A comparative study of clustering algorithms for intermittent heating demand considering time series

Occupants' behavior has a significant impact on building energy consumption. Various clustering algorithms, represented by the K-means algorithm based on the “Euclidean” distance, are widely used to extract occupants' behavior patterns. However, these commonly used algorithms cannot effectively capture the temporal characteristics of occupants' behavior. Although some studies have started to adopt clustering algorithms considering time series for research, the differences in clustering effects of different algorithms on occupants' behavior are still unclear, especially in the clustering application study of behavior with intermittent characteristics, this intermittent behavior is a behavior that does not always exist, and the timing of its occurrence changes. Therefore, this study takes the heating demand behavior of occupants with obvious intermittency as a case and compares the clustering effect of the common clustering algorithm (K-means) and three clustering algorithms considering time series (K-shape, Dynamic Time Warping (DTW), and Derivative Dynamic Time Warping (DDTW)). The study also explores possible reasons leading to differences and introduces three evaluation methods for clustering results of intermittent heating demand. This study provides a reference for selecting appropriate clustering algorithms to study occupants' intermittent behavior and has practical application value in improving the extraction and prediction of behavior patterns of occupants with intermittent characteristics.