PCR: um modelo híbrido para previsão de grandezas elétricas aplicado em estudo de caso de um regulador de tensão em operação

Abstract

The growing demand for electricity and requirements imposed by regulatory agencies have led the conventional electricity distribution system to evolve into the context of smart grids. Data acquisition and analysis are central issues for the evolution of the distribution system. Machine learning technologies are gaining ground in applied studies, making this an emerging topic. The related studies do not address the preparation and analysis of data before application in artificial intelligence models. In this context, this work presents a model capable of performing the pre-processing, classification, and prediction of electrical quantities. Compared with related works, there is an indication that this is the first study that addresses the pre-processing and grouping of data by similarity to increase the effectiveness of artificial intelligence models. The Fuzzy C-Means method for data classification allows outliers to be found more assertively. The Grubbs method identifies critical operating points of the system. The regression stage presents predictions made with LSTM neural networks up to fourtime steps ahead with a percentage absolute average error of 0.16% using a real database of an electric power distribution utility.