Clustering Load Profiles for Demand Response Applications

TLDR

Smart grid technologies enable residential and commercial loads to participate in demand response, making data dimension reduction and classification essential for DR success. The study proposes a load‑profile clustering method that uses information entropy, piecewise aggregate approximation, and spectral clustering to classify load data. The method models typical daily loads with a variable temporal resolution technique and applies an improved spectral clustering based on multi‑scale distance and shape similarities, demonstrated on a case study of 100 commercial HVAC datasets. The approach achieves feasible data dimension reduction, reasonable profile selection and classification, and stable operation performance.

Abstract

With the development of smart grid technologies, residential and commercial loads have large potentialities to participate in demand response (DR) programs. This makes the data dimension reduction techniques and classification processing critical for the success of DR development. A novel load profile clustering method is proposed for load data classification which is based on the information entropy, piecewise aggregate approximation, and spectral clustering (SC). The variable temporal resolution technique is presented to model typical daily load datasets, and then an improved SC based on multi-scale similarities of distance and shape characteristics is proposed for clustering to obtain reasonable load classification. A case study with one hundred of commercial heating, ventilation, and air conditioning data analysis illustrates the approach. The results prove that the proposed method is feasible in terms of data dimension reduction, reasonable profile selection and classification, and the operation stability.