Chiller system optimization using k nearest neighbour regression

Research output: Contribution to journalArticlepeer-review

9 Citations (Scopus)

Abstract

This study applies the k nearest neighbour (kNN) regression to ascertain optimal operating strategies for a chiller system, hence lowering its carbon emissions. First, 19 operating variables were identified for the regression of the coefficient of performance (COP)—the total cooling capacity divided by the total power of system components. Cross validation involved splitting ten partitions of test samples from a set of operating data with 14537 time points. The optimal k = 3 is evaluated with the lowest mean square error of 0.00148. Using the random kNN, the most significant variables are found to be the temperature and flow rate of return chilled water, the load of individual chillers is the next, followed by the condenser water leaving temperature in system optimization. Based on the existing building cooling load profile with ambient conditions, the optimal variables with the maximum COP are identified from neighbour conditions at the 99th percentile of the distance distribution. The optimal schedule of switching chillers and temperature settings improve the seasonal COP from 3.27 to 3.90, reducing the carbon emissions by 178160 kg CO2e. The novelty of this study is the application of the kNN regression with enhanced distance selection criterion to identify specific optimization strategies which fulfil the existing design and operating constraints of engineering systems.

Original languageEnglish
Article number127050
JournalJournal of Cleaner Production
Volume303
DOIs
Publication statusPublished - 20 Jun 2021

Keywords

  • Coefficient of performance
  • k nearest neighbour
  • Supervised machine learning
  • Water-cooled chiller

Fingerprint

Dive into the research topics of 'Chiller system optimization using k nearest neighbour regression'. Together they form a unique fingerprint.

Cite this