Pattern recognition as a tool to support decision making in the management of the electric sector. Part II: A new method based on clustering of multivariate time series

Abstract

This work presents a new method for the clustering and pattern recognition of multivariate time series (CPT-M) based on multivariate statistics. The algorithm comprises four steps that extract essential features of multivariate time series of residential users with emphasis on seasonal and temporal profile, among others. The method was successfully implemented and tested in the context of an energy efficiency program carried out by the Electric Company of Alagoas (Brazil) that considers, among others, the analysis of the impact of replacing refrigerators in low-income consumers’ homes in several towns located within the state of Alagoas (Brazil). The results were compared with a well-known method of time series clustering already established in the literature, the Fuzzy C-Means (FCM). Unlike C-means models of clustering, the CPT-M method is also capable to obtain directly the number of clusters. The analysis confirmed that the CPT-M method was capable to identify a greater diversity of patterns, showing the potential of this method in better recognition of consumption patterns considering simultaneously the effect of other variables in additional to load curves. This represents an important aspect to the process of decision making in the energy distribution sector.