A new clustering method with an ensemble of weighted distance metrics to discover daily patterns of indoor air quality

Abstract

This paper presents a novel clustering method with an ensemble distance metric to identify meaningful daily patterns of building indoor air quality (IAQ). The novelty of this work is to construct an ensemble of weighted distance metrics with weights being adaptive to the characteristics of IAQ data. The ensemble distance was constructed based on Euclidean distance, Pearson correlation coefficient and cross-correlation coefficient to capture absolute levels, overall trends, and volatilities of IAQ data. The weights for individual distance metrics are automatically adapted to the characteristics of IAQ data based on the Z-scores of the distances among all pair-wise samples in the dataset. In addition, IAQ daily time series are segmented into short sequences to reduce the computational cost. Two datasets containing one-week IAQ observations were used to verify the proposed method. The results showed that the proposed method is able to capture insightful information on the distribution and frequency of the peaks, volatilities, and absolute levels of various IAQ parameters by identifying the patterns with unique characteristics from the IAQ daily sequences. The use of the ensemble distance metric offered better performance with an improved pattern consistency rate of 11–22% when compared with using a single distance metric for IAQ data. The discovered patterns can provide an increased understanding of the pollutant levels, the IAQ changing trends, and the distributions in the peaks and valleys of IAQ pollutants and, hence, can support the optimal operation of building HVAC systems.