Data analytics, modelling, and optimization on HVAC systems

• Main Author: Ong, Zi Heng
• Other Authors: Soh Yeng Chai
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2023
• Subjects: Engineering::Electrical and electronic engineering
• Online Access: https://hdl.handle.net/10356/167343
• Institution: Nanyang Technological University

A Heating, Ventilation, and Air Conditioning (HVAC) system provides heating and cooling services to buildings and other enclosed spaces. They are designed to control temperature, air quality, and humidity to create a comfortable indoor environment based on set points. In this industry-supported project, the HVAC system of the building, Keppel Bay Tower (KBT), which has 18 floors, is to be studied for optimization. The project first aims to use data gathered from the building’s Internet of Things sensors to perform data analytics to garner insights on data that may prove useful in identifying the optimality conditions of the system. After data pre processing, the Air Handling Unit (AHU), Chiller (CH), Chiller Water Pump (CHWP), Condenser Water Pump (CWP), and Cooling Tower Fans (CTF) were identified to be the relevant components with useful data. The presence of missing, incomplete, and spurious data was also identified. Relevant measure points were then identified, and through correlation and data analysis, it was found that in general, the CH and CWP component types contributed most efficiently to cooling. Both general and specific trends for each AHU were also uncovered for monitoring, with emphasis placed on components that play a more significant role in cooling. Data-driven models were then developed along with a feature selection study for energy consumption prediction and used to optimize the operations of the HVAC system. The models developed include a Random Forest for Model 0 and Neural Network models for Models 1 and 2. The models were run with a 20% reduction in total power consumption simulated. Poor model evaluation results were produced by Model 0 along with inaccurate predictions of the output temperature, as this model only took in the input temperature and the highest correlated feature of total power. Feature selection was then carried out, and Models 1 and 2 were set to simulate the Chilled Water (CHW) cycle and AHU activities respectfully, with many more inputs. Hence, Model 1 was able to produce highly accurate predictions of the temperature of the CHW supply, and Model 2 was also able to produce accurate predictions of output temperature given Model 1’s input. This process identified that a 20% reduction in power consumption is a viable option to optimize the HVAC system while ensuring comfortable temperatures are kept on weekdays and possibly weekends.