Application of data fusion and FDD for improving the performance of chiller sequencing control

Cooling load based chiller sequencing control is in principle the best among those existent methods. However, site experiences show that the cooling load direct measurement, i.e., cooling load calculated based on the supply chilled water and the return chilled water temperatures and the chilled water flow rate is easy to be corrupted by measurement noise, outliers, systematic errors and various soft or hard faults. Inaccurate cooling load measurement will lead to poor performance of chiller sequencing control. Therefore, this paper presents a control strategy with enhanced robustness and energy efficiency. Data fusion technique is introduced to mitigate/eliminate the negative effects of measurement noises, outliers and systematic errors, which aims at acquiring a more accurate and reliable fused cooling load. The fused measurement of building cooling load is obtained by combining the complementary advantages of direct cooling load measurement with those of indirect cooling load measurement calculated from an inverse model. Meanwhile, a parameter named confidence degree Yf which is used to indicate the quality of the fused building cooling load measurement is generated. For improving the performance of the chiller sequencing control, two schemes based on data fusion are developed. The first one is to use the fused cooling load measurement instead of the direct measurement in chiller sequencing control. The chiller sequencing control performance will be enhanced due to the improved accuracy/reliability of the fused building cooling load measurement. The second one is to integrate sensor fault detection and diagnosis (FDD) technique into chiller sequencing control for guaranteeing the essential sensors operating healthily. Since the confidence degree has a low value when systematic errors or measurement faults occur in the sensor measurement for cooling load direct measurement, the sensor fault diagnosis algorithm will be triggered by the low confidence degree. After the faults are detected, the fault diagnosis algorithm is implemented to isolate the particular misbehaving sensors. The developed schemes are tested using a simulated central chiller plant with six identical chillers. The first case study shows that the performance of chiller sequencing control will be strengthened by using the fused building cooling load measurement. The second case study shows the effectiveness of the sensor diagnosis strategy by isolating all the systematic errors and faults. Full scale site application is being carried out in a super high rise building in Kowloon.