Conventionally, operations of chillers seriously rely on engineers’ practical experiences. However, various uncertainties, including changeable weather and complicated chiller combinations, lead to inconsistent decisions of switching chiller machines. This technique employs AI and big data analytics to provide decision supports of chiller configuration optimization for industries and reduce the variability of decisions as well as energy waste. Different from most algorithm solutions which provide one-way prediction or only suggestions but no self-learning. Our technique uses real data to feedback itself and applies model health examination, trying to find out root cause and self-repairing. Keeping the model healthy can make sure the suggestion of the decision is in good quality. In the meantime, it takes expert rules and criteria, which might changes from time to time, into account, making the system more intelligent and the suggested decision more expertized.

1.Domain integrated model :
Well embedded domain knowledge and expert experiences in the model, making the suggestions keep up with the times and adapted to local conditions.
2.AI self-calibration:
The system gets feedback from the real data and makes self-improvement with error detection and model evolution. Keeping the model healthy can make sure the suggestion of the decision is in good quality.
3.Precise modeling:
Exclusive research for best chiller efficient performance interval and indoor-outdoor temperature converting formula making the model more precise and accurate.

In high energy-consuming industries such as semiconductor and TFT-LCD manufacturing high-tech industries, the chiller water system is indispensable to factories but requires huge energy consumption. This AI decision support technique can conduct pre-assessment without additional facilities investments. Including wafer fab, backend fab, and panel fab can apply this technique for energy saving.