An efficient thermal comfort prediction method for indoor airflow environment using a CFD-based deep learning model

Thermal comfort in indoor environments significantly affects human health and productivity, while there remains room for improvement in enhancing thermal comfort around individuals. This study proposed an efficient thermal comfort prediction method based on the Convolutional Neural Networks-Long Short-Term Memory (CNN-LSTM) model to rapidly and accurately assess indoor thermal comfort. As demonstrated with a high-speed train, the computational fluid dynamics (CFD) technology is combined to establish the dataset. Five design parameters (the ratio and angle of the upper inlets, supply air temperature and humidity, and external temperature) and four evaluation indices (air velocity, air temperature, Predicted Mean Vote, and Draft Rate) are considered in assessing the accuracy of the method on the test dataset. The results indicate that CNN-LSTM achieves consistent and accurate predictive performance, with average mean absolute error (MAE) close to 0.01 m/s, 0.2 °C, 0.1, and 1.0. On the generalization test set, the predictive performance of CNN-LSTM decreases slightly, but the average of the determination coefficients ($R^2$) still approaches 0.89. The thermal comfort prediction method developed in this study demonstrates significant advantages in predictive performance, showing great potential for application in the construction of healthy and comfortable indoor environments in buildings, aircraft, subways, etc.