Exploring the advantages of artificial neural networks in predicting children's thermal perception and their potential application

Abstract

While numerous thermal comfort models have been developed to predict human thermal comfort levels in outdoor areas under varying weather conditions, these indexes are generally designed for adults. To assess the suitability of thermal comfort models, the Universal Thermal Climate Index and a multiple linear regression (MLR) model based on Predicted Mean Vote factors, to predict children's outdoor thermal sensation votes (TSV), field investigations were conducted in a Harbin park across multiple seasons. In addition, two new artificial neural network (ANN) models, with single and double hidden layers, were developed and validated to address a wider range of input parameters than the traditional models, clothing levels and metabolic rates, as well as accounting for a wider range of ages, body weights and heights. The results demonstrated that: 1) the ANN models outperformed the traditional models; 2) The two-hidden-layer ANN model slightly outperformed the one-hidden-layer model; 3) sensitivity analysis identified the top four parameters influencing the prediction of children's TSV in Harbin as mean radiant temperature (0.259), air temperature (0.200), globe temperature (0.161), and children's metabolic rate (0.110). These findings will offer valuable insights for optimizing thermal environments in urban parks, reducing children's thermal stress, and advancing intelligent park services.