Sustainable Development Goals
Abstract/Objectives
This technology integrates AI, big data, and machine learning to predict the cooling demand of factories by considering external climatic conditions. It analyzes data to estimate the energy efficiency of each chiller and develops optimal settings and scheduling combinations for chillers, water pumps, and cooling towers within the chilled water system. Practical needs such as time-of-use electricity pricing, optimal chiller load ranges, and maintenance are also considered, as shown in Figure 1. The relationship models between dew point temperature, cooling demand, parameter setting combinations, and power consumption are established and continuously enhanced, as illustrated in Figure 2. This ensures chillers maintain high efficiency, reduce energy consumption and electricity costs, improve energy efficiency, and achieve the goals of energy saving, carbon reduction, and net-zero sustainability.
Results/Contributions

This technology is applicable to areas requiring temperature control and various types of chillers, without the need for additional investment in software or hardware. By leveraging AI, big data, and machine learning to capture information and optimizes the quality of complex decision-making combinations, thereby creating additional benefits in energy savings, carbon reduction, and electricity cost reduction.

Cooling demand is influenced by external temperature, factory size, and utilization conditions. The chiller system, which uses water as a medium, mainly consists of chillers, water pumps, and cooling towers connected by pipelines. The chillers provide cooling, the cooling towers dissipate heat, and the water pumps provide the kinetic energy for circulating water between the chillers, cooling towers, and usage points. The interaction effects of these factors are complex and vary for each plant. After years of research, this technology has been continuously refined using various machine learning methods, including SARIMAX (Seasonal Autoregressive Integrated Moving Average with Exogenous Regressors), MARS (Multivariate Adaptive Regression Splines), random forests, extreme gradient boosting, and ensemble learning. These methods are employed to establish a digital twin model that predicts cooling ton demand and suggests parameter settings for the chilled water system. This helps reduce power consumption, lower electricity costs, and decrease carbon emissions. It also controls process equipment and site temperature to ensure product quality.

This technology holds significant value for high-energy-consuming high-tech industries and has been proven effective in several panel, semiconductor, and PCB factories. It has won the 2020 17th National Innovation Award and the 2021 Ministry of Science and Technology Future Tech Award, as shown in Figure 3. In 2023, Business Weekly reported extensively on Macronix's collaboration with National Tsing Hua University on this technology, which has been validated over many years, positioning the company as a leader in AI-driven carbon reduction, as shown in Figure 4. The PCB giant Zhen Ding Technology Group has also established the "ZD-NTHU Joint Research Center" and is closely collaborating on this technology. Its effectiveness has been validated and will be expanded to more than a dozen plants within the group.

Keywords
1.Chiller System Energy Optimization 2.Energy Consumption Optimization 3.Net-Zero Emissions 4.Sustainable Development 5.Artificial Intelligence 6.Machine Learning 7.Digital Twins 8.Smart Manufacturing
Contact Information
盧映宇
dalab@ie.nthu.edu.tw