of indoor quality and thermal comfort for university classrooms using data-based machine learning

Abstract

Improving indoor environment quality on university classrooms is a hot research topic. The on-site experiment was conducted on indoor environmental quality regarding temperature, humidity, air pollutants, light and acoustics during different seasonal conditions. The result shows that nearly 25% of indoor particulate matters exceeded the GB18883 standard when the outdoor environmental pollution was severe under natural ventilation conditions in autumn. More than 20% of students experienced symptoms of drowsiness, dizziness, chest tightness, poor breathing, as well as depression and irritability. From the analysis of occupant demand, indoor air pollution and thermal comfort are the most anticipated areas for students to improve their learning environment. This paper proposes an optimal IEQ prediction model integrated with students’ satisfaction and indoor environmental features using machine-learning classification algorithms. The back-propagation neural network shows the high prediction accuracy among different algorithms. The traditional PMV-PPD model shows an accuracy rate of only 28% for thermal sensation prediction, while the highest prediction accuracy obtained through machine learning algorithms is about 75%. Moreover, the influence of individual’s thermal adaptation ability, including gender, long-term thermal experience, and psychological factors, and environmental factors was analyzed in this study.