Similarity-Based Clustering for Identification and Segmentation of Responsive Electricity Customers

The identification and segmentation of responsive electricity customers have been formulated here as a binary time series clustering (TSC) problem. The assumption of a stationary environment in kernel methods can complicate the mapping of non-stationary time series data to a high-dimensional feature space, leading to a degradation in the performance of kernel K-means clustering. Hence, a similarity-based non-linear TSC is proposed to capture consumers’ reactions to demand response (DR) signals. K-means with Dynamic Time Warping (DTW) is employed as a nonlinear TSC approach, and to extend it beyond standard sample-to-centroid comparisons, a similarity matrix is suggested in place of raw time series, enabling sample-to-sample comparisons. It is based on distance and correlation matrices ensembling one-to-many and two-to-two comparisons to map original data to the similarity space. By analyzing consumption data from the Low Carbon London project, we demonstrate the effectiveness of our approach in identifying responsive consumers with different responsive levels.