Machine learning-based predictive model for temperature and comfort parameters in indoor enviroment using experimantal data

Abstract

The research introduces an artificial neural network model that predicts temperature and assesses thermal comfort metrics for a cooling room, demonstrating how machine learning advancements can enhance thermal efficiency and cost-effectiveness in building design. The study utilized the Levenberg-Marquardt (LM) artificial neural network (ANN) approach to derive the average temperature and thermal comfort metrics collected under actual operating settings. The Predicted Mean Vote (PMV) and Predicted Percentage Dissatisfied (PPD)values were measured at three distinct sites and then compared to the trial findings. The model uses a dataset of 205 observations, with 143 cases used for training and 31 examples for testing and validation. The ANN model demonstrated effective training, with negligible errors in estimated error values. The mean squared error values for average temperature and thermal comfort parameters were 0.0342, 0.0376, 0.0571, 0.0029, and 0.2296. The R values for temperature measurements are 0.9947 and 0.9923, 0.9847 and 0.9437, and 0.9737, demonstrating a highly effective engineering method. The ANN model provided precise predictions for temperature and thermal comfort metrics, such as PMV and PPD in a cooling chamber, with a tolerance of ± 15 %. The LM approach, a machine learning methodology, produced excellent outcomes, particularly at lower temperatures, with 15 % of the data exceeding this range.