| Title |
Five Machine Learning Models for HVAC Systems in an Existing Office Building |
| Authors |
Ra, Seon-Jung ; Shin, Han-Sol ; Suh, Won-Jun ; Chu, Han-Gyeong ; Park, Cheol-Soo |
| DOI |
https://doi.org/10.5659/JAIK_SC.2017.33.10.69 |
| Keywords |
Existing building ; machine learning ; data-driven model ; HVAC system ; Simulation model |
| Abstract |
The first principles-based simulation model, e.g. dynamic simulation, is influenced by model uncertainty, simplification of the reality, lack of information, a modeler’s subjective assumptions, etc. Recently, a data-driven machine learning model has received a growing attention for simulation of existing buildings. The data-driven model is advantageous that it is simpler and requires less inputs than the first principles based model. In this study, the authors applied five different machine learning techniques (Artificial Neural Network, Support Vector Machine, Gaussian Process, Random Forest, and Genetic Programming) to HVAC systems (chiller, cooling tower, pump, ice thermal storage system and air handling unit) installed in an existing office building. It was found that the five machine learning models are good enough to predict the dynamic behavior of the HVAC systems. The machine learning model made by Genetic Programming is most accurate among the five machine learning models. The models made by Support Vector Machine and Gaussian Process Model require significant computation time and thus are limited in terms of the number of inputs. The accuracy of the model made by Random Forest is dependent on the set of inputs. |