An Approach to Cluster Electrical Load Profiles Based on Piecewise Symbolic Aggregation

Abstract

With the extensive establishment of advanced metering infrastructure (AMI) and the development of big data technology, large volume of electricity load profile data collected from smart meter reveal information about customer's electricity consumption behavior. The precise knowledge of customer's load profiles clustering technology is helpful for power company to develop differentiated user services and reasonable power dispatching. In general, customers's load profiles exist several typical load profiles(TLPs) and a few abnormal or irregular load profiles, but the conventional clustering method can not distinguish them accurately. Therefore, this paper proposes an approach based on piecewise symbolic aggregation. Firstly, this method reduces the dimension of load profiles by a time series segmentation method based on Pearson correlation coefficient (PCC), and all load profiles are divided into several subsequences. Then, each sub-sequences is replaced by a character according to the size of each sub-sequence's mean value. After that, each load profile will be replaced by a symbolic string, and the load profiles with same symbolic string will be cluster into a group. Finally, this paper performs a case analysis with a Irish dataset, and the results show that the proposed approach can improve the clustering quality of electrical load profiles.