scope:

ABSTRACT

This paper presents a new approach for the development and use of benchmarking models in the context of ongoing commissioning. Different techniques are explored to establish the benchmarking models: (1) a static approach, which is based on predefined training set size and established different models for weekdays and weekend and holidays, or (2) window techniques, which are either augmented or sliding. The performance of each approach is evaluated for two chillers installed in an existing central cooling and heating plant. Both chillers have identical capacity and performance characteristics; however, they have quite different operating hours. For chiller CH-1, the models established with a 21-day data set for the augmented window technique and the 14-day data set for the sliding window technique give the best testing results. For chiller CH-2, the static model established with one week of data and the augmented window model with seven days of data provides the most accurate testing results.