Load profile segmentation for electricity market settlement

Abstract

An unsupervised learning method is used to create clusters for electricity load profiles within a group of real customers. A time-series analysis method (hierarchical clustering) is adopted. A case study is conducted with real consumption data from residential, commercial, and industrial consumers to show the effectiveness of the proposed clustering method for load profiling. After the data cleansing, filtering, and normalization processes, the input dataset is divided into several clusters based on their profile differences. Later, various results are obtained to reflect different consumption patterns within a profile group by the selected distance measurement methods such as Euclidean and Dynamic Time Warping. The results obtained in the case study show that the proposed mathematical algorithm can be used to create realistic and scalable profiling subgroups (with percentages of similar consumptions in each cluster) instead of the traditional methods which cluster all profiles in a single big cluster. The proposed algorithm is used for a case study of Turkey; however, this study is adaptable to other European markets.