Predicting chiller system performance using ARIMA-regression models

A proper selection of predictor variables would enhance the exploratory analysis of time series models while prompting practical strategies to optimize chiller system performance. This study explores essential operating variables to predict the time series of the coefficient of performance (COP) of a chiller system expressed as the cooling capacity output divided by the total electric power input of all components. Based on a huge set of historical operating data, hybrid ARIMA-regression models were developed by fitting 14 predictor variables other than the past COP-related terms. The most significant variable influencing predictability involves the part load ratio (PLR) and its order-3 lag terms lasting for 45 min. The system COP fluctuation is mainly governed by the PLR variation due to operating unnecessary chillers and non-pair up operation of system components. When chiller sequencing is properly implemented, the PLRs shift up with tempered variation. The paired component combinations lower the system electric power to maximize the system COP. The annual average system COP increases to 3.5538 from 3.3212 with a predicted electricity saving of 8.2955%. The novelty of this study is to highlight which variables and component operating statuses help improve the predictability of time series models while prioritizing practical strategies for performance improvement.

Automated summary

This study explores the predictor variables for time series models of chiller systems and finds that the part load ratio is a key factor influencing system performance and predictability. Proper chiller sequencing can increase system efficiency.