Measurement and verification of energy performance for chiller system retrofit with k nearest neighbour regression

Abstract

There is no generic performance indicator for chiller systems running in different cooling load profiles. This study applies the k nearest neighbour (kNN) regression to analyse the pre- and post-energy performances of a chiller system retrofit. The system consists of five sets of chillers, pumps and cooling towers with two different capacities. The retrofit involves recommissioning the control system and replacing faulty variable speed drives. Data were logged at 15-min intervals for the evaluation of system coefficient of performance (SCOP)—the total cooling capacity divided by the total electric use of system components. A total of 15506 sets of operating conditions were gathered year-round before and after retrofit. For each post-operating condition, Euclidean distance was examined to search for five neighbours based on the system capacity, the dry bulb temperature and relative humidity of outdoor air in pre-operating conditions. The SCOP improves by 0.01–88.30% in 79.63% of the post-operating conditions when comparing with neighbour pre-operating conditions. The chiller part load ratio is ranked the first in the improvement, followed by the number of operating chillers. LOESS (local regression) curves facilitate tracking changes and boundaries of significant variables for the improvement. The carbon emissions reduce by 224386 kg CO2e after retrofit. The novelty of this study rests on aligning cooling load profiles with the kNN regression to compare directly SCOPs before and after retrofit.