Data-driven Building HVAC System for Personalized Occupant Satisfaction and Efficient Energy Use

Abstract

The predicted mean vote (PMV) model is widely used to measure thermal comfort for humans, which uses heating, ventilation, and air conditioning (HVAC) systems. However, the PMV model has limitations in satisfying individual person’s thermal comfort. As a result, a recent survey of occupants in buildings showed that the percentage of thermal discomfort is significantly high, despite the active use of the HVAC system. To address this issue, we propose a personalized thermal comfort prediction model based on machine learning that utilizes data from thermal sensation votes, indoor temperature, and humidity. We did an experiment for the data acquisition system, and four students participated. With these data, we develop a personalized thermal comfort prediction model. Among the five machine learning models, i.e., artificial neural network (ANN), linear regression (LR), support vector machine (SVM), ANN is selected showing best performamce. We formulate an optimization problem for the proposed personalized HVAC system, and its solution is derived using a genetic algorithm. The results of the thermal comfort of the personalized model are compared to the PMV model. It shows significant differences between the thermal comfort of the personalized model and the PMV model. Also, the thermal comfort performance and cost are evaluated through a building simulation.