Abstract/Summary

Although local air movement acts as a critical factor to enhance human thermal comfort and energy efficiency, the various factors influencing such movement have led to inconsistent publications on how to evaluate and design localised airflow systems in practice. This study aims to identify the main impacting factors for a localised airflow system and predict a cooling performance based on machine learning algorithms. Three typical localised airflow forms, i.e. an isothermal air supply (IASN), non-isothermal air supply (NIASN), and floor fan (FF), were deployed. The experiments were conducted under a variety of temperature/humidity/air velocity conditions in a well-controlled climate chamber, and a database including 1305 original samples was built. The primary results indicated that a classification tree C5.0 model showed a better prediction performance (83.99%) for a localised airflow system, with 17 input parameters in the model. Through a sensitivity analysis, 8 feature variables were quantified as having significant main effect responses on subjects’ thermal sensation votes (TSV), and three environmental factors (temperature, air velocity, and relative humidity) were identified as having the most significant effects. Using the 8 sensitive factors, the C5.0 model was modified with 82.30% accuracy for subject TSV prediction. A tree model demonstrating the decision rules in the C5.0 model was obtained, with air velocity (=0 m/s,＞0 m/s) as the first feature variable, and root node and temperature (≤28 °C,＞28 °C) as the second feature variable and leaf node, respectively. The outcomes that provide the most influential variables and a machine learning model are beneficial for evaluating personal thermal comfort at individual levels and for guiding the application of a localised airflow system in buildings.