Development and Verification of a Comfort Index Model Based on Deep Learning
Abstract：To provide more scientific and reasonable comfort service products, taking the comfort station in the ecological base of Haizhu District as an example, a comfort index model was constructed using artificial intelligence algorithms such as deep neural networks and extreme gradient boosting trees. The test results showed that: (1) The general human comfort index has significant deviations compared to the biological comfort instrument index due to the consideration of solar radiation characteristics, and this deviation shows a consistent tendency to be lower, especially obvious during the daytime. (2) The comfort models based on wind, temperature, and humidity constructed by deep neural networks and extreme gradient boosting trees have some improvement compared to the human comfort index model, but the deviation is still large during the daytime. (3) By introducing radar reflectivity, precipitation, and time factors to represent solar radiation characteristics into the deep neural network comfort model, this improvement significantly enhanced the model's prediction accuracy, with the mean absolute error reduced to 2.7, a 50% reduction compared to the human comfort index model, and the average error distribution approaching a quasi-normal distribution centered around zero.
Keywords：Applied MeteorologyDeep Neural NetworkExtreme Gradient Boosting TreeHuman Comfort IndexHaizhu District, Guangzhou
Publication Date：2025-08-20
Online Publishing Date：2025-09-15(First online date of this platform, not the publication date of the document)
Pages：4( 67-70 )