A HYBRID APPROACH BASED ON BUILDING PHYSICS AND MACHINE LEARNING FOR THERMAL COMFORT PREDICTION IN SMART BUILDINGS

Abstract

One of the most important challenges facing the world is the application of modern technology in order to create smart buildings that achieve sustainable development goals (SDGs). Thermal comfort and reduction of energy consumption in buildings are considered important factors which, in turn, are reflected in creating a healthy environment and improving human productivity. Internet of Things (IoT) provides an ideal solution for collecting real-time data on the factors affecting indoor thermal comfort and energy consumption. However, comfort level is subjective and depends on many factors, which may not be learned by conventional models, an integrated model depending on thermal comfort factors is needed. In this work, a hybrid physics-based model incorporated with machine learning techniques is used for the prediction of thermal comfort inside buildings. XGBoost (eXtreme Gradient Boost) algorithm method was used due to its abilities to handle complex problems. A calculated dataset was extracted from the physics-based model gathered with the environmental variables data such as humidity, moisture, temperature, and air velocity collected from IoT devices. The results show an improvement in the prediction of the thermal comfort approach as compared with the conventional models. The XGBoost algorithm can exhibit an effective solution for eliminating deficiencies of traditional models and can be used when designing smart buildings, simulating, and evaluating the designed buildings, controlling energy consumption, and achieving thermal comfort.