| Title |
Machine Learning Models for Prediction and Control of an Ice Thermal Storage System in an Existing Building |
| Authors |
Shin, Han-Sol ; Suh, Won-Jun ; Chu, Han-Gyeong ; Ra, Seon-Jung ; Park, Cheol-Soo |
| DOI |
https://doi.org/10.5659/JAIK_SC.2017.33.11.63 |
| Keywords |
ice thermal storage system ; machine learning ; model predictive control ; office building ; simulation model |
| Abstract |
In South Korea, an ice thermal storage system is popular because night-time electricity rate is cheaper than daytime rate. A spherical ice ball system is one of the most popular ice thermal storage systems used in Korea. However, it is difficult to estimate the degree of freezing and defrosting of the spherical ice ball system and thus, excessive icing commonly occurs in order to prevent any shortage of stored ice. If this rule-of thumb control can be replaced by a simulation model-based control, there would be significant potential for energy savings. In this study, the authors developed 25 machine learning simulation models for the spherical ice thermal storage system installed in a 30-story office building (gross floor area: 32,600m2) located in Seoul, Korea. Five different machine learning algorithms (Artificial Neural Network, Support Vector Machine, Gaussian Process, Random Forest, and Genetic Programming) were used for five different input scenarios, respectively. The 25 machine learning models are accurate enough to predict the amount of icing required for the following daytime. In addition, with the use of Model Predictive Control (MPC), 16.8% of excessive icing during overnight can be reduced and 15% of cooling energy (chiller, cooling tower, Brine pump, etc.) can be saved. |