Comparison of Thermal Comfort Models Based on Principal Component Analysis and Ant Colony Optimization

As indoor thermal environment control technology advances, the demand for thermal comfort in buildings has progressively heightened. This study focused on the population in office buildings within Nanjing, measuring a range of parameters that potentially influenced human thermal comfort, including indoor and outdoor temperatures, relative humidity, light intensity, carbon dioxide levels, and air conditioning set temperature. Utilizing machine learning in the realm of thermal comfort, and with the comfort levels derived from survey questionnaires as the evaluation benchmark, a predictive model was developed based on Principal Component Analysis (PCA) and Ant Colony Optimization (ACO). This model was employed to forecast the overall comfort levels in office buildings, exploring the selection of various model parameters and assessing their performance. The findings indicated that the K-Nearest Neighbors (KNN) model with a 9-PCA dataset demonstrated the most robust overall performance, with average metrics of 84.0% Accuracy, 0.396 R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup>, and 0.17 MSE. Among individual models, the Random Forest Regression (RFR) model with a 9-PCA dataset stood out with the best performance, achieving average accuracy, R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup>, and MSE of 87.3%, 0.627, and 0.108, respectively. The performance and stability of ACO-optimized combinatorial model is better than that of RFR, exhibiting a 12.0% reduction in MSE compared to the RFR model.