New Perspectives on Clustering for Demand Response

Demand response (DR) programs have received significant attention with proliferation of smart meters and increasing need for demand-side flexibility to complement the growing share of renewable generation. A critical element in DR program is the consumer selection; adhoc selection of consumers may not yield any tangible results in actual deployment. Clustering on features derived from smart meter data has shown potential for facilitating the consumer selection for DR. This paper furthers the understanding of this approach by looking at issues associated with the clustering process. Specifically, the paper identifies the problem of defining characteristic profiles for consumers exhibiting multiple consumption patterns. The characteristic profile is a key element for clustering as well as for evaluating behavior consistency. A new method for extracting characteristic profile is presented and metrics for consistency in consumption patterns are redefined. We also propose several useful attributes to quantify the peak load contributions associated with a consumer cluster. We apply the proposed techniques to Dataport smart meter data to bring fresh insights on clustering techniques that segregate consumers based on their consumption and behavioral patterns. We demonstrate how clusters formed using our proposed definition of characteristic profile show bettering clustering consistency. Our results also show how the proposed consistency metrics and peak attributes are useful for capturing the consumer predictability and peak contribution for a more meaningful DR program design.