Enhancing indoor learning environments through intelligent air conditioning control

The paper presents an approach utilizing machine learning (ML) for sound classification to optimize room ventilation in educational settings. The first part introduces a convolutional neural network (CNN) model adept at classifying spectrogram images derived from audio recordings into distinct sound classes, achieving consistent accuracies of 0.85 or higher. In the second part, the study delves into configuring a controller employing insights from the ML model to regulate air volume flow and minimize disruptive noise levels from air conditioning units (ACUs). Through a comprehensive parameter study using a transient classroom simulation, optimal control strategies are identified. Findings suggest that a control strategy averaging sound classes over a period coupled with specific intervals for air volume flow adjustments, particularly at a duration of 30 seconds, achieves favourable outcomes in balancing air quality and noise levels. On an evaluation scale between 0 (best) and 2 (worst) consolidating air quality and noise level quality ratings, values of about 0.2 can be reached.