5.14 Patterns Recognition in Energy Management

Abstract

This chapter aims to discuss issues related to the unsupervised recognition of patterns in the general scope and in particular the problems belonging to the electric sector. In this aspect, an initial discussion was made on the main concepts and procedures of the clustering analysis. For time factor on group formation, some clustering algorithms are required. In engineering, pattern recognition in time series has been used for the detection and diagnosis of faults, optimization of trajectories, and characterization of consumption profiles, mainly electric power. It is worth mentioning that univariate time series (STU) clustering algorithms use standard approaches using point-prototype grouping models. However, pattern recognition in multivariate time series (STM) represents a more complex problem (nonpoint-prototype problem) with intrinsic characteristics. This chapter presents a method for the selection, typification, and clustering of STU capable of recognizing consumption patterns in the electricity sector. For the STM, a special clustering method was proposed based on multivariate statistics. Both the method was successfully implemented and tested in the context of an energy efficiency program carried out by the Energy Company of Maranhao and Alagoas (Brazil), respectively. The results reveal the viability of the method in recognizing patterns consistent with the reality of the electricity sector. The proposed method is also useful to support decision-making at management level.