Industry’s First AI Powered Chilled Water Energy Saving System Saves 30GWh of Electricity per Year
TSMC fulfills its commitment on green manufacturing and environmental protection through innovation and development. In 2017, TSMC developed its industry’s first “optimal energy-saving control program” for its chilled water system. Furthermore, the Company pioneered introduction of artificial intelligence (AI) in its energy saving system, using neural network algorithm of machine learning (ML) to build a low energy consumption model in 2018. Finally, TSMC successfully created its industry’s first AI powered chilled water control system that further increased energy efficiency by 2%, estimating 30GWh of electricity to be saved each year.
Introduce AI to Solve the Two Major Efficiency Problems of Chilled Water System
Problem occurs in the chilled water system after a long time of use, such as aging compressors and increased friction loss in pipelines. The optimal energy-saving control program, which was developed with traditional prediction model in 2017, has greatly improved energy efficiency. However, due to the limitation of the prediction model that the system was originally designed with, the system won’t be able to precisely analyze or solve the problems caused by long time of use, which are the most important efficiency problems of chilled water system.
In addition, the equipment required for a chilled water system includes chiller, chilled water pumps, cooling water pumps, cooling tower, heat recovery, and other ancillary components. The energy consumption between each equipment is correlated due to the setting temperature changes, which is also affected by external environment condition, so the relevant parameters of energy consumption are nearly up to 10K. Therefore, it is difficult to accurately predict a set of optimal temperature for energy consuming efficiency by engineers’ experience or using traditional engineering formulas in such high dimensional and complex nonlinear chilled water system. This is another major efficiency problem of chilled water system.
Instantly and Accurately Determine the Optimal Energy Efficiency Parameters through Machine Learning
In order to determine the optimal energy efficiency of the chilled water system, TSMC’s facility team used the massive amount of operational data, which was generated by an “optimal energy-saving control program,” to create a new model with neural network algorithm of machine learning. The new model has successfully determined 90 key parameters related to energy efficiency and electric consumption from over 1,000 related parameters of chilled water system.
To make good use of these 90 key parameters, facility team repeatedly explored the correlations between those 90 parameters, and evaluated their importance through 4.15 million data modeling trainings in 15 months. Therefore, the new model has successfully taken care of both efficiency problems of aging machine and nonlinear system complexity, which the old system wasn’t able to resolve, and determined the best temperature for optimal energy efficiency with the new energy consumption prediction model in the new chilled water system. Compared with the previous “optimal energy-saving control program” developed in 2017, the energy efficiency has been increased 2% by the new system.