Development of indoor/outdoor environment and dynamic clothing insulation-based thermal comfort prediction model using artificial neural network

Abstract

This study presents the development of a predicted mean vote (PMV) prediction model based on an artificial neural network (ANN), utilizing dynamic clothing insulation calculated via a linear regression model and easily measurable indoor and outdoor environmental factors. Additionally, the cooling and heating performance of an air-cooled variable refrigerant flow (VRF) heat pump system was modeled under various load conditions. The validity of the building model for PMV prediction was established by comparing simulation outcomes with actual building power consumption. Four scenarios were designed by varying the combinations of input variables required for PMV prediction, and each scenario's performance was evaluated. Among these, Scenario 3, which only considered dynamic clothing volume alongside simple indoor and outdoor variables, demonstrated improved predictive accuracy compared to the more comprehensive Scenario 1. Furthermore, Scenario 4, which included CO₂ concentration as an additional variable, exhibited the best prediction performance. The model effectively achieved high thermal comfort prediction across different American Society of Heating, Refrigerating and Air-Conditioning Engineers (ASHRAE) climate zones, showing an average coefficient of variation of the root mean square error (CVRMSE) of 7.83 %, 5.16 %, and 6.78 %, and a standard deviation percentage error of 4.34 %, 4.94 %, and 3.16 %, respectively. The results indicate that the developed model not only aligns well with the predicted PMV distribution and mean values but also captures the variability observed in real-world measurements. This demonstrates the model’s capability to accurately forecast PMV using readily measurable environmental factors through ANN.